To see the other types of publications on this topic, follow the link: Explainable artifical intelligence.
Dissertations / Theses on the topic 'Explainable artifical intelligence'
Create a spot-on reference in APA, MLA, Chicago, Harvard, and other styles
Consult the top 39 dissertations / theses for your research on the topic 'Explainable artifical intelligence.'
Next to every source in the list of references, there is an 'Add to bibliography' button. Press on it, and we will generate automatically the bibliographic reference to the chosen work in the citation style you need: APA, MLA, Harvard, Chicago, Vancouver, etc.
You can also download the full text of the academic publication as pdf and read online its abstract whenever available in the metadata.
Browse dissertations / theses on a wide variety of disciplines and organise your bibliography correctly.
1
Nilsson, Linus. "Explainable Artificial Intelligence for Reinforcement Learning Agents." Thesis, KTH, Skolan för elektroteknik och datavetenskap (EECS), 2021. http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-294162.
Full textAbstract:
Following the success that machine learning has enjoyed over the last decade, reinforcement learning has become a prime research area for automation and solving complex tasks. Ranging from playing video games at a professional level to robots collaborating in picking goods in warehouses, the applications of reinforcement learning are numerous. The systems are however, very complex and the understanding of why the reinforcement learning agents solve the tasks given to them in certain ways are still largely unknown to the human observer. This makes the actual use of the agents limited to non-critical tasks and the information that could be learnt by them hidden. To this end, explainable artificial intelligence (XAI) has been a topic that has received more attention in the last couple of years, in an attempt to be able to explain the machine learning systems to the human operators. In this thesis we propose to use model-agnostic XAI techniques combined with clustering techniques on simple Atari games, as well as proposing an automated evaluation for how well the explanations explain the behavior of the agents. This in an effort to uncover to what extent model-agnostic XAI can be used to gain insight into the behavior of reinforcement learning agents. The tested methods were RISE, t-SNE and Deletion. The methods were evaluated on several different agents trained on playing the Atari-breakout game and the results show that they can be used to explain the behavior of the agents on a local level (one individual frame of a game sequence), global (behavior over the entire game sequence) as well as uncovering different strategies used by the agents and as training time differs between agents.Efter framgångarna inom maskininlärning de senaste årtiondet har förstärkningsinlärning blivit ett primärt forskningsämne för att lösa komplexa uppgifter och inom automation. Tillämpningarna är många, allt från att spela datorspel på en professionell nivå till robotar som samarbetar för att plocka varor i ett lager. Dock så är systemen väldigt komplexa och förståelsen kring varför en agent väljer att lösa en uppgift på ett specifikt sätt är okända för en mänsklig observatör. Detta gör att de praktiska tillämpningarna av dessa agenter är begränsade till icke-kritiska system och den information som kan användas för att lära ut nya sätt att lösa olika uppgifter är dolda. Utifrån detta så har förklarbar artificiell intelligens (XAI) blivit ett område inom forskning som fått allt mer uppmärksamhet de senaste åren. Detta för att kunna förklara maskininlärningssystem för den mänskliga användaren. I denna examensrapport föreslår vi att använda modelloberoende XAI tekniker kombinerat klustringstekniker på enkla Atarispel, vi föreslår även ett sätt att automatisera hur man kan utvärdera hur väl en förklaring förklarar beteendet hos agenterna. Detta i ett försök att upptäcka till vilken grad modelloberoende XAI tekniker kan användas för att förklara beteenden hos förstärkningsinlärningsagenter. De testade metoderna var RISE, t-SNE och Deletion. Metoderna utvärderades på flera olika agenter, tränade att spelaAtari-breakout. Resultatet visar att de kan användas för att förklara beteendet hos agenterna på en lokal nivå (en individuell bild ur ett spel), globalt beteende (över den totala spelsekvensen) samt även att metoderna kan hitta olika strategier användna av de olika agenterna där mängden träning de fått skiljer sig.
2
Vincenzi, Leonardo. "eXplainable Artificial Intelligence User Experience: contesto e stato dell’arte." Bachelor's thesis, Alma Mater Studiorum - Università di Bologna, 2021. http://amslaurea.unibo.it/23338/.
Full textAbstract:
Il grande sviluppo del mondo dell’Intelligenza Artificiale unito alla sua vastissima applicazione in molteplici ambiti degli ultimi anni, ha portato a una sempre maggior richiesta di spiegabilità dei sistemi di Machine Learning. A seguito di questa necessità il campo dell’eXplainable Artificial Intelligence ha compiuto passi importanti verso la creazione di sistemi e metodi per rendere i sistemi intelligenti sempre più trasparenti e in un futuro prossimo, per garantire sempre più equità e sicurezza nelle decisioni prese dall’AI, si prevede una sempre più rigida regolamentazione verso la sua spiegabilità.
Per compiere un ulteriore salto di qualità, il recente campo di studio XAI UX si pone come obiettivo principale l’inserimento degli utenti al centro dei processi di progettazione di sistemi di AI, attraverso la combinazione di tecniche di spiegabilità offerte dall'eXplainable AI insieme allo studio di soluzioni UX.
Il nuovo focus sull’utente e la necessità di creare team multidisciplinari, e quindi con maggiori barriere comunicative tra le persone, impongono ancora un largo studio sia da parte degli esperti XAI sia da parte della comunità HCI, e sono attualmente le principali difficoltà da risolvere.
All’interno dell’elaborato si fornisce una visione attuale sullo stato dell'arte della XAI UX introducendo le motivazioni, il contesto e i vari ambiti di ricerca che comprende.
3
PICCHIOTTI, NICOLA. "Explainable Artificial Intelligence: an application to complex genetic diseases." Doctoral thesis, Università degli studi di Pavia, 2021. http://hdl.handle.net/11571/1447637.
Full text
4
ABUKMEIL, MOHANAD. "UNSUPERVISED GENERATIVE MODELS FOR DATA ANALYSIS AND EXPLAINABLE ARTIFICIAL INTELLIGENCE." Doctoral thesis, Università degli Studi di Milano, 2022. http://hdl.handle.net/2434/889159.
Full textAbstract:
For more than a century, the methods of learning representation and the exploration of the intrinsic structures of data have developed remarkably and currently include supervised, semi-supervised, and unsupervised methods. However, recent years have witnessed the flourishing of big data, where typical dataset dimensions are high, and the data can come in messy, missing, incomplete, unlabeled, or corrupted forms. Consequently, discovering and learning the hidden structure buried inside such data becomes highly challenging.
From this perspective, latent data analysis and dimensionality reduction play a substantial role in decomposing the exploratory factors and learning the hidden structures of data, which encompasses the significant features that characterize the categories and trends among data samples in an ordered manner. That is by extracting patterns, differentiating trends, and testing hypotheses to identify anomalies, learning compact knowledge, and performing many different machine learning (ML) tasks such as classification, detection, and prediction.
Unsupervised generative learning (UGL) methods are a class of ML characterized by their possibility of analyzing and decomposing latent data, reducing dimensionality, visualizing the manifold of data, and learning representations with limited levels of predefined labels and prior assumptions. Furthermore, explainable artificial intelligence (XAI) is an emerging field of ML that deals with explaining the decisions and behaviors of learned models. XAI is also associated with UGL models to explain the hidden structure of data, and to explain the learned representations of ML models. However, the current UGL models lack large-scale generalizability and explainability in the testing stage, which leads to restricting their potential in ML and XAI applications.
To overcome the aforementioned limitations, this thesis proposes innovative methods that integrate UGL and XAI to enable data factorization and dimensionality reduction to improve the generalizability of the learned ML models. Moreover, the proposed methods enable visual explainability in modern applications as anomaly detection and autonomous driving systems. The main research contributions are listed as follows:
• A novel overview of UGL models including blind source separation (BSS), manifold learning (MfL), and neural networks (NNs). Also, the overview considers open issues and challenges among each UGL method.
• An innovative method to identify the dimensions of the compact feature space via a generalized rank in the application of image dimensionality reduction.
• An innovative method to hierarchically reduce and visualize the manifold of data to improve the generalizability in limited data learning scenarios, and computational complexity reduction applications.
• An original method to visually explain autoencoders by reconstructing an attention map in the application of anomaly detection and explainable autonomous driving systems.
The novel methods introduced in this thesis are benchmarked on publicly available datasets, and they outperformed the state-of-the-art methods considering different evaluation metrics. Furthermore, superior results were obtained with respect to the state-of-the-art to confirm the feasibility of the proposed methodologies concerning the computational complexity, availability of learning data, model explainability, and high data reconstruction accuracy.
5
Rouget, Thierry. "Learning explainable concepts in the presence of a qualitative model." Thesis, University of Ottawa (Canada), 1995. http://hdl.handle.net/10393/9762.
Full textAbstract:
This thesis addresses the problem of learning concept descriptions that are interpretable, or explainable. Explainability is understood as the ability to justify the learned concept in terms of the existing background knowledge. The starting point for the work was an existing system that would induce only fully explainable rules. The system performed well when the model used during induction was complete and correct. In practice, however, models are likely to be imperfect, i.e. incomplete and incorrect. We report here a new approach that achieves explainability with imperfect models. The basis of the system is the standard inductive search driven by an accuracy-oriented heuristic, biased towards rule explainability. The bias is abandoned when there is heuristic evidence that a significant loss of accuracy results from constraining the search to explainable rules only. The users can express their relative preference for accuracy vs. explainability. Experiments with the system indicate that, even with a partially incomplete and/or incorrect model, insisting on explainability results in only a small loss of accuracy. We also show how the new approach described can repair a faulty model using evidence derived from data during induction.
6
Gjeka, Mario. "Uno strumento per le spiegazioni di sistemi di Explainable Artificial Intelligence." Bachelor's thesis, Alma Mater Studiorum - Università di Bologna, 2020.
Find full textAbstract:
L'obiettivo di questa tesi è quello di mostrare l’importanza delle spiegazioni in un sistema intelligente. Il bisogno di avere un'intelligenza artificiale spiegabile e trasparente sta crescendo notevolmente, esigenza evidenziata dalla ricerca delle aziende di sviluppare sistemi informatici intelligenti trasparenti e spiegabili.
7
Amarasinghe, Kasun. "Explainable Neural Networks based Anomaly Detection for Cyber-Physical Systems." VCU Scholars Compass, 2019. https://scholarscompass.vcu.edu/etd/6091.
Full textAbstract:
Cyber-Physical Systems (CPSs) are the core of modern critical infrastructure (e.g. power-grids) and securing them is of paramount importance. Anomaly detection in data is crucial for CPS security. While Artificial Neural Networks (ANNs) are strong candidates for the task, they are seldom deployed in safety-critical domains due to the perception that ANNs are black-boxes. Therefore, to leverage ANNs in CPSs, cracking open the black box through explanation is essential.
The main objective of this dissertation is developing explainable ANN-based Anomaly Detection Systems for Cyber-Physical Systems (CP-ADS). The main objective was broken down into three sub-objectives: 1) Identifying key-requirements that an explainable CP-ADS should satisfy, 2) Developing supervised ANN-based explainable CP-ADSs, 3) Developing unsupervised ANN-based explainable CP-ADSs.
In achieving those objectives, this dissertation provides the following contributions: 1) a set of key-requirements that an explainable CP-ADS should satisfy, 2) a methodology for deriving summaries of the knowledge of a trained supervised CP-ADS, 3) a methodology for validating derived summaries, 4) an unsupervised neural network methodology for learning cyber-physical (CP) behavior, 5) a methodology for visually and linguistically explaining the learned CP behavior.
All the methods were implemented on real-world and benchmark datasets. The set of key-requirements presented in the first contribution was used to evaluate the performance of the presented methods. The successes and limitations of the presented methods were identified. Furthermore, steps that can be taken to overcome the limitations were proposed. Therefore, this dissertation takes several necessary steps toward developing explainable ANN-based CP-ADS and serves as a framework that can be expanded to develop trustworthy ANN-based CP-ADSs.
8
PANIGUTTI, Cecilia. "eXplainable AI for trustworthy healthcare applications." Doctoral thesis, Scuola Normale Superiore, 2022. https://hdl.handle.net/11384/125202.
Full textAbstract:
Acknowledging that AI will inevitably become a central element of clinical practice,
this thesis investigates the role of eXplainable AI (XAI) techniques in developing
trustworthy AI applications in healthcare. The first part of this thesis focuses on the
societal, ethical, and legal aspects of the use of AI in healthcare. It first compares
the different approaches to AI ethics worldwide and then focuses on the practical
implications of the European ethical and legal guidelines for AI applications in
healthcare. The second part of the thesis explores how XAI techniques can help meet
three key requirements identified in the initial analysis: transparency, auditability,
and human oversight. The technical transparency requirement is tackled by enabling
explanatory techniques to deal with common healthcare data characteristics
and tailor them to the medical field. In this regard, this thesis presents two novel
XAI techniques that incrementally reach this goal by first focusing on multi-label
predictive algorithms and then tackling sequential data and incorporating domainspecific
knowledge in the explanation process. This thesis then analyzes the ability
to leverage the developed XAI technique to audit a fictional commercial black-box
clinical decision support system (DSS). Finally, the thesis studies AI explanation’s
ability to effectively enable human oversight by studying the impact of explanations
on the decision-making process of healthcare professionals.
9
Keneni, Blen M. Keneni. "Evolving Rule Based Explainable Artificial Intelligence for Decision Support System of Unmanned Aerial Vehicles." University of Toledo / OhioLINK, 2018. http://rave.ohiolink.edu/etdc/view?acc_num=toledo1525094091882295.
Full text
10
Hammarström, Tobias. "Towards Explainable Decision-making Strategies of Deep Convolutional Neural Networks : An exploration into explainable AI and potential applications within cancer detection." Thesis, Uppsala universitet, Avdelningen för visuell information och interaktion, 2020. http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-424779.
Full textAbstract:
The influence of Artificial Intelligence (AI) on society is increasing, with applications in highly sensitive and complicated areas. Examples include using Deep Convolutional Neural Networks within healthcare for diagnosing cancer. However, the inner workings of such models are often unknown, limiting the much-needed trust in the models. To combat this, Explainable AI (XAI) methods aim to provide explanations of the models' decision-making. Two such methods, Spectral Relevance Analysis (SpRAy) and Testing with Concept Activation Methods (TCAV), were evaluated on a deep learning model classifying cat and dog images that contained introduced artificial noise. The task was to assess the methods' capabilities to explain the importance of the introduced noise for the learnt model. The task was constructed as an exploratory step, with the future aim of using the methods on models diagnosing oral cancer. In addition to using the TCAV method as introduced by its authors, this study also utilizes the CAV-sensitivity to introduce and perform a sensitivity magnitude analysis. Both methods proved useful in discerning between the model’s two decision-making strategies based on either the animal or the noise. However, greater insight into the intricacies of said strategies is desired. Additionally, the methods provided a deeper understanding of the model’s learning, as the model did not seem to properly distinguish between the noise and the animal conceptually. The methods thus accentuated the limitations of the model, thereby increasing our trust in its abilities. In conclusion, the methods show promise regarding the task of detecting visually distinctive noise in images, which could extend to other distinctive features present in more complex problems. Consequently, more research should be conducted on applying these methods on more complex areas with specialized models and tasks, e.g. oral cancer.
11
Houzé, Étienne. "Explainable Artificial Intelligence for the Smart Home : Enabling Relevant Dialogue between Users and Autonomous Systems." Thesis, KTH, Skolan för elektroteknik och datavetenskap (EECS), 2019. http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-252705.
Full textAbstract:
Smart home technologies aim at providing users control over their house, including but not limited to temperature, light, security, energy consumption. Despite offering the possibility of lowering power consumption (and thus the energy bill), smart home systems still struggle to convince a broad public, often being regarded as intrusive or not trustworthy. Gaining sympathy from users would require a solution to provide relevant explanations without relying on distant computing (which would imply sending private data online). We therefore propose an architecture where the autonomic controller system is extended with an intelligent layer aiming at translating autonomic values into concepts and finding causality relations in order to explain the logic behind the decisions and the state of the system to the user.Smart-home-teknik syftar till att ge användarna bättre kontroll över sina hem till exempel med avseende på temperatur, ljus, säkerhet eller förbrukning av energi. Denna teknik har svårt att övertyga många kunder trots löften om energibesparing och komfort. Tekniken ses som påträngande. En lösning som ökar användarnas välvilja genom att ge relevanta förklaringar men utan att skicka privata data online är önskvärd. I detta examensarbete föreslår vi en arkitektur där ett intelligent lager läggs ovanpå det autonoma systemet. Det intelligenta lagret försöker dels översätta autonoma termer till begrepp som användaren förstår och dels att hitta kausala relationer som förklarar systemets tillstånd och beslut som fattas av systemet.
12
Pierrard, Régis. "Explainable Classification and Annotation through Relation Learning and Reasoning." Electronic Thesis or Diss., université Paris-Saclay, 2020. http://www.theses.fr/2020UPAST008.
Full textAbstract:
Avec les succés récents de l’apprentissage profond et les interactions toujours plus nombreuses entre êtres humains et intelligences artificielles, l’explicabilité est devenue une préoccupation majeure. En effet, il est difficile de comprendre le comportement des réseaux de neurones profonds, ce qui les rend inadaptés à une utilisation dans les systèmes critiques. Dans cette thèse, nous proposons une approche visant à classifier ou annoter des signaux tout en expliquant les résultats obtenus. Elle est basée sur l’utilisation d’un modèle transparent, dont le raisonnement est clair, et de relations floues interprétables qui permettent de représenter l’imprécision du langage naturel.Au lieu d’apprendre sur des exemples sur lesquels les relations ont été annotées, nous proposons de définir un ensemble de relations au préalable. L’évaluation de ces relations sur les exemples de la base d’entrainement est accélérée grâce à deux heuristiques que nous présentons. Ensuite, les relations les plus pertinentes sont extraites en utilisant un nouvel algorithme de frequent itemset mining flou. Ces relations permettent de construire des règles pour la classification ou des contraintes pour l’annotation. Ainsi, une explication en langage naturel peut être générée.Nous présentons des expériences sur des images et des séries temporelles afin de montrer la généricité de notre approche. En particulier, son application à l’annotation d’organe explicable a été bien évaluée par un ensemble de participants qui ont jugé les explications convaincantes et cohérentesWith the recent successes of deep learning and the growing interactions between humans and AIs, explainability issues have risen. Indeed, it is difficult to understand the behaviour of deep neural networks and thus such opaque models are not suited for high-stake applications. In this thesis, we propose an approach for performing classification or annotation and providing explanations. It is based on a transparent model, whose reasoning is clear, and on interpretable fuzzy relations that enable to express the vagueness of natural language.Instead of learning on training instances that are annotated with relations, we propose to rely on a set of relations that was set beforehand. We present two heuristics that make the process of evaluating relations faster. Then, the most relevant relations can be extracted using a new fuzzy frequent itemset mining algorithm. These relations enable to build rules, for classification, and constraints, for annotation. Since the strengths of our approach are the transparency of the model and the interpretability of the relations, an explanation in natural language can be generated.We present experiments on images and time series that show the genericity of the approach. In particular, the application to explainable organ annotation was received positively by a set of participants that judges the explanations consistent and convincing
13
Hedström, Anna. "Explainable Artificial Intelligence : How to Evaluate Explanations of Deep Neural Network Predictions using the Continuity Test." Thesis, KTH, Skolan för elektroteknik och datavetenskap (EECS), 2020. http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-281279.
Full textAbstract:
With a surging appetite to leverage deep learning models as means to enhance decisionmaking, new requirements for interpretability are set. Renewed research interest has thus been found within the machine learning community to develop explainability methods that can estimate the influence of a given input feature to the prediction made by a model. Current explainability methods of deep neural networks have nonetheless shown to be far from fault-proof and the question of how to properly evaluate explanations has largely remained unsolved. In this thesis work, we have taken a deep look into how popular, state-of-the-art explainability methods are evaluated with a specific focus on one evaluation technique that has proved particularly challenging - the continuity test. The test captures the sensitivity of explanations by measuring how an explanation, in relation to a model prediction, changes when its input is perturbed. While it might sound like a reasonable expectation that the methods behave consistently in their input domain, we show in experiments on both toy-and real-world data, on image classification tasks, that there is little to no empirical association between how explanations and networks respond to perturbed inputs. As such, we challenge the proposition that explanations and model outcomes can, and should be, compared. As a second line of work, we also point out how and why it is problematic that commonly applied ad-hoc perturbation techniques tend to produce samples that lie far from the data distribution. In the pursuit for better, more plausible image perturbations for the continuity test, we therefore present an alternative approach that relies on sampling in latent space, as learned by a probabilistic, generative model. To this end, we hope that the work presented in this thesis will not only be helpful in identifying limitations of current evaluation techniques, but that the work also contributes with ideas of how to improve them.Med ett ökat intresse att använda djupa neurala nätverk i olika delar av samhället, ställs nya krav på dess tolkbarhet. Förnyat forskningsintresse har således infunnit sig för att utveckla förklaringsmodeller som kan redogöra varför ett beslut har tagits av en algoritm. Förklaringsmodellerna har dock visat sig vara långt ifrån felfria och frågan om hur man utvärderar dem har i stort sett förblivit obesvarad. I detta arbete har vi undersökt hur populära förklaringsmodeller evalueras med ett specifikt fokus på en evalueringsteknik som har visat sig vara särskilt problematisk - “kontinuitetstestet”. Testet fångar upp förklaringarnas “sensitivitet” genom att mäta hur en förklaring, i förhållande till modellens prediktion, förändras när dess input modifieras. Även om det kan låta som en rimlig förväntning att förklaringsmetoderna skall uppträda konsekvent i sin inputdomän, så visar vi i bildklassificeringsexperiment på både syntetisk och real data, att det empiriska sambandet mellan hur förklaringar och nätverk svarar på modifierad input är obetydlig. Därmed utmanar vi idén om att förklaringar och modellresultat kan och bör jämföras. Som en andra inriktning i detta arbete påpekar vi också hur och varför det är problematiskt att vanligt förekommande bildmanipuleringstekniker tenderar att producera bilder som ligger långt ifrån den ursprungliga datadistributionen. I strävan efter mer trovärdiga och bättre bildmanipuleringstekniker för kontinuitetstestet, presenterar vi därför ett alternativt tillvägagångssätt som förlitar sig på sampling i det latenta rummet, producerat av en probabilistisk generativ modell. För detta ändamål hoppas vi därför att arbetet som presenteras i denna examensarbete inte bara belyser brister i aktuella utvärderingstekniker, utan också bidrar med idéer om hur man kan förbättra dem.
14
BITETTO, ALESSANDRO. "The role of Explainable Artificial Intelligence in risk assessment: a study on the economic and epidemiologic impact." Doctoral thesis, Università degli studi di Pavia, 2022. http://hdl.handle.net/11571/1452624.
Full textAbstract:
The growing application of black-box Artificial Intelligence algorithms in many real-world application is raising the importance of understanding how the models make their decision. The research field that aims to "open" the black-box and to make the predictions more interpretable, is referred as eXplainable Artificial Intelligence (XAI). Another important field of research, strictly related to XAI, is the compression of information, also referred as dimensionality reduction. Having a synthetic set of few variables that captures the behaviour and the relationships of many more variables can be an effective tool for XAI as well. Thus, the contribution of the present thesis is the development of new approaches in the field of explainability, working on the two complementary pillars of dimensionality reduction and variables importance. The convergence of the two pillars copes with the aim of helping decision makers with the interpretation of the results.
This thesis is composed of seven chapters: an introduction and a conclusion plus five self contained sections reporting the corresponding papers. Chapter 1 proposes a PCA-based method to create a synthetic index to measure the condition of a country’s financial system, providing policy makers and financial institutions with a monitoring and policy tool that is easy to implement and update. In chapter 2, a Dynamic Factor Model is used to produce a synthetic index that is able to capture the time evolution of cross-country dependencies of financial variables. The index is proved to increase the accuracy in predicting the ease in accessing to financial funding. In chapter 3, a set of variables covering health, environmental safety infrastructures, demographic, economic and institutional effectiveness is used to test two methodologies to build an Epidemiological Susceptibility Risk index. The predictive power of both indexes is tested on forecasting task involving Macroeconomic variables. In chapter 4, the credit riskiness of Small Medium Enterprises (henceforth SMEs) is assessed by testing and assessing the increase of performance of a machine learning historical random forest model compared to an ordered probit model. The relevance of each variable in predicting SME credit risk is assessed by using Shapley values. In chapter 5, a dataset of Italian unlisted firms provides evidence of the importance of using market information when assessing the credit risk for SMEs. A non-linear dimensionality reduction technique is applied to assign market volatility from listed peers and to evaluate Merton's probability of default (PD). Results show the increase in accuracy of predicting the default of unlisted firms when using the evaluated PD. Moreover, the way PD affects the defaults is explored by assessing its contribution to the predicted outcome by the means of Shapley values.
15
Verenich, Ilya. "Explainable predictive monitoring of temporal measures of business processes." Thesis, Queensland University of Technology, 2018. https://eprints.qut.edu.au/124037/1/Ilya_Verenich_Thesis.pdf.
Full textAbstract:
This thesis explores data-driven, predictive approaches to monitor business process performance. These approaches allow process stakeholders to prevent or mitigate potential performance issues or compliance violations in real time, as early as possible. To help users understand the rationale for the predictions and build trust in them, the thesis proposes two techniques for explainable predictive process monitoring: one based on deep learning, the other driven by process models. This is achieved by decomposing a prediction into its elementary components. The techniques are compared against state-of-the-art baselines and a trade-off between accuracy and explainability of the predictions is evaluated.
16
Bracchi, Luca. "I-eXplainer: applicazione web per spiegazioni interattive." Bachelor's thesis, Alma Mater Studiorum - Università di Bologna, 2020. http://amslaurea.unibo.it/20424/.
Full textAbstract:
Lo scopo di questa tesi è dimostrare che è possibile trasformare una spiegazione statica relativa all'output di un algoritmo spiegabile in una spiegazione interattiva con una maggior grado di efficacia. La spiegazione cercata sarà human readable ed esplorabile.
Per dimostrare la tesi userò la spiegazione di uno degli algoritmi spiegabili del toolkit AIX360 e andrò a generarne una statica che chiamerò spiegazione di base. Questa verrà poi espansa ed arricchita grazie a una struttura dati costruita per contenere informazioni formattate e connesse tra loro. Tali informazioni verranno richieste dall'utente, dandogli la possibilità di costruire una spiegazione interattiva che si sviluppa secondo il suo desiderio.
17
Costa, Bueno Vicente. "Fuzzy Horn clauses in artificial intelligence: a study of free models, and applications in art painting style categorization." Doctoral thesis, Universitat Autònoma de Barcelona, 2021. http://hdl.handle.net/10803/673374.
Full textAbstract:
Aquesta tesi doctoral contribueix a l’estudi de les clàusules de Horn en lògiques difuses, així com al seu ús en representació difusa del coneixement aplicada al disseny d’un algorisme de classificació de pintures segons el seu estil artístic. En la primera part del treball ens centrem en algunes nocions rellevants per a la programació lògica, com ho són per exemple els models lliures i les estructures de Herbrand en lògica matemàtica difusa. Així doncs, provem l’existència de models lliures en classes universals difuses de Horn, i demostrem que tota teoria difusa universal de Horn sense igualtat té un model de Herbrand. A més, introduïm dues nocions de minimalitat per a models lliures, i demostrem que aquestes nocions són equivalents en el cas de les fully named structures. En la segona part de la tesi doctoral, utilitzem les clàusules de Horn combinades amb el modelatge qualitatiu com a marc de representació difusa del coneixement per a la categorització d’estils de pintura artística. Finalment, dissenyem un classificador de pintures basat en clàusules de Horn avaluades, descriptors qualitatius de colors i explicacions. Aquest algorisme, anomenat l-SHE, proporciona raons dels resultats obtinguts i mostra percentatges competitius de precisió a l’experimentació.La presente tesis doctoral contribuye al estudio de las cláusulas de Horn en lógicas difusas, así como a su uso en representación difusa del conocimiento aplicada al diseño de un algoritmo de clasificación de pinturas según su estilo artístico. En la primera parte del trabajo nos centramos en algunas nociones relevantes para la programación lógica, como lo son por ejemplo los modelos libres y las estructuras de Herbrand en lógica matemática difusa. Así pues, probamos la existencia de modelos libres en clases universales difusas de Horn y demostramos que toda teoría difusa universal de Horn sin igualdad tiene un modelo de Herbrand. Asimismo, introducimos dos nociones de minimalidad para modelos libres, y demostramos que estas nociones son equivalentes en el caso de las fully named structures. En la segunda parte de la tesis doctoral, utilizamos cláusulas de Horn combinadas con el modelado cualitativo como marco de representación difusa del conocimiento para la categorización de estilos de pintura artística. Finalmente, diseñamos un clasificador de pinturas basado en cláusulas de Horn evaluadas, descriptores cualitativos de colores y explicaciones. Este algoritmo, que llamamos l-SHE, proporciona razones de los resultados obtenidos y obtiene porcentajes competitivos de precisión en la experimentación.
This PhD thesis contributes to the systematic study of Horn clauses of predicate fuzzy logics and their use in knowledge representation for the design of an art painting style classification algorithm. We first focus the study on relevant notions in logic programming, such as free models and Herbrand structures in mathematical fuzzy logic. We show the existence of free models in fuzzy universal Horn classes, and we prove that every equality-free consistent universal Horn fuzzy theory has a Herbrand model. Two notions of minimality of free models are introduced, and we show that these notions are equivalent in the case of fully named structures. Then, we use Horn clauses combined with qualitative modeling as a fuzzy knowledge representation framework for art painting style categorization. Finally, we design a style painting classifier based on evaluated Horn clauses, qualitative color descriptors, and explanations. This algorithm, called l-SHE, provides reasons for the obtained results and obtains percentages of accuracy in the experimentation that are competitive.
Universitat Autònoma de Barcelona. Programa de Doctorat en Ciència Cognitiva i Llenguatge
18
Varela, Rial Alejandro 1993. "In silico modeling of protein-ligand binding." Doctoral thesis, TDX (Tesis Doctorals en Xarxa), 2022. http://hdl.handle.net/10803/673579.
Full textAbstract:
The affinity of a drug to its target protein is one of the key properties of a drug. Although there are experimental methods to measure the binding affinity, they are expensive and relatively slow. Hence, accurately predicting this property with software tools would be very beneficial to drug discovery. In this thesis, several applications have been developed to model and predict the binding mode of a ligand to a protein, to evaluate the feasibility of that prediction and to perform model interpretability in deep neural networks trained on protein-ligand complexes.La afinidad de un fármaco a su proteína diana es una de las propiedades clave de un fármaco. Actualmente, existen métodos experimentales para medir la afinidad, pero son muy costosos y relativamente lentos. Así, predecir esta propiedad con precisión empleando herramientas de software sería muy beneficioso para el descubrimiento de fármacos. En esta tesis se han desarrollado aplicaciones de software para modelar y predecir el modo de unión de ligando a proteína, para evaluar cómo de factible es tal predicción y para interpretar redes neuronales profundas entrenadas en complejos proteína-ligando.
19
Palmisano, Enzo Pio. "A First Study of Transferable and Explainable Deep Learning Models for HPC Systems." Master's thesis, Alma Mater Studiorum - Università di Bologna, 2020. http://amslaurea.unibo.it/20099/.
Full textAbstract:
Il lavoro descritto in questa tesi è basato sullo studio di modelli di Deep Learning applicati all’anomaly detection per la rilevazioni di stati anomali nei sistemi HPC (High-Performance Computing). In particolare, l’obiettivo è studiare la trasferibilità di un modello e di spiegarne il risultato prodotto attraverso tecniche dell’Explainable Artificial Intelligence. I sistemi HPC sono dotati di numerosi sensori in grado di monitorare il corretto funzionamento in tempo reale. Tuttavia, a causa dell’elevato grado di complessità di questi sistemi, si rende necessario l’uso di tecniche innovative e capaci di prevedere guasti, errori e qualsiasi tipo di anomalia per ridurre i costi di manutenzione e per rendere il servizio sempre disponibile. Negli anni ci sono stati numerosi studi per la realizzazioni di modelli di Deep Learning utili allo scopo, ma ciò che manca a questi modelli è la capacità di generalizzare a condizioni diverse da quelle riscontrate durante la fase di training. Nella prima parte di questo lavoro andremo, quindi, ad analizzare un modello già sviluppato per studiarne la trasferibilità e la sua generalizzazione per un’applicazione più ampia rispetto al dominio su cui è costruito. Un ulteriore problema è dato dalle modalità di risposta di un modello ad un determinato input. Molto spesso risulta essere incomprensibile anche per coloro che hanno costruito il modello, la risposta prodotta. Perciò, attraverso le tecniche dell’Explainable Artificial Intelligence, vengono studiati e analizzati i vari output del modello per comprenderne il funzionamento e motivare i risultati corretti ed errati. Il sistema HPC che andremo ad analizzare è chiamato MARCONI ed è di proprietà del consorzio no-profit Cineca di Bologna. Il Cineca è composto da 70 università italiane, quattro centri di ricerca nazionali e il Ministero di Università e ricerca (MIUR). Il Cineca rappresenta uno dei più potenti centri di supercalcolo per la ricerca scientifica in Italia.
20
Sabbatini, Federico. "Interpretable Prediction of Galactic Cosmic-Ray Short-Term Variations with Artificial Neural Networks." Master's thesis, Alma Mater Studiorum - Università di Bologna, 2020. http://amslaurea.unibo.it/22147/.
Full textAbstract:
Monitoring the galactic cosmic-ray flux variations is a crucial issue for all those space missions for which cosmic rays constitute a limitation to the performance of the on-board instruments. If it is not possible to study the galactic cosmic-ray short-term fluctuations on board, it is necessary to benefit of models that are able to predict these flux modulations.
Artificial neural networks are nowadays the most used tools to solve a wide range of different problems in various disciplines, including medicine, technology, business and many others. All artificial neural networks are black boxes, i.e.\ their internal logic is hidden to the user. Knowledge extraction algorithms are applied to the neural networks in order to obtain explainable models when this lack of explanation constitutes a problem.
This thesis work describes the implementation and optimisation of an explainable model for predicting the galactic cosmic-ray short-term flux variations observed on board the European Space Agency mission LISA Pathfinder. The model is based on an artificial neural network that benefits as input data of solar wind speed and interplanetary magnetic field intensity measurements gathered by the National Aeronautics and Space Administration space missions Wind and ACE orbiting nearby LISA Pathfinder. The knowledge extraction is performed by applying to the underlying neural network both the ITER algorithm and a linear regressor. ITER was selected after a deep investigation of the available literature. The model presented here provides explainable predictions with errors smaller than the LISA Pathfinder statistical uncertainty.
21
Mualla, Yazan. "Explaining the Behavior of Remote Robots to Humans : An Agent-based Approach." Thesis, Bourgogne Franche-Comté, 2020. http://www.theses.fr/2020UBFCA023.
Full textAbstract:
Avec l’émergence et la généralisation des systèmes d'intelligence artificielle, comprendre le comportement des agents artificiels, ou robots intelligents, devient essentiel pour garantir une collaboration fluide entre l'homme et ces agents. En effet, il n'est pas simple pour les humains de comprendre les processus qui ont amenés aux décisions des agents. De récentes études dans le domaine l’intelligence artificielle explicable, particulièrement sur les modèles utilisant des objectifs, ont confirmé qu'expliquer le comportement d’un agent à un humain favorise la compréhensibilité de l'agent par ce dernier et augmente son acceptabilité. Cependant, fournir des informations trop nombreuses ou inutiles peut également semer la confusion chez les utilisateurs humains et provoquer des malentendus. Pour ces raisons, la parcimonie des explications a été présentée comme l'une des principales caractéristiques facilitant une interaction réussie entre l’homme et l’agent. Une explication parcimonieuse est définie comme l'explication la plus simple et décrivant la situation de manière adéquate. Si la parcimonie des explications fait l'objet d'une attention croissante dans la littérature, la plupart des travaux ne sont réalisés que de manière conceptuelle.Dans le cadre d'une méthodologie de recherche rigoureuse, cette thèse propose un mécanisme permettant d’expliquer le comportement d’une intelligence artificielle de manière parcimonieuse afin de trouver un équilibre entre simplicité et adéquation. En particulier, il introduit un processus de formulation des explications, sensible au contexte et adaptatif, et propose une architecture permettant d’expliquer les comportements des agents à des humains (HAExA). Cette architecture permet de rendre ce processus opérationnel pour des robots distants représentés comme des agents utilisant une architecture de type Croyance-Désir-Intention.Pour fournir des explications parcimonieuses, HAExA s'appuie d'abord sur la génération d'explications normales et contrastées, et ensuite sur leur mise à jour et leur filtrage avant de les communiquer à l'humain. Nous validons nos propositions en concevant et menant des études empiriques d'interaction homme-machine utilisant la simulation orientée-agent. Nos études reposent sur des mesures bien établies pour estimer la compréhension et la satisfaction des explications fournies par HAExA. Les résultats sont analysés et validés à l'aide de tests statistiques paramétriques et non paramétriquesWith the widespread use of Artificial Intelligence (AI) systems, understanding the behavior of intelligent agents and robots is crucial to guarantee smooth human-agent collaboration since it is not straightforward for humans to understand the agent’s state of mind. Recent studies in the goal-driven Explainable AI (XAI) domain have confirmed that explaining the agent’s behavior to humans fosters the latter’s understandability of the agent and increases its acceptability. However, providing overwhelming or unnecessary information may also confuse human users and cause misunderstandings. For these reasons, the parsimony of explanations has been outlined as one of the key features facilitating successful human-agent interaction with a parsimonious explanation defined as the simplest explanation that describes the situation adequately. While the parsimony of explanations is receiving growing attention in the literature, most of the works are carried out only conceptually.This thesis proposes, using a rigorous research methodology, a mechanism for parsimonious XAI that strikes a balance between simplicity and adequacy. In particular, it introduces a context-aware and adaptive process of explanation formulation and proposes a Human-Agent Explainability Architecture (HAExA) allowing to make this process operational for remote robots represented as Belief-Desire-Intention agents. To provide parsimonious explanations, HAExA relies first on generating normal and contrastive explanations and second on updating and filtering them before communicating them to the human.To evaluate the proposed architecture, we design and conduct empirical human-computer interaction studies employing agent-based simulation. The studies rely on well-established XAI metrics to estimate how understood and satisfactory the explanations provided by HAExA are. The results are properly analyzed and validated using parametric and non-parametric statistical testing
22
Ankaräng, Marcus, and Jakob Kristiansson. "Comparison of Logistic Regression and an Explained Random Forest in the Domain of Creditworthiness Assessment." Thesis, KTH, Skolan för elektroteknik och datavetenskap (EECS), 2021. http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-301907.
Full textAbstract:
As the use of AI in society is developing, the requirement of explainable algorithms has increased. A challenge with many modern machine learning algorithms is that they, due to their often complex structures, lack the ability to produce human-interpretable explanations. Research within explainable AI has resulted in methods that can be applied on top of non- interpretable models to motivate their decision bases. The aim of this thesis is to compare an unexplained machine learning model used in combination with an explanatory method, and a model that is explainable through its inherent structure. Random forest was the unexplained model in question and the explanatory method was SHAP. The explainable model was logistic regression, which is explanatory through its feature weights. The comparison was conducted within the area of creditworthiness and was based on predictive performance and explainability. Furthermore, the thesis intends to use these models to investigate what characterizes loan applicants who are likely to default. The comparison showed that no model performed significantly better than the other in terms of predictive performance. Characteristics of bad loan applicants differed between the two algorithms. Three important aspects were the applicant’s age, where they lived and whether they had a residential phone. Regarding explainability, several advantages with SHAP were observed. With SHAP, explanations on both a local and a global level can be produced. Also, SHAP offers a way to take advantage of the high performance in many modern machine learning algorithms, and at the same time fulfil today’s increased requirement of transparency.I takt med att AI används allt oftare för att fatta beslut i samhället, har kravet på förklarbarhet ökat. En utmaning med flera moderna maskininlärningsmodeller är att de, på grund av sina komplexa strukturer, sällan ger tillgång till mänskligt förståeliga motiveringar. Forskning inom förklarar AI har lett fram till metoder som kan appliceras ovanpå icke- förklarbara modeller för att tolka deras beslutsgrunder. Det här arbetet syftar till att jämföra en icke- förklarbar maskininlärningsmodell i kombination med en förklaringsmetod, och en modell som är förklarbar genom sin struktur. Den icke- förklarbara modellen var random forest och förklaringsmetoden som användes var SHAP. Den förklarbara modellen var logistisk regression, som är förklarande genom sina vikter. Jämförelsen utfördes inom området kreditvärdighet och grundades i prediktiv prestanda och förklarbarhet. Vidare användes dessa modeller för att undersöka vilka egenskaper som var kännetecknande för låntagare som inte förväntades kunna betala tillbaka sitt lån. Jämförelsen visade att ingen av de båda metoderna presterande signifikant mycket bättre än den andra sett till prediktiv prestanda. Kännetecknande särdrag för dåliga låntagare skiljde sig åt mellan metoderna. Tre viktiga aspekter var låntagarens °ålder, vart denna bodde och huruvida personen ägde en hemtelefon. Gällande förklarbarheten framträdde flera fördelar med SHAP, däribland möjligheten att kunna producera både lokala och globala förklaringar. Vidare konstaterades att SHAP gör det möjligt att dra fördel av den höga prestandan som många moderna maskininlärningsmetoder uppvisar och samtidigt uppfylla dagens ökade krav på transparens.
23
Matz, Filip, and Yuxiang Luo. "Explaining Automated Decisions in Practice : Insights from the Swedish Credit Scoring Industry." Thesis, KTH, Skolan för industriell teknik och management (ITM), 2021. http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-300897.
Full textAbstract:
The field of explainable artificial intelligence (XAI) has gained momentum in recent years following the increased use of AI systems across industries leading to bias, discrimination, and data security concerns. Several conceptual frameworks for how to reach AI systems that are fair, transparent, and understandable have been proposed, as well as a number of technical solutions improving some of these aspects in a research context. However, there is still a lack of studies examining the implementation of these concepts and techniques in practice. This research aims to bridge the gap between prominent theory within the area and practical implementation, exploring the implementation and evaluation of XAI models in the Swedish credit scoring industry, and proposes a three-step framework for the implementation of local explanations in practice. The research methods used consisted of a case study with the model development at UC AB as a subject and an experiment evaluating the consumers' levels of trust and system understanding as well as the usefulness, persuasive power, and usability of the explanation for three different explanation prototypes developed. The framework proposed was validated by the case study and highlighted a number of key challenges and trade-offs present when implementing XAI in practice. Moreover, the evaluation of the XAI prototypes showed that the majority of consumers prefers rulebased explanations, but that preferences for explanations is still dependent on the individual consumer. Recommended future research endeavors include studying a longterm XAI project in which the models can be evaluated by the open market and the combination of different XAI methods in reaching a more personalized explanation for the consumer.Under senare år har antalet AI implementationer stadigt ökat i flera industrier. Dessa implementationer har visat flera utmaningar kring nuvarande AI system, specifikt gällande diskriminering, otydlighet och datasäkerhet vilket lett till ett intresse för förklarbar artificiell intelligens (XAI). XAI syftar till att utveckla AI system som är rättvisa, transparenta och begripliga. Flera konceptuella ramverk har introducerats för XAI som presenterar etiska såväl som politiska perspektiv och målbilder. Dessutom har tekniska metoder utvecklats som gjort framsteg mot förklarbarhet i forskningskontext. Däremot saknas det fortfarande studier som undersöker implementationer av dessa koncept och tekniker i praktiken. Denna studie syftar till att överbrygga klyftan mellan den senaste teorin inom området och praktiken genom en fallstudie av ett företag i den svenska kreditupplysningsindustrin. Detta genom att föreslå ett ramverk för implementation av lokala förklaringar i praktiken och genom att utveckla tre förklaringsprototyper. Rapporten utvärderar även prototyperna med konsumenter på följande dimensioner: tillit, systemförståelse, användbarhet och övertalningsstyrka. Det föreslagna ramverket validerades genom fallstudien och belyste ett antal utmaningar och avvägningar som förekommer när XAI system utvecklas för användning i praktiken. Utöver detta visar utvärderingen av prototyperna att majoriteten av konsumenter föredrar regelbaserade förklaringar men indikerar även att preferenser mellan konsumenter varierar. Rekommendationer för framtida forskning är dels en längre studie, vari en XAI modell introduceras på och utvärderas av den fria marknaden, dels forskning som kombinerar olika XAI metoder för att generera mer personliga förklaringar för konsumenter.
24
Leoni, Cristian. "Interpretation of Dimensionality Reduction with Supervised Proxies of User-defined Labels." Thesis, Linnéuniversitetet, Institutionen för datavetenskap och medieteknik (DM), 2021. http://urn.kb.se/resolve?urn=urn:nbn:se:lnu:diva-105622.
Full textAbstract:
Research on Machine learning (ML) explainability has received a lot of focus in recent times. The interest, however, mostly focused on supervised models, while other ML fields have not had the same level of attention. Despite its usefulness in a variety of different fields, unsupervised learning explainability is still an open issue. In this paper, we present a Visual Analytics framework based on eXplainable AI (XAI) methods to support the interpretation of Dimensionality reduction methods. The framework provides the user with an interactive and iterative process to investigate and explain user-perceived patterns for a variety of DR methods by using XAI methods to explain a supervised method trained on the selected data. To evaluate the effectiveness of the proposed solution, we focus on two main aspects: the quality of the visualization and the quality of the explanation. This challenge is tackled using both quantitative and qualitative methods, and due to the lack of pre-existing test data, a new benchmark has been created. The quality of the visualization is established using a well-known survey-based methodology, while the quality of the explanation is evaluated using both case studies and a controlled experiment, where the generated explanation accuracy is evaluated on the proposed benchmark. The results show a strong capacity of our framework to generate accurate explanations, with an accuracy of 89% over the controlled experiment. The explanation generated for the two case studies yielded very similar results when compared with pre-existing, well-known literature on ground truths. Finally, the user experiment generated high quality overall scores for all assessed aspects of the visualization.
25
Hazarika, Subhashis. "Statistical and Machine Learning Approaches For Visualizing and Analyzing Large-Scale Simulation Data." The Ohio State University, 2019. http://rave.ohiolink.edu/etdc/view?acc_num=osu1574692702479196.
Full text
26
Saluja, Rohit. "Interpreting Multivariate Time Series for an Organization Health Platform." Thesis, KTH, Skolan för elektroteknik och datavetenskap (EECS), 2020. http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-289465.
Full textAbstract:
Machine learning-based systems are rapidly becoming popular because it has been realized that machines are more efficient and effective than humans at performing certain tasks. Although machine learning algorithms are extremely popular, they are also very literal and undeviating. This has led to a huge research surge in the field of interpretability in machine learning to ensure that machine learning models are reliable, fair, and can be held liable for their decision-making process. Moreover, in most real-world problems just making predictions using machine learning algorithms only solves the problem partially. Time series is one of the most popular and important data types because of its dominant presence in the fields of business, economics, and engineering. Despite this, interpretability in time series is still relatively unexplored as compared to tabular, text, and image data. With the growing research in the field of interpretability in machine learning, there is also a pressing need to be able to quantify the quality of explanations produced after interpreting machine learning models. Due to this reason, evaluation of interpretability is extremely important. The evaluation of interpretability for models built on time series seems completely unexplored in research circles. This thesis work focused on achieving and evaluating model agnostic interpretability in a time series forecasting problem. The use case discussed in this thesis work focused on finding a solution to a problem faced by a digital consultancy company. The digital consultancy wants to take a data-driven approach to understand the effect of various sales related activities in the company on the sales deals closed by the company. The solution involved framing the problem as a time series forecasting problem to predict the sales deals and interpreting the underlying forecasting model. The interpretability was achieved using two novel model agnostic interpretability techniques, Local interpretable model- agnostic explanations (LIME) and Shapley additive explanations (SHAP). The explanations produced after achieving interpretability were evaluated using human evaluation of interpretability. The results of the human evaluation studies clearly indicate that the explanations produced by LIME and SHAP greatly helped lay humans in understanding the predictions made by the machine learning model. The human evaluation study results also indicated that LIME and SHAP explanations were almost equally understandable with LIME performing better but with a very small margin. The work done during this project can easily be extended to any time series forecasting or classification scenario for achieving and evaluating interpretability. Furthermore, this work can offer a very good framework for achieving and evaluating interpretability in any machine learning-based regression or classification problem.Maskininlärningsbaserade system blir snabbt populära eftersom man har insett att maskiner är effektivare än människor när det gäller att utföra vissa uppgifter. Även om maskininlärningsalgoritmer är extremt populära, är de också mycket bokstavliga. Detta har lett till en enorm forskningsökning inom området tolkbarhet i maskininlärning för att säkerställa att maskininlärningsmodeller är tillförlitliga, rättvisa och kan hållas ansvariga för deras beslutsprocess. Dessutom löser problemet i de flesta verkliga problem bara att göra förutsägelser med maskininlärningsalgoritmer bara delvis. Tidsserier är en av de mest populära och viktiga datatyperna på grund av dess dominerande närvaro inom affärsverksamhet, ekonomi och teknik. Trots detta är tolkningsförmågan i tidsserier fortfarande relativt outforskad jämfört med tabell-, text- och bilddata. Med den växande forskningen inom området tolkbarhet inom maskininlärning finns det också ett stort behov av att kunna kvantifiera kvaliteten på förklaringar som produceras efter tolkning av maskininlärningsmodeller. Av denna anledning är utvärdering av tolkbarhet extremt viktig. Utvärderingen av tolkbarhet för modeller som bygger på tidsserier verkar helt outforskad i forskarkretsar. Detta uppsatsarbete fokuserar på att uppnå och utvärdera agnostisk modelltolkbarhet i ett tidsserieprognosproblem. Fokus ligger i att hitta lösningen på ett problem som ett digitalt konsultföretag står inför som användningsfall. Det digitala konsultföretaget vill använda en datadriven metod för att förstå effekten av olika försäljningsrelaterade aktiviteter i företaget på de försäljningsavtal som företaget stänger. Lösningen innebar att inrama problemet som ett tidsserieprognosproblem för att förutsäga försäljningsavtalen och tolka den underliggande prognosmodellen. Tolkningsförmågan uppnåddes med hjälp av två nya tekniker för agnostisk tolkbarhet, lokala tolkbara modellagnostiska förklaringar (LIME) och Shapley additiva förklaringar (SHAP). Förklaringarna som producerats efter att ha uppnått tolkbarhet utvärderades med hjälp av mänsklig utvärdering av tolkbarhet. Resultaten av de mänskliga utvärderingsstudierna visar tydligt att de förklaringar som produceras av LIME och SHAP starkt hjälpte människor att förstå förutsägelserna från maskininlärningsmodellen. De mänskliga utvärderingsstudieresultaten visade också att LIME- och SHAP-förklaringar var nästan lika förståeliga med LIME som presterade bättre men med en mycket liten marginal. Arbetet som utförts under detta projekt kan enkelt utvidgas till alla tidsserieprognoser eller klassificeringsscenarier för att uppnå och utvärdera tolkbarhet. Dessutom kan detta arbete erbjuda en mycket bra ram för att uppnå och utvärdera tolkbarhet i alla maskininlärningsbaserade regressions- eller klassificeringsproblem.
27
Giuliani, Luca. "Extending the Moving Targets Method for Injecting Constraints in Machine Learning." Master's thesis, Alma Mater Studiorum - Università di Bologna, 2021. http://amslaurea.unibo.it/23885/.
Full textAbstract:
Informed Machine Learning is an umbrella term that comprises a set of methodologies in which domain knowledge is injected into a data-driven system in order to improve its level of accuracy, satisfy some external constraint, and in general serve the purposes of explainability and reliability. The said topid has been widely explored in the literature by means of many different techniques. Moving Targets is one such a technique particularly focused on constraint satisfaction: it is based on decomposition and bi-level optimization and proceeds by iteratively refining the target labels through a master step which is in charge of enforcing the constraints, while the training phase is delegated to a learner. In this work, we extend the algorithm in order to deal with semi-supervised learning and soft constraints. In particular, we focus our empirical evaluation on both regression and classification tasks involving monotonicity shape constraints. We demonstrate that our method is robust with respect to its hyperparameters, as well as being able to generalize very well while reducing the number of violations on the enforced constraints. Additionally, the method can even outperform, both in terms of accuracy and constraint satisfaction, other state-of-the-art techniques such as Lattice Models and Semantic-based Regularization with a Lagrangian Dual approach for automatic hyperparameter tuning.
28
cruciani, federica. "EXplainable Artificial Intelligence: enabling AI in neurosciences and beyond." Doctoral thesis, 2023. https://hdl.handle.net/11562/1085066.
Full textAbstract:
The adoption of AI models in medicine and neurosciences has the potential to play a significant role not only in bringing scientific advancements but also in clinical decision-making. However, concerns mounts due to the eventual biases AI could have which could result in far-reaching consequences particularly in a critical field like biomedicine. It is challenging to achieve usable intelligence because not only it is fundamental to learn from prior data, extract knowledge and guarantee generalization capabilities, but also to disentangle the underlying explanatory factors in order to deeply understand the variables leading to the final decisions. There hence has been a call for approaches to open the AI `black box' to increase trust and reliability on the decision-making capabilities of AI algorithms. Such approaches are commonly referred to as XAI and are starting to be applied in medical fields even if not yet fully exploited. With this thesis we aim at contributing to enabling the use of AI in medicine and neurosciences by taking two fundamental steps: (i) practically pervade AI models with XAI (ii) Strongly validate XAI models. The first step was achieved on one hand by focusing on XAI taxonomy and proposing some guidelines specific for the AI and XAI applications in the neuroscience domain. On the other hand, we faced concrete issues proposing XAI solutions to decode the brain modulations in neurodegeneration relying on the morphological, microstructural and functional changes occurring at different disease stages as well as their connections with the genotype substrate. The second step was as well achieved by firstly defining four attributes related to XAI validation, namely stability, consistency, understandability and plausibility. Each attribute refers to a different aspect of XAI ranging from the assessment of explanations stability across different XAI methods, or highly collinear inputs, to the alignment of the obtained explanations with the state-of-the-art literature. We then proposed different validation techniques aiming at practically fulfilling such requirements. With this thesis, we contributed to the advancement of the research into XAI aiming at increasing awareness and critical use of AI methods opening the way to real-life applications enabling the development of personalized medicine and treatment by taking a data-driven and objective approach to healthcare.
29
Pereira, Filipe Inácio da Costa. "Explainable artificial intelligence - learning decision sets with sat." Master's thesis, 2018. http://hdl.handle.net/10451/34903.
Full textAbstract:
Tese de mestrado, Engenharia Informática (Interação e Conhecimento) Universidade de Lisboa, Faculdade de Ciências, 2018Artificial Intelligence is a core research topic with key significance in technological growth. With the increase of data, we have more efficient models that in a few seconds will inform us of their prediction on a given input set. The more complex techniques nowadays with better results are Black Box Models. Unfortunately, these can’t provide an explanation behind their prediction, which is a major drawback for us humans. Explainable Artificial Intelligence, whose objective is to associate explanations with decisions made by autonomous agents, breaks this lack of transparency. This can be done by two approaches, either by creating models that are interpretable by themselves or by creating frameworks that justify and interpret any prediction made by any given model. This thesis describes the implementation of two interpretable models (Decision Sets and Decision Trees) based on Logic Reasoners, either SAT (Satisfiability) or SMT (Satisfiability Modulo Theories) solvers. This work was motivated by an in-depth analysis of past work in the area of Explainable Artificial Intelligence, with the purpose of seeking applications of logic in this domain. The Decision Sets approach focuses on the training data, as does any other model, and encoding the variables and constraints as a CNF (Conjuctive Normal Form) formula which can then be solved by a SAT/SMT oracle. This approach focuses on minimizing the number of rules (or Disjunctive Normal Forms) for each binary class representation and avoiding overlap, whether it is training sample or feature-space overlap, while maintaining interpretable explanations and perfect accuracy. The Decision Tree model studied in this work consists in computing a minimum size decision tree, which would represent a 100% accurate classifier given a set of training samples. The model is based on encoding the problem as a CNF formula, which can be tackled with the efficient use of a SAT oracle.
30
Coke, Ricardo Miguel Branco. "Organization of information in neural networks." Master's thesis, 2021. http://hdl.handle.net/10348/11093.
Full textAbstract:
Dissertação submetida à UNIVERSIDADE DE TRAS-OS-MONTES E ALTO DOURO para obtenção do grau de MESTRE em Engenharia Electrotécnica e de ComputadoresO mundo está a evoluir e a tecnologia tal como a conhecemos está a ultrapassar barreiras que antes pensávamos não poderem ser quebradas. A Inteligência Artificial tem sido uma das áreas que tem alcançado um nível de sucesso que proporciona a melhor das expectativas em muitas tarefas em todas as áreas de trabalho. No domínio de Machine Learning, as redes neuronais têm sido um dos conceitos que se tem destacado, resultando em precisões extremamente elevadas em problemas provenientes de todos os sectores. No entanto, a falta de interpretabilidade é um dos problemas inerentes que estas redes complexas com capacidade de produzir predições trouxe. Os modelos actuais de redes neuronais são muitas vezes considerados como ”caixas negras” devido à falta de ”explicabilidade” e transparência, uma vez que o conhecimento obtido pela aprendizagem do modelo com os dados fornecidos como entrada, são imprevisíveis e em sistemas complexos, inexplicáveis. A complexidade de uma rede está intrinsecamente ligada a alguns parâmetros como o número de camadas escondidas, função de ativação escolhida, que influencia o quão flexível é a rede na classificação de padrões, podendo fazer as devidas classificações em espaços dimensionais maiores com maior facilidade, com custo mais elevado em recursos, tempo e menos interpretabilidade ao computar o problema. Desde então sempre se tentou arranjar soluções alternativas, modificações à estrutura e método de aprendizagem das redes neuronais de modo a compensar os recursos adicionais necessitados. A falta de compreensão sobre a razão pela qual os modelos chegam a determinadas conclusões prejudica a implementação destes modelos em áreas críticas como medicina de precisão, direito ou finanças, onde o raciocínio para as conclusões apresentadas é extremamente necessário. Este problema levou à criação de um novo subcampo no domínio de Machine Learning onde a Inteligência Artificial explicável é uma prioridade. Esta dissertação apresenta uma revisão da literatura existente e contribuições disponíveis no campo da Inteligência Artificial explicável e no campo de modificações da estrutura das redes neuronais típicas. Baseado na investigação mais recente do Professor Paulo Salgado, que foi providenciada para compreensão e validação, sobre canais complementares de informação em Redes Neuronais e um algoritmo baseado neste modelo complementar, uma primeira contribuição desta dissertação foi formalizar os aspectos teóricos dessa investigação assim como a sua compreensão, seguido da validação do modelo com métricas típicas do domínio de Machine Learning.
The world is evolving and technology as we know its surpassing barriers that we once thought couldn’t be broken. Artificial Intelligence has been one of the areas that has achieved a level of success that delivers the best of expectations in many tasks across the field. In the domain of Machine Learning, Neural Networks have been one of the concepts that have stood out, providing extremely high accuracy when solving problems across all work sectors. However, the lack of interpretability is one of the inherent problems it has brought. The present neural network models are often considered as ”black-boxes” derived from the lack of explainability and transparency, since the knowledge the structure learns from the data provided as input is unpredictable and in complex systems, unexplainable. The neural network complexity is intrinsically associated with some parameters, such as number of hidden layers, activation function, which influence how flexible are the network pattern classifications allowing for easier classifications in higher spatial dimensions at the cost of more resources, time and interpretability when computing the problem. In order to solve the increased computational needs and lack of interpretability, new solutions and structure modifications to the typical networks have been researched and implemented. The lack of understanding on why the models reaches certain conclusions impairs these models to be deployed in critical areas such as precision medicine, law or finances where reasoning for the presented conclusions are needed. This problem led to the creation of a new sub-field in the domain of Machine Learning where explainable artificial intelligence is the priority. This dissertation presents a review of the existing literature and contributions available in the field of XAI and regarding structural modifications to the typical networks. Based on the most recent Professor Paulo Salgado’s investigation, which were provided for study and validation, about complementary information channels on Neural Networks and an algorithm based on this complementary sub model, the comprehension and formalization of the theoretical methods was another contribute of this dissertation followed by the validation of the model with typical Machine Learning performance metrics.
31
"Towards Building an Intelligent Tutor for Gestural Languages using Concept Level Explainable AI." Doctoral diss., 2020. http://hdl.handle.net/2286/R.I.57347.
Full textAbstract:
abstract: Languages, specially gestural and sign languages, are best learned in immersive environments with rich feedback. Computer-Aided Language Learning (CALL) solu- tions for spoken languages have successfully incorporated some feedback mechanisms, but no such solution exists for signed languages. Computer Aided Sign Language Learning (CASLL) is a recent and promising field of research which is made feasible by advances in Computer Vision and Sign Language Recognition(SLR). Leveraging existing SLR systems for feedback based learning is not feasible because their decision processes are not human interpretable and do not facilitate conceptual feedback to learners. Thus, fundamental research is needed towards designing systems that are modular and explainable. The explanations from these systems can then be used to produce feedback to aid in the learning process.
In this work, I present novel approaches for the recognition of location, movement and handshape that are components of American Sign Language (ASL) using both wrist-worn sensors as well as webcams. Finally, I present Learn2Sign(L2S), a chat- bot based AI tutor that can provide fine-grained conceptual feedback to learners of ASL using the modular recognition approaches. L2S is designed to provide feedback directly relating to the fundamental concepts of ASL using an explainable AI. I present the system performance results in terms of Precision, Recall and F-1 scores as well as validation results towards the learning outcomes of users. Both retention and execution tests for 26 participants for 14 different ASL words learned using learn2sign is presented. Finally, I also present the results of a post-usage usability survey for all the participants. In this work, I found that learners who received live feedback on their executions improved their execution as well as retention performances. The average increase in execution performance was 28% points and that for retention was 4% points.Dissertation/Thesis
Doctoral Dissertation Engineering 2020
32
"Explainable AI in Workflow Development and Verification Using Pi-Calculus." Doctoral diss., 2020. http://hdl.handle.net/2286/R.I.55566.
Full textAbstract:
abstract: Computer science education is an increasingly vital area of study with various challenges that increase the difficulty level for new students resulting in higher attrition rates. As part of an effort to resolve this issue, a new visual programming language environment was developed for this research, the Visual IoT and Robotics Programming Language Environment (VIPLE). VIPLE is based on computational thinking and flowchart, which reduces the needs of memorization of detailed syntax in text-based programming languages. VIPLE has been used at Arizona State University (ASU) in multiple years and sections of FSE100 as well as in universities worldwide. Another major issue with teaching large programming classes is the potential lack of qualified teaching assistants to grade and offer insight to a student’s programs at a level beyond output analysis.
In this dissertation, I propose a novel framework for performing semantic autograding, which analyzes student programs at a semantic level to help students learn with additional and systematic help. A general autograder is not practical for general programming languages, due to the flexibility of semantics. A practical autograder is possible in VIPLE, because of its simplified syntax and restricted options of semantics. The design of this autograder is based on the concept of theorem provers. To achieve this goal, I employ a modified version of Pi-Calculus to represent VIPLE programs and Hoare Logic to formalize program requirements. By building on the inference rules of Pi-Calculus and Hoare Logic, I am able to construct a theorem prover that can perform automated semantic analysis. Furthermore, building on this theorem prover enables me to develop a self-learning algorithm that can learn the conditions for a program’s correctness according to a given solution program.Dissertation/Thesis
Doctoral Dissertation Computer Science 2020
33
Neves, Maria Inês Lourenço das. "Opening the black-box of artificial intelligence predictions on clinical decision support systems." Master's thesis, 2021. http://hdl.handle.net/10362/126699.
Full textAbstract:
Cardiovascular diseases are the leading global death cause. Their treatment and prevention
rely on electrocardiogram interpretation, which is dependent on the physician’s variability.
Subjectiveness is intrinsic to electrocardiogram interpretation and hence, prone to
errors. To assist physicians in making precise and thoughtful decisions, artificial intelligence
is being deployed to develop models that can interpret extent datasets and provide
accurate decisions. However, the lack of interpretability of most machine learning models
stands as one of the drawbacks of their deployment, particularly in the medical domain.
Furthermore, most of the currently deployed explainable artificial intelligence methods
assume independence between features, which means temporal independence when dealing
with time series. The inherent characteristic of time series cannot be ignored as it
carries importance for the human decision making process.
This dissertation focuses on the explanation of heartbeat classification using several
adaptations of state-of-the-art model-agnostic methods, to locally explain time series classification.
To address the explanation of time series classifiers, a preliminary conceptual
framework is proposed, and the use of the derivative is suggested as a complement to
add temporal dependency between samples. The results were validated on an extent
public dataset, through the 1-D Jaccard’s index, which consists of the comparison of the
subsequences extracted from an interpretable model and the explanation methods used.
Secondly, through the performance’s decrease, to evaluate whether the explanation fits
the model’s behaviour. To assess models with distinct internal logic, the validation was
conducted on a more transparent model and more opaque one in both binary and multiclass
situation. The results show the promising use of including the signal’s derivative
to introduce temporal dependency between samples in the explanations, for models with
simpler internal logic.As doenças cardiovasculares são, a nível mundial, a principal causa de morte e o seu tratamento e prevenção baseiam-se na interpretação do electrocardiograma. A interpretação do electrocardiograma, feita por médicos, é intrinsecamente subjectiva e, portanto, sujeita a erros. De modo a apoiar a decisão dos médicos, a inteligência artificial está a ser usada para desenvolver modelos com a capacidade de interpretar extensos conjuntos de dados e fornecer decisões precisas. No entanto, a falta de interpretabilidade da maioria dos modelos de aprendizagem automática é uma das desvantagens do recurso à mesma, principalmente em contexto clínico. Adicionalmente, a maioria dos métodos inteligência artifical explicável assumem independência entre amostras, o que implica a assunção de independência temporal ao lidar com séries temporais. A característica inerente das séries temporais não pode ser ignorada, uma vez que apresenta importância para o processo de tomada de decisão humana. Esta dissertação baseia-se em inteligência artificial explicável para tornar inteligível a classificação de batimentos cardíacos, através da utilização de várias adaptações de métodos agnósticos do estado-da-arte. Para abordar a explicação dos classificadores de séries temporais, propõe-se uma taxonomia preliminar, e o uso da derivada como um complemento para adicionar dependência temporal entre as amostras. Os resultados foram validados para um conjunto extenso de dados públicos, por meio do índice de Jaccard em 1-D, com a comparação das subsequências extraídas de um modelo interpretável e os métodos inteligência artificial explicável utilizados, e a análise de qualidade, para avaliar se a explicação se adequa ao comportamento do modelo. De modo a avaliar modelos com lógicas internas distintas, a validação foi realizada usando, por um lado, um modelo mais transparente e, por outro, um mais opaco, tanto numa situação de classificação binária como numa situação de classificação multiclasse. Os resultados mostram o uso promissor da inclusão da derivada do sinal para introduzir dependência temporal entre as amostras nas explicações fornecidas, para modelos com lógica interna mais simples.
34
"Foundations of Human-Aware Planning -- A Tale of Three Models." Doctoral diss., 2018. http://hdl.handle.net/2286/R.I.51791.
Full textAbstract:
abstract: A critical challenge in the design of AI systems that operate with humans in the loop is to be able to model the intentions and capabilities of the humans, as well as their beliefs and expectations of the AI system itself. This allows the AI system to be "human- aware" -- i.e. the human task model enables it to envisage desired roles of the human in joint action, while the human mental model allows it to anticipate how its own actions are perceived from the point of view of the human. In my research, I explore how these concepts of human-awareness manifest themselves in the scope of planning or sequential decision making with humans in the loop. To this end, I will show (1) how the AI agent can leverage the human task model to generate symbiotic behavior; and (2) how the introduction of the human mental model in the deliberative process of the AI agent allows it to generate explanations for a plan or resort to explicable plans when explanations are not desired. The latter is in addition to traditional notions of human-aware planning which typically use the human task model alone and thus enables a new suite of capabilities of a human-aware AI agent. Finally, I will explore how the AI agent can leverage emerging mixed-reality interfaces to realize effective channels of communication with the human in the loop.Dissertation/Thesis
Doctoral Dissertation Computer Science 2018
35
Brito, João Pedro da Cruz. "Deep Adversarial Frameworks for Visually Explainable Periocular Recognition." Master's thesis, 2021. http://hdl.handle.net/10400.6/11850.
Full textAbstract:
Machine Learning (ML) models have pushed stateoftheart performance closer to (and
even beyond) human level. However, the core of such algorithms is usually latent and
hardly understandable. Thus, the field of Explainability focuses on researching and adopting techniques that can explain the reasons that support a model’s predictions. Such explanations of the decisionmaking process would help to build trust between said model
and the human(s) using it. An explainable system also allows for better debugging, during
the training phase, and fixing, upon deployment. But why should a developer devote time
and effort into refactoring or rethinking Artificial Intelligence (AI) systems, to make them
more transparent? Don’t they work just fine?
Despite the temptation to answer ”yes”, are we really considering the cases where these
systems fail? Are we assuming that ”almost perfect” accuracy is good enough? What if,
some of the cases where these systems get it right, were just a small margin away from
a complete miss? Does that even matter? Considering the evergrowing presence of ML
models in crucial areas like forensics, security and healthcare services, it clearly does.
Motivating these concerns is the fact that powerful systems often operate as blackboxes,
hiding the core reasoning underneath layers of abstraction [Gue]. In this scenario, there
could be some seriously negative outcomes if opaque algorithms gamble on the presence
of tumours in Xray images or the way autonomous vehicles behave in traffic.
It becomes clear, then, that incorporating explainability with AI is imperative. More recently, the politicians have addressed this urgency through the General Data Protection
Regulation (GDPR) [Com18]. With this document, the European Union (EU) brings forward several important concepts, amongst which, the ”right to an explanation”. The definition and scope are still subject to debate [MF17], but these are definite strides to formally
regulate the explainable depth of autonomous systems.
Based on the preface above, this work describes a periocular recognition framework that
not only performs biometric recognition but also provides clear representations of the features/regions that support a prediction. Being particularly designed to explain nonmatch
(”impostors”) decisions, our solution uses adversarial generative techniques to synthesise
a large set of ”genuine” image pairs, from where the most similar elements with respect to
a query are retrieved. Then, assuming the alignment between the query/retrieved pairs,
the elementwise differences between the query and a weighted average of the retrieved
elements yields a visual explanation of the regions in the query pair that would have to
be different to transform it into a ”genuine” pair. Our quantitative and qualitative experiments validate the proposed solution, yielding recognition rates that are similar to the
stateoftheart, while adding visually pleasing explanations.
36
Palesi, Luciano Alessandro Ipsaro. "Human Centered Big Data Analytics for Smart Applications." Doctoral thesis, 2022. http://hdl.handle.net/2158/1282883.
Full textAbstract:
This thesis is concerned with Smart Applications.
Smart applications are all those applications that incorporate data-driven, actionable insights in the user experience, and they allow in different contexts users to complete actions or make decisions efficiently. The differences between smart applications and traditional applications are mainly that the former are dynamic and evolve on the basis of intuition, user feedback or new data.
Moreover, smart applications are data-driven and linked to the context of use. There are several aspects to be considered in the development of intelligent applications, such as machine learning algorithms for producing insights, privacy, data security and ethics.
The purpose of this thesis is to study and develop human centered algorithms and systems in different contexts (retail, industry, environment and smart city) with particular attention to big data analysis and prediction techniques. The second purpose of this thesis is to study and develop techniques for the interpretation of results in order to make artificial intelligence algorithms "explainable".
Finally, the third and last purpose is to develop solutions in GDPR compliant environments and then secure systems that respect user privacy.
37
Ribeiro, Manuel António de Melo Chinopa de Sousa. "Neural and Symbolic AI - mind the gap! Aligning Artificial Neural Networks and Ontologies." Master's thesis, 2020. http://hdl.handle.net/10362/113651.
Full textAbstract:
Artificial neural networks have been the key to solve a variety of different problems.
However, neural network models are still essentially regarded as black boxes, since they
do not provide any human-interpretable evidence as to why they output a certain re sult. In this dissertation, we address this issue by leveraging on ontologies and building
small classifiers that map a neural network’s internal representations to concepts from
an ontology, enabling the generation of symbolic justifications for the output of neural
networks. Using two image classification problems as testing ground, we discuss how to
map the internal representations of a neural network to the concepts of an ontology, exam ine whether the results obtained by the established mappings match our understanding
of the mapped concepts, and analyze the justifications obtained through this method.
38
Ciravegna, Gabriele. "On the Two-fold Role of Logic Constraints in Deep Learning." Doctoral thesis, 2022. http://hdl.handle.net/2158/1264916.
Full textAbstract:
Deep Learning (DL) is a special class of Artificial Intelligence (AI) algorithms, studying the training of Deep Neural Networks (DNNs). Thanks to the modularity of their structure, these models are effective in a variety of problems ranging from computer vision to speech recognition. Particularly in the last few years, DL has achieved impressive results. Nonetheless, the excitement around the field may remain disappointed since there are still many open issues.
In this thesis, we consider the Learning from Constraints framework. In this setting, learning is conceived as the problem of finding task functions while respecting a number of constraints representing the available knowledge. This setting allows considering different types of knowledge (including, but not exclusively, the supervisions) and mitigating some of the DL limits. DNN deployment, indeed, is still precluded in those contexts where manual labelling is expensive. Active Learning aims at solving this problem by requiring supervision only on few unlabelled samples. In this scenario, we propose to take consider domain knowledge. Indeed, the relationships among classes offer a way to spot incoherent predictions, i.e., predictions where the model may most likely need supervision. We develop a framework where first-order-logic knowledge is converted into constraints and their violation is checked as a guide for sample selection. Another DL limit is the fragility of DNNs when facing adversarial examples, carefully perturbed samples causing misclassifications at test time. As in the previous case, we propose to employ domain knowledge since it offers a natural guide to detect adversarial examples. Indeed, while the domain knowledge is fulfilled over the training data, the same does not hold true outside this distribution. Therefore, a constrained classifier can naturally reject predictions associated to incoherent predictions, i.e., in this case, adversarial examples.
While some relationships are known properties of the considered environments, DNNs can also autonomously develop new relation patterns. Therefore, we also propose a novel Learning of Constraints formulation which aims at understanding which logic constraints are present among the task functions. This also allow explaining DNNs, otherwise commonly considered black-box classifiers. Indeed, the lack of transparency is a major limit of DL, preventing its application in many safety-critical domains. In a first case, we propose a pair of neural networks, where one learns the relationships among the outputs of the other one, and provides First-Order Logic (FOL)-based descriptions. Different typologies of explanations are evaluated in distinct experiments, showing that the proposed approach discovers new knowledge and can improve the classifier performance. In a second case, we propose an end-to-end differentiable approach, extracting logic explanations from the same classifier. The method relies on an entropy-based layer which automatically identifies the most relevant concepts. This enables the distillation of concise logic explanations in several safety-critical domains, outperforming state-of-the-art white-box models.
39
Olsson, Anton, and Gustaf Joelsson. "Requirements Analysis For AI Solutions : A Study On How Requirements Analysis Is Executed When Developing AI Solutions." Thesis, 2019. http://urn.kb.se/resolve?urn=urn:nbn:se:hb:diva-25542.
Full textAbstract:
Requirements analysis is an essential part of the System Development Life Cycle (SDLC) in order to achieve success in a software development project. There are several methods, techniques and frameworks used when expressing, prioritizing and managing requirements in IT projects. It is widely established that it is difficult to determine requirements for traditional systems, so a question naturally arises on how the requirements analysis is executed as AI solutions (that even fewer individuals can grasp) are being developed. Little research has been made on how the vital requirements phase is executed during development of AI solutions. This research aims to investigate the requirements analysis phase during the development of AI solutions. To explore this topic, an extensive literature review was made, and in order to collect new information, a number of interviews were performed with five suitable organizations (i.e, organizations that develop AI solutions). The results from the research concludes that the requirements analysis does not differ between development of AI solutions in comparison to development of traditional systems. However, the research showed that there were some deviations that can be deemed to be particularly unique for the development of AI solutions that affects the requirements analysis. These are:(1) the need for an iterative and agile systems development process, with an associated iterative and agile requirements analysis, (2) the importance of having a large set of quality data, (3) the relative deprioritization of user involvement, and (4) the difficulty of establishing timeframe, results/feasibility and the behavior of the AI solution beforehand.