Literatura científica selecionada sobre o tema "Combinatorial optimization layers"
Crie uma referência precisa em APA, MLA, Chicago, Harvard, e outros estilos
Consulte a lista de atuais artigos, livros, teses, anais de congressos e outras fontes científicas relevantes para o tema "Combinatorial optimization layers".
Ao lado de cada fonte na lista de referências, há um botão "Adicionar à bibliografia". Clique e geraremos automaticamente a citação bibliográfica do trabalho escolhido no estilo de citação de que você precisa: APA, MLA, Harvard, Chicago, Vancouver, etc.
Você também pode baixar o texto completo da publicação científica em formato .pdf e ler o resumo do trabalho online se estiver presente nos metadados.
Artigos de revistas sobre o assunto "Combinatorial optimization layers"
1
Kim, Bomi, Taehyeon Kim e Yoonsik Choe. "Bayesian Optimization Based Efficient Layer Sharing for Incremental Learning". Applied Sciences 11, n.º 5 (1 de março de 2021): 2171. http://dx.doi.org/10.3390/app11052171.
Texto completo da fonteResumo:
Incremental learning is a methodology that continuously uses the sequential input data to extend the existing network’s knowledge. The layer sharing algorithm is one of the representative methods which leverages general knowledge by sharing some initial layers of the existing network. To determine the performance of the incremental network, it is critical to estimate how much the initial convolutional layers in the existing network can be shared as the fixed feature extractors. However, the existing algorithm selects the sharing configuration through improper optimization strategy but a brute force manner such as searching for all possible sharing layers case. This is a non-convex and non-differential problem. Accordingly, this can not be solved using powerful optimization techniques such as the gradient descent algorithm or other convex optimization problem, and it leads to high computational complexity. To solve this problem, we firstly define this as a discrete combinatorial optimization problem, and propose a novel efficient incremental learning algorithm-based Bayesian optimization, which guarantees the global convergence in a non-convex and non-differential optimization. Additionally, our proposed algorithm can adaptively find the optimal number of sharing layers via adjusting the threshold accuracy parameter in the proposed loss function. With the proposed method, the global optimal sharing layer can be found in only six or eight iterations without searching for all possible layer cases. Hence, the proposed method can find the global optimal sharing layers by utilizing Bayesian optimization, which achieves both high combined accuracy and low computational complexity.
2
Deng, Beiming, Lijia Xia e Hui Cheng. "Bilayer Real Time Multi-Robot Communication Maintenance Deployment Framework for Robot Swarms". Journal of Physics: Conference Series 2850, n.º 1 (1 de setembro de 2024): 012011. http://dx.doi.org/10.1088/1742-6596/2850/1/012011.
Texto completo da fonteResumo:
Abstract The communication maintenance problem of robot swarms is important to multi-robot control in applications like rescue and area exploration. In this paper, we propose a robot-relay-based framework to keep the robot swarm connected from the view of Line-of-Sight communication. This framework mainly consists of a graphic calculation layer and a numerical optimization layer. The combination of the two layers is intent on blending the combinatorial identity of the problem and the advantage of the differentiable nature of this numerical optimization problem and forming a pipeline for generating a swarm deployment solution with low communication loss, low energy consumption, and high efficiency.
3
Lábadi, Zoltán, Noor Taha Ismaeel, Péter Petrik e Miklós Fried. "Compositional Optimization of Sputtered SnO2/ZnO Films for High Coloration Efficiency". International Journal of Molecular Sciences 25, n.º 19 (8 de outubro de 2024): 10801. http://dx.doi.org/10.3390/ijms251910801.
Texto completo da fonteResumo:
We performed an electrochromic investigation to optimize the composition of reactive magnetron-sputtered mixed layers of zinc oxide and tin oxide (ZnO-SnO2). Deposition experiments were conducted as a combinatorial material synthesis approach. The binary system for the samples of SnO2-ZnO represented the full composition range. The coloration efficiency (CE) was determined for the mixed oxide films with the simultaneous measurement of layer transmittance, in a conventional three-electrode configuration, and an electric current was applied by using organic propylene carbonate electrolyte cells. The optical parameters and composition were measured and mapped by using spectroscopic ellipsometry (SE). Scanning Electron Microscopy (SEM) and Energy-Dispersive X-ray Spectroscopy (EDS) measurements were carried out to check the SE results, for (TiO2-SnO2). Pure metal targets were placed separately from each other, and the indium–tin-oxide (ITO)-covered glass samples and Si-probes on a glass holder were moved under the two separated targets (Zn and Sn) in a reactive argon–oxygen (Ar-O2) gas mixture. This combinatorial process ensured that all the compositions (from 0 to 100%) were achieved in the same sputtering chamber after one sputtering preparation cycle. The CE data evaluated from the electro-optical measurements plotted against the composition displayed a characteristic maximum at around 29% ZnO. The accuracy of our combinatorial approach was 5%.
4
Chebakov, Sergey V., e Liya V. Serebryanaya. "Finding algorithm of optimal subset structure based on the Pareto layers in the knapsack problem". Journal of the Belarusian State University. Mathematics and Informatics, n.º 2 (30 de julho de 2020): 97–104. http://dx.doi.org/10.33581/2520-6508-2020-2-97-104.
Texto completo da fonteResumo:
An algorithm is developed for finding the structure of the optimal subset in the knapsack problem based on the proposed multicriteria optimization model. A two-criteria relation of preference between elements of the set of initial data is introduced. This set has been split into separate Pareto layers. The depth concept of the elements dominance of an individual Pareto layer is formulated. Based on it, conditions are determined under which the solution to the knapsack problem includes the first Pareto layers. They are defined on a given set of initial data. The structure of the optimal subset is presented, which includes individual Pareto layers. Pareto layers are built in the introduced preference space. This does not require algorithms for enumerating the elements of the initial set. Such algorithms are used when finding only some part of the optimal subset. This reduces the number of operations required to solve the considered combinatorial problem. The method for determining the found Pareto layers shows that the number of operations depends on the volume of the knapsack and the structure of the Pareto layers, into which the set of initial data in the entered two-criteria space is divided.
5
Wu, Kee Rong, e Chung Wei Yeh. "Solution to the 0-1 Multidimensional Knapsack Problem Based on DNA Computation". Applied Mechanics and Materials 58-60 (junho de 2011): 1767–72. http://dx.doi.org/10.4028/www.scientific.net/amm.58-60.1767.
Texto completo da fonteResumo:
We proposed a two-layer scheme of Deoxyribonucleic acid (DNA) based computation, DNA-01MKP, to solve the typical NP-hard combinatorial optimization problem, 0-1 multidimensional knapsack problem (0-1 MKP). DNA-01MKP consists of two layers of procedures: (1) translation of the problem equations to strands and (2) solution of problems. For layer 1, we designed flexible well-formatted strands to represent the problem equations; for layer 2, we constructed the DNA algorithms to solve the 0-1 MKP. Our results revealed that this molecular computation scheme is able to solve the complicated operational problem with a reasonable time complexity of O(n×k), though it needs further experimental verification in the future. By adjusting the DNA-based procedures, the scheme may be used to resolve different NP-hard problems.
6
Cao, Zhanmao, Qisong Huang e Chase Wu. "Maximize concurrent data flows in multi-radio multi-channel wireless mesh networks". Computer Science and Information Systems 17, n.º 3 (2020): 759–77. http://dx.doi.org/10.2298/csis200216019c.
Texto completo da fonteResumo:
Multi-radio multi-channel (MRMC) wireless mesh networks (WMNs) have emerged as the broadband networks to provide access to the Internet for ubiquitous computing with the support for a large number of data flows. Many applications in WMNs can be abstracted as a multi-flow coexistence problem to carry out multiple concurrent data transfers. More specifically, links in different channel layers must be concatenated to compose multiple data transfer paths based on nodes? free interfaces and available channels. This is typically formulated as a combinatorial optimization problem with various stages including channel assignment, path computing, and link scheduling. This paper analyzes traffic behaviors and designs a coexisting algorithm to maximize the number of concurrent data flows. Simulations are conducted in combinatorial cases of channel and radio with various traffic requests of multiple pairs. The experimental results show the efficacy of the coexisting algorithm over a randomly generated topology. This scheme can be used to develop routing and scheduling solutions for various multi-flow network applications through prior computing.
7
Rahman, B. S., e D. K. Lieu. "Optimization of Magnetic Pole Geometry for Field Harmonic Control in Electric Motors". Journal of Vibration and Acoustics 116, n.º 2 (1 de abril de 1994): 173–78. http://dx.doi.org/10.1115/1.2930409.
Texto completo da fonteResumo:
A principal source of vibration in permanent magnet motors and generators is the induced stress from the rotating permanent magnets. The harmonic content of this forcing function may excite resonant modes of vibration in the motor or surrounding structure. Thus attenuation of specific harmonics is of considerable interest. This paper describes a method for optimal shaping of the permanent magnets to eliminate one or more of these harmonics. The analytical model for an optimized 4-pole motor consisted of segmented PMs and a solid ring stator. The permanent magnets were modeled as a number of thin radially cut annular layers with specific sector angles. Changing the shape of the PMs resulted in a different flux density field and thus a different frequency spectrum of the forcing function. Attenuation of specified higher harmonics could be achieved at the expense of increasing other harmonics. For a 4-pole motor, the optimization algorithm was fairly successful at eliminating any one of the 8th, 12th or 16th harmonics. The algorithm used was developed to solve combinatorial optimization problems, and drew heavily upon principles from statistical mechanics. The final pole geometry is dependent upon the choice of the cost function used in the optimization algorithm.
8
Inga, Esteban, Juan Inga e Andres Ortega. "Novel Approach Sizing and Routing of Wireless Sensor Networks for Applications in Smart Cities". Sensors 21, n.º 14 (9 de julho de 2021): 4692. http://dx.doi.org/10.3390/s21144692.
Texto completo da fonteResumo:
Citizens are expected to require the growth of multiple Internet of Things (IoT) -based applications to improve public and private services. According to their concept, smart cities seek to improve the efficiency, reliability, and resilience of these services. Consequently, this paper searches for a new vision for resolving problems related to the quick deployment of a wireless sensor network (WSN) by using a sizing model and considering the capacity and coverage of the concentrators. Additionally, three different routing models of these technology resources are presented as alternatives for each WSN deployment to ensure connectivity between smart meters and hubs required for smart metering. On the other hand, these solutions must reduce costs when this type of wireless communication network is deployed. The present work proposes various optimization models that consider the physical and network layers in order to integrate different wireless communication technologies, thus reducing costs in terms of the minimum number of data aggregation points. Using a heterogeneous wireless network can reduce resource costs and energy consumption in comparison to a single cellular technology, as proposed in previous works. This work proposes a sizing model and three different models for routing wireless networks. In each case, constraints are evaluated and can be associated with different real-world scenarios. This document provides an optimization model that encompasses all of the proposed constraints; due to the combinatorial nature of the problem, this would require a heuristic technique.
9
Farhi, Edward, Jeffrey Goldstone, Sam Gutmann e Leo Zhou. "The Quantum Approximate Optimization Algorithm and the Sherrington-Kirkpatrick Model at Infinite Size". Quantum 6 (7 de julho de 2022): 759. http://dx.doi.org/10.22331/q-2022-07-07-759.
Texto completo da fonteResumo:
The Quantum Approximate Optimization Algorithm (QAOA) is a general-purpose algorithm for combinatorial optimization problems whose performance can only improve with the number of layers p. While QAOA holds promise as an algorithm that can be run on near-term quantum computers, its computational power has not been fully explored. In this work, we study the QAOA applied to the Sherrington-Kirkpatrick (SK) model, which can be understood as energy minimization of n spins with all-to-all random signed couplings. There is a recent classical algorithm by Montanari that, assuming a widely believed conjecture, can efficiently find an approximate solution for a typical instance of the SK model to within (1−ϵ) times the ground state energy. We hope to match its performance with the QAOA.Our main result is a novel technique that allows us to evaluate the typical-instance energy of the QAOA applied to the SK model. We produce a formula for the expected value of the energy, as a function of the 2p QAOA parameters, in the infinite size limit that can be evaluated on a computer with O(16p) complexity. We evaluate the formula up to p=12, and find that the QAOA at p=11 outperforms the standard semidefinite programming algorithm. Moreover, we show concentration: With probability tending to one as n→∞, measurements of the QAOA will produce strings whose energies concentrate at our calculated value. As an algorithm running on a quantum computer, there is no need to search for optimal parameters on an instance-by-instance basis since we can determine them in advance. What we have here is a new framework for analyzing the QAOA, and our techniques can be of broad interest for evaluating its performance on more general problems where classical algorithms may fail.
10
Zhang, Xu, Pan Guo, Hua Zhang e Jin Yao. "Hybrid Particle Swarm Optimization Algorithm for Process Planning". Mathematics 8, n.º 10 (11 de outubro de 2020): 1745. http://dx.doi.org/10.3390/math8101745.
Texto completo da fonteResumo:
Process planning is a typical combinatorial optimization problem. When the scale of the problem increases, combinatorial explosion occurs, which makes it difficult for traditional precise algorithms to solve the problem. A hybrid particle swarm optimization (HPSO) algorithm is proposed in this paper to solve problems of process planning. A hierarchical coding method including operation layer, machine layer and logic layer is designed in this algorithm. Each layer of coding corresponds to the decision of a sub-problem of process planning. Several genetic operators of the genetic algorithm are designed to replace the update formula of particle position and velocity in the particle swarm optimization algorithm. The results of the benchmark example in case study show that the algorithm proposed in this paper has better performance.
Teses / dissertações sobre o assunto "Combinatorial optimization layers"
1
Bouvier, Louis. "Apprentissage structuré et optimisation combinatoire : contributions méthodologiques et routage d'inventaire chez Renault". Electronic Thesis or Diss., Marne-la-vallée, ENPC, 2024. http://www.theses.fr/2024ENPC0046.
Texto completo da fonteResumo:
Cette thèse découle des défis de recherche opérationnelle de la chaîne logistique Renault. Pour y répondre, nous apportons des contributions à l’architecture et à l’entraînement des réseaux neuronaux avec des couches d’optimisation combinatoire (CO). Nous les combinons avec de nouvelles matheuristiques pour aborder les problèmes de routage d’inventaire de Renault. La Partie I est dédiée aux applications des réseaux neuronaux avec des couches CO en recherche opérationnelle. Nous introduisons une méthode pour approximer les contraintes. Nous utilisons de telles couches pour encoder des politiques pour des processus de décision markoviens à grands espaces d’états et d’actions. Alors que la plupart des études sur les couches CO reposent sur l’apprentissage super- visé, nous introduisons un schéma primal-dual pour la minimisation du risque empirique. Notre algorithme est compatible avec l’apprentissage profond, adapté à de grands espaces combinatoires, et générique. La Partie II est dédiée à la logistique retour des emballages Renault en Europe. Notre politique pour les décisions opérationnelles est basée sur une nouvelle matheuristique pour la variante déterministe du problème. Nous montrons son efficacité sur des instances à grande échelle, que nous publions, avec notre code et nos solutions. Une version de notre politique est utilisée quotidiennement en production depuis mars 2023. Nous abordons aussi la contractualisation de routes au niveau tactique. L’ampleur du problème empêche l’utilisation d’approches classiques d’optimisation stochastique. Nous introduisons un nouvel algorithme basé sur les contributions de la Partie I pour la minimisation du risque empiriqueThis thesis stems from operations research challenges faced by Renault supply chain. Toaddress them, we make methodological contributions to the architecture and training of neural networks with combinatorial optimization (CO) layers. We combine them with new matheuristics to solve Renault’s industrial inventory routing problems.In Part I, we detail applications of neural networks with CO layers in operations research. We notably introduce a methodology to approximate constraints. We also solve some off- policy learning issues that arise when using such layers to encode policies for Markov decision processes with large state and action spaces. While most studies on CO layers rely on supervised learning, we introduce a primal-dual alternating minimization scheme for empirical risk minimization. Our algorithm is deep learning-compatible, scalable to large combinatorial spaces, and generic. In Part II, we consider Renault European packaging return logistics. Our rolling-horizon policy for the operational-level decisions is based on a new large neighborhood search for the deterministic variant of the problem. We demonstrate its efficiency on large-scale industrialinstances, that we release publicly, together with our code and solutions. We combine historical data and experts’ predictions to improve performance. A version of our policy has been used daily in production since March 2023. We also consider the tactical-level route contracting process. The sheer scale of this industrial problem prevents the use of classic stochastic optimization approaches. We introduce a new algorithm based on methodological contributions of Part I for empirical risk minimization
2
Kelareva, Galina Vladislavovna. "Development and applications of multi-layered genetic algorithms to multi-dimensional optimisation problems". Thesis, 2003. https://eprints.utas.edu.au/20554/7/whole_KelarevaGalinaVladislavovna2003.pdf.
Texto completo da fonteResumo:
Genetic algorithms represent a global optimisation method, imitating the principles of
natural evolution: selection and survival of the fittest. Genetic algorithms operate on a
randomly initialised population of potential solutions to a problem. The solutions
develop by passing valuable genetic information to succeeding generations.
Genetic algorithms are known as a robust technique suitable for a variety of
optimisation problems. However, when applied to complex combinatorial problems
with multiple parameters, conventional genetic algorithms are usually slow and
ineffective due to the large search space.
This thesis proposes a novel approach to the development of a genetic algorithm and
applies this approach to a maintenance scheduling problem in a power generation
system. Problem specific knowledge is utilised to divide the problem into several layers,
with each layer representing a part of the initial problem. Solutions are progressively
developed, with each layer algorithm finding partial solutions that satisfy specified
criteria. These partial solutions are then used as building blocks in the next layer, to
progressively build up complete solutions.
The resulting multi-layered genetic algorithm is able to concentrate its search efforts in
areas where good quality solutions are likely to be present, therefore producing better
results than traditional genetic algorithms. Further developments of the multi-layered
genetic algorithm are also suggested in this thesis. The algorithm is combined with a
local search method, and heuristic rules are used for initialisation of the population. The
combined method results in an effective and fast exploration of the problem's search
space and is suitable for a variety of optimisation problems.
The proposed algorithm is implemented using MATLAB programming language and
tested on a real power generation system. A number of implementation issues, such as specific chromosome structure and a varying generation gap; interchangeable solutions
and gene convergence; weeding out duplicates from the population and reducing the
search space without losing the quality of representing the problem domain, are all
discussed. Specifics of a local search method and its representation are also examined.
Special attention is paid to developing efficient evaluation and neighbourhood
exploration procedures.
Capítulos de livros sobre o assunto "Combinatorial optimization layers"
1
Cai, Xuhong, Li Jiang, Songhu Guo, Hejiao Huang e Hongwei Du. "A Two-Layers Heuristic Search Algorithm for Milk Run with a New PDPTW Model". In Combinatorial Optimization and Applications, 379–92. Cham: Springer International Publishing, 2020. http://dx.doi.org/10.1007/978-3-030-64843-5_26.
Texto completo da fonte
2
Ruthmair, Mario, e Günther R. Raidl. "A Layered Graph Model and an Adaptive Layers Framework to Solve Delay-Constrained Minimum Tree Problems". In Integer Programming and Combinatoral Optimization, 376–88. Berlin, Heidelberg: Springer Berlin Heidelberg, 2011. http://dx.doi.org/10.1007/978-3-642-20807-2_30.
Texto completo da fonte
3
Awasthi, Abhishek, Jörg Lässig, Thomas Weise e Oliver Kramer. "Tackling Common Due Window Problem with a Two-Layered Approach". In Combinatorial Optimization and Applications, 772–81. Cham: Springer International Publishing, 2016. http://dx.doi.org/10.1007/978-3-319-48749-6_59.
Texto completo da fonte
4
Snapper, Marc L., e Amir H. Hoveyda. "Combinatorial approaches to chiral catalyst discovery". In Combinatorial Chemistry, 433–56. Oxford University PressOxford, 2000. http://dx.doi.org/10.1093/oso/9780199637546.003.0016.
Texto completo da fonteResumo:
Abstract In most combinatorial approaches to drug discovery, lead compounds are generally identified from large, structurally diverse libraries (1). Preliminary findings are then often optimized through the subsequent design and examination of more limited libraries that focus and expand on the initial results. In a similar manner, this layered approach to drug discovery and optimization can be adapted to catalyst development (Scheme 1, see also Chapters 14 and 15). In the first phase, a wide range of catalyst candidates can be screened to select specific complexes that effect a reaction of interest. Once a catalyst has been identified that demonstrates the desired reactivity, efforts can then tum toward optimizing the conditions and system to yield the desired selectivity. This two-tiered development strategy can offer distinct advantages in the discovery and identification of catalysts for asymmetric reactions. If the catalyst discovery and optimization protocols are sufficiently rapid and reliable, there is no prerequisite for finding general solutions to catalytic reactions; each reaction can enjoy a catalyst that is designed specifically for the substrate and transformation of interest.
5
SUZUKI, KYOTARO, HIDEHARU AMANO e YOSHIYASU TAKEFUJI. "MULTI-LAYER CHANNEL ROUTING PROBLEMS". In Neural Computing for Optimization and Combinatorics, 79–99. WORLD SCIENTIFIC, 1996. http://dx.doi.org/10.1142/9789812832122_0005.
Texto completo da fonte
6
James, Tabitha, e Cesar Rego. "Path Relinking with Multi-Start Tabu Search for the Quadratic Assignment Problem". In Recent Algorithms and Applications in Swarm Intelligence Research, 52–70. IGI Global, 2013. http://dx.doi.org/10.4018/978-1-4666-2479-5.ch004.
Texto completo da fonteResumo:
This paper introduces a new path relinking algorithm for the well-known quadratic assignment problem (QAP) in combinatorial optimization. The QAP has attracted considerable attention in research because of its complexity and its applicability to many domains. The algorithm presented in this study employs path relinking as a solution combination method incorporating a multistart tabu search algorithm as an improvement method. The resulting algorithm has interesting similarities and contrasts with particle swarm optimization methods. Computational testing indicates that this algorithm produces results that rival the best QAP algorithms. The authors additionally conduct an analysis disclosing how different strategies prove more or less effective depending on the landscapes of the problems to which they are applied. This analysis lays a foundation for developing more effective future QAP algorithms, both for methods based on path relinking and tabu search, and for hybrids of such methods with related processes found in particle swarm optimization.
Trabalhos de conferências sobre o assunto "Combinatorial optimization layers"
1
Kul'ment'ev, Аlexander. "Artificial intelligence optimization method for nuclear fuel triso-elements in high-temperature reactor". In IXth INTERNATIONAL SAMSONOV CONFERENCE “MATERIALS SCIENCE OF REFRACTORY COMPOUNDS”. Frantsevich Ukrainian Materials Research Society, 2024. http://dx.doi.org/10.62564/m4-ak2225.
Texto completo da fonteResumo:
In the present era nuclear energy, has unique advantages compared to other energy sources. Now significant research and development related to TRISO-coated fuels is underway worldwide as part of the activities of the Generation IV International Forum on Very-High-Temperature Reactors. The focus is largely on extending the capabilities of the TRISO-coated fuel system for higher operating temperatures (1250°C) and higher burnups (10 – 20 %). Of greatest concern is the influence of higher fuel temperatures and burnups on fission product interactions with the SiC layer leading to the release of fission products. One of the possible solution consist in addition additional layers with special properties. For example, to prevent the corrosion of the SiC layer by fission product palladium, several types of new combinations of the coating layers have been proposed and tested. The idea is to add a layer that traps palladium by chemical reaction inside the SiC layer. Earlier several kinds of additional layers have been selected: an SiC + PyC layer and an SiC layer. For optimization of TRISO particle it is necessary to determine the number of additional layers, their thickness and composition. This is combinatorial optimization problem (continuous + discrete). Traditional methods rely on manual adjustment and human experience, which is inefficient and difficult to obtain the optimal solution. Therefore it is necessary to develop an automated design method. In the present report variant of such method is proposed based on artificial intelligence approach. There are several meta-heuristic algorithms such as genetic algorithm, neural network and particle swarm optimization algorithm (PSO) which have the ability to solve continuous, discrete and combinatorial optimization problems. Namely PSO algorithm looks especially attractive. Early this method was proven to be reliable and effective in nuclear power problems by applying it in designing a Savannah marine reactor shielding.
2
Cao, Tianxiao, Lu Sun, Canh Hao Nguyen e Hiroshi Mamitsuka. "Learning Low-Rank Tensor Cores with Probabilistic ℓ0-Regularized Rank Selection for Model Compression". In Thirty-Third International Joint Conference on Artificial Intelligence {IJCAI-24}. California: International Joint Conferences on Artificial Intelligence Organization, 2024. http://dx.doi.org/10.24963/ijcai.2024/418.
Texto completo da fonteResumo:
Compressing deep neural networks is of great importance for real-world applications on resource-constrained devices. Tensor decomposition is one promising answer that retains the functionality and most of the expressive power of the original deep models by replacing the weights with their decomposed cores. Decomposition with optimal ranks can achieve a good compression-accuracy trade-off, but it is expensive to optimize due to its discrete and combinatorial nature. A common practice is to set all ranks equal and tune one hyperparameter, but it may significantly harm the flexibility and generalization. In this paper, we propose a novel automatic rank selection method for deep model compression that allows learning model weights and decomposition ranks simultaneously. We propose to penalize the ℓ0 (quasi-)norm of the slices of decomposed tensor cores during model training. To avoid combinatorial optimization, we develop a probabilistic formulation and apply an approximate Bernoulli gate to each of the slices of tensor cores, which can be implemented in an end-to-end and scalable framework via gradient descent. It enables the automatic rank selection to be incorporated with arbitrary tensor decompositions and neural network layers such as linear layers, convolutional layers, and embedding layers. Comprehensive experiments on various tasks, including image classification, text sentiment classification, and neural machine translation, demonstrate the superior effectiveness of the proposed method over baselines.
3
Bin Sazali, Muhammad Arif, Nahrul Khair Alang Md Rashid e Khaidzir Hamzah. "Ant Colony Optimization of Multilayer Shielding for Mixed Neutron and Gamma Radiations: A Preliminary Study". In 2017 25th International Conference on Nuclear Engineering. American Society of Mechanical Engineers, 2017. http://dx.doi.org/10.1115/icone25-67765.
Texto completo da fonteResumo:
Mixed neutron and gamma radiations require different shielding materials as their interaction with materials is different. Composites were developed in order to combine the shielding capabilities of different materials. However, their homogeneity is difficult to be assured which can lead to pinholes where radiation can penetrate. To avoid this problem, several materials arranged in layers can be used to shield against mixed radiations. Since the multilayer shielding can be made from any material in many configurations, the ant colony optimization (ACO) is a promising method because it deals with combinatorial optimization problems. The candidate materials are HDPE, boron, cadmium, gadolinium, tungsten, bismuth, and iron. Preliminary MCNP simulations were done to observe the effect of arrangements, thicknesses, and types of materials on the radiation spectrum. It was found that: (1) the final layer should be made of high density material, (2) an increase beyond certain thicknesses did not result in a significant increase in attenuation, and (3) there should be an optimum combination of material that can effectively shield against both neutrons and gamma rays.
4
Jurewicz, Mateusz, e Leon Derczynski. "Set Interdependence Transformer: Set-to-Sequence Neural Networks for Permutation Learning and Structure Prediction". In Thirty-First International Joint Conference on Artificial Intelligence {IJCAI-22}. California: International Joint Conferences on Artificial Intelligence Organization, 2022. http://dx.doi.org/10.24963/ijcai.2022/434.
Texto completo da fonteResumo:
The task of learning to map an input set onto a permuted sequence of its elements is challenging for neural networks. Set-to-sequence problems occur in natural language processing, computer vision and structure prediction, where interactions between elements of large sets define the optimal output. Models must exhibit relational reasoning, handle varying cardinalities and manage combinatorial complexity. Previous attention-based methods require n layers of their set transformations to explicitly represent n-th order relations. Our aim is to enhance their ability to efficiently model higher-order interactions through an additional interdependence component. We propose a novel neural set encoding method called the Set Interdependence Transformer, capable of relating the set's permutation invariant representation to its elements within sets of any cardinality. We combine it with a permutation learning module into a complete, 3-part set-to-sequence model and demonstrate its state-of-the-art performance on a number of tasks. These range from combinatorial optimization problems, through permutation learning challenges on both synthetic and established NLP datasets for sentence ordering, to a novel domain of product catalog structure prediction. Additionally, the network's ability to generalize to unseen sequence lengths is investigated and a comparative empirical analysis of the existing methods' ability to learn higher-order interactions is provided.
5
Monteiro, Daniel Pereira, Lucas Nardelli de Freitas Botelho Saar, Larissa Ferreira Rodrigues Moreira e Rodrigo Moreira. "On Enhancing Network Throughput using Reinforcement Learning in Sliced Testbeds". In Workshop de Pesquisa Experimental da Internet do Futuro, 1–7. Sociedade Brasileira de Computação - SBC, 2024. http://dx.doi.org/10.5753/wpeif.2024.2094.
Texto completo da fonteResumo:
Novel applications demand high throughput, low latency, and high reliability connectivity and still pose significant challenges to slicing orchestration architectures. The literature explores network slicing techniques that employ canonical methods, artificial intelligence, and combinatorial optimization to address errors and ensure throughput for network slice data plane. This paper introduces the Enhanced Mobile Broadband (eMBB)-Agent as a new approach that uses Reinforcement Learning (RL) in a vertical application to enhance network slicing throughput to fit Service-Level Agreements (SLAs). The eMBB-Agent analyzes application transmission variables and proposes actions within a discrete space to adjust the reception window using a Deep Q-Network (DQN). This paper also presents experimental results that examine the impact of factors such as the channel error rate, DQN model layers, and learning rate on model convergence and achieved throughput, providing insights on embedding intelligence in network slicing.
6
Potebnia, Artem. "Construction of the comprehensive multi-layer graph model of the search spaces associated with the combinatorial optimization problems". In 2017 4th International Scientific-Practical Conference Problems of Infocommunications. Science and Technology (PIC S&T). IEEE, 2017. http://dx.doi.org/10.1109/infocommst.2017.8246398.
Texto completo da fonte
7
Lin, Chen-Chou, Yi-Chih Chow e Yu-Yu Huang. "Geometry Optimization of Cylindrical Flaps of Oscillating Wave Surge Converters Using Artificial Neural Network Models". In ASME 2019 13th International Conference on Energy Sustainability collocated with the ASME 2019 Heat Transfer Summer Conference. American Society of Mechanical Engineers, 2019. http://dx.doi.org/10.1115/es2019-3878.
Texto completo da fonteResumo:
Abstract This paper presents an optimization algorithm based on the Artificial Neural Network (ANN) to determine the optimal shape, size, and density for the cylindrical flap of the Bottom-Hinged Oscillating Wave Surge Converter (BH-OWSC) that can extract maximal wave power under a given wave condition. Eight parameters are selected, and their upper and lower bounds are set at the initial stage, and then 64 cases with different combinatorial parametric settings are generated by the Design of Experiment process. The 64 cases are then fed into FLOW-3D to simulate the operations of the BH-OWSC under the given wave condition for calculating the capture factor, establishing a database for subsequent ANN data training purpose. To search the maximal capture factor in the specific range of the flap models, we fed 107 random models with various levels of design parameters into the ANN model, which adopts the backpropagation architecture and one hidden layer with ten neuron cells. After three complete random searches, and by simulating the ANN-derived flap’s geometry using FLOW-3D, the result shows that a maximal capture factor of 1.824 can be obtained. The major geometric features of the flap with maximal capture factor are (1) the cylinder axis of the flap inclines to the opposite direction of incident wave propagation, (2) the cylinder’s sectional diameters are about the same size, and (3) the smaller flap density the better power capturing performance.