Letteratura scientifica selezionata sul tema "Adaptive gradient methods with momentum"
Cita una fonte nei formati APA, MLA, Chicago, Harvard e in molti altri stili
Consulta la lista di attuali articoli, libri, tesi, atti di convegni e altre fonti scientifiche attinenti al tema "Adaptive gradient methods with momentum".
Accanto a ogni fonte nell'elenco di riferimenti c'è un pulsante "Aggiungi alla bibliografia". Premilo e genereremo automaticamente la citazione bibliografica dell'opera scelta nello stile citazionale di cui hai bisogno: APA, MLA, Harvard, Chicago, Vancouver ecc.
Puoi anche scaricare il testo completo della pubblicazione scientifica nel formato .pdf e leggere online l'abstract (il sommario) dell'opera se è presente nei metadati.
Articoli di riviste sul tema "Adaptive gradient methods with momentum":
1
Abdulkadirov, R. I., e P. A. Lyakhov. "A new approach to training neural networks using natural gradient descent with momentum based on Dirichlet distributions". Computer Optics 47, n. 1 (febbraio 2023): 160–69. http://dx.doi.org/10.18287/2412-6179-co-1147.
Abstract (sommario):
In this paper, we propose a natural gradient descent algorithm with momentum based on Dirichlet distributions to speed up the training of neural networks. This approach takes into account not only the direction of the gradients, but also the convexity of the minimized function, which significantly accelerates the process of searching for the extremes. Calculations of natural gradients based on Dirichlet distributions are presented, with the proposed approach introduced into an error backpropagation scheme. The results of image recognition and time series forecasting during the experiments show that the proposed approach gives higher accuracy and does not require a large number of iterations to minimize loss functions compared to the methods of stochastic gradient descent, adaptive moment estimation and adaptive parameter-wise diagonal quasi-Newton method for nonconvex stochastic optimization.
2
REHMAN, MUHAMMAD ZUBAIR, e NAZRI MOHD NAWI. "STUDYING THE EFFECT OF ADAPTIVE MOMENTUM IN IMPROVING THE ACCURACY OF GRADIENT DESCENT BACK PROPAGATION ALGORITHM ON CLASSIFICATION PROBLEMS". International Journal of Modern Physics: Conference Series 09 (gennaio 2012): 432–39. http://dx.doi.org/10.1142/s201019451200551x.
Abstract (sommario):
Despite being widely used in the practical problems around the world, Gradient Descent Back-propagation algorithm comes with problems like slow convergence and convergence to local minima. Previous researchers have suggested certain modifications to improve the convergence in gradient Descent Back-propagation algorithm such as careful selection of input weights and biases, learning rate, momentum, network topology, activation function and value for 'gain' in the activation function. This research proposed an algorithm for improving the working performance of back-propagation algorithm which is 'Gradient Descent with Adaptive Momentum (GDAM)' by keeping the gain value fixed during all network trials. The performance of GDAM is compared with 'Gradient Descent with fixed Momentum (GDM)' and 'Gradient Descent Method with Adaptive Gain (GDM-AG)'. The learning rate is fixed to 0.4 and maximum epochs are set to 3000 while sigmoid activation function is used for the experimentation. The results show that GDAM is a better approach than previous methods with an accuracy ratio of 1.0 for classification problems like Wine Quality, Mushroom and Thyroid disease.
3
Chen, Ruijuan, Xiaoquan Tang e Xiuting Li. "Adaptive Stochastic Gradient Descent Method for Convex and Non-Convex Optimization". Fractal and Fractional 6, n. 12 (29 novembre 2022): 709. http://dx.doi.org/10.3390/fractalfract6120709.
Abstract (sommario):
Stochastic gradient descent is the method of choice for solving large-scale optimization problems in machine learning. However, the question of how to effectively select the step-sizes in stochastic gradient descent methods is challenging, and can greatly influence the performance of stochastic gradient descent algorithms. In this paper, we propose a class of faster adaptive gradient descent methods, named AdaSGD, for solving both the convex and non-convex optimization problems. The novelty of this method is that it uses a new adaptive step size that depends on the expectation of the past stochastic gradient and its second moment, which makes it efficient and scalable for big data and high parameter dimensions. We show theoretically that the proposed AdaSGD algorithm has a convergence rate of O(1/T) in both convex and non-convex settings, where T is the maximum number of iterations. In addition, we extend the proposed AdaSGD to the case of momentum and obtain the same convergence rate for AdaSGD with momentum. To illustrate our theoretical results, several numerical experiments for solving problems arising in machine learning are made to verify the promise of the proposed method.
4
Zhang, Yue, Seong-Yoon Shin, Xujie Tan e Bin Xiong. "A Self-Adaptive Approximated-Gradient-Simulation Method for Black-Box Adversarial Sample Generation". Applied Sciences 13, n. 3 (18 gennaio 2023): 1298. http://dx.doi.org/10.3390/app13031298.
Abstract (sommario):
Deep neural networks (DNNs) have famously been applied in various ordinary duties. However, DNNs are sensitive to adversarial attacks which, by adding imperceptible perturbation samples to an original image, can easily alter the output. In state-of-the-art white-box attack methods, perturbation samples can successfully fool DNNs through the network gradient. In addition, they generate perturbation samples by only considering the sign information of the gradient and by dropping the magnitude. Accordingly, gradients of different magnitudes may adopt the same sign to construct perturbation samples, resulting in inefficiency. Unfortunately, it is often impractical to acquire the gradient in real-world scenarios. Consequently, we propose a self-adaptive approximated-gradient-simulation method for black-box adversarial attacks (SAGM) to generate efficient perturbation samples. Our proposed method uses knowledge-based differential evolution to simulate gradients and the self-adaptive momentum gradient to generate adversarial samples. To estimate the efficiency of the proposed SAGM, a series of experiments were carried out on two datasets, namely MNIST and CIFAR-10. Compared to state-of-the-art attack techniques, our proposed method can quickly and efficiently search for perturbation samples to misclassify the original samples. The results reveal that the SAGM is an effective and efficient technique for generating perturbation samples.
5
Long, Sheng, Wei Tao, Shuohao LI, Jun Lei e Jun Zhang. "On the Convergence of an Adaptive Momentum Method for Adversarial Attacks". Proceedings of the AAAI Conference on Artificial Intelligence 38, n. 13 (24 marzo 2024): 14132–40. http://dx.doi.org/10.1609/aaai.v38i13.29323.
Abstract (sommario):
Adversarial examples are commonly created by solving a constrained optimization problem, typically using sign-based methods like Fast Gradient Sign Method (FGSM). These attacks can benefit from momentum with a constant parameter, such as Momentum Iterative FGSM (MI-FGSM), to enhance black-box transferability. However, the monotonic time-varying momentum parameter is required to guarantee convergence in theory, creating a theory-practice gap. Additionally, recent work shows that sign-based methods fail to converge to the optimum in several convex settings, exacerbating the issue. To address these concerns, we propose a novel method which incorporates both an innovative adaptive momentum parameter without monotonicity assumptions and an adaptive step-size scheme that replaces the sign operation. Furthermore, we derive a regret upper bound for general convex functions. Experiments on multiple models demonstrate the efficacy of our method in generating adversarial examples with human-imperceptible noise while achieving high attack success rates, indicating its superiority over previous adversarial example generation methods.
6
Zhang, Jiahui, Xinhao Yang, Ke Zhang e Chenrui Wen. "An Adaptive Deep Learning Optimization Method Based on Radius of Curvature". Computational Intelligence and Neuroscience 2021 (10 novembre 2021): 1–10. http://dx.doi.org/10.1155/2021/9882068.
Abstract (sommario):
An adaptive clamping method (SGD-MS) based on the radius of curvature is designed to alleviate the local optimal oscillation problem in deep neural network, which combines the radius of curvature of the objective function and the gradient descent of the optimizer. The radius of curvature is considered as the threshold to separate the momentum term or the future gradient moving average term adaptively. In addition, on this basis, we propose an accelerated version (SGD-MA), which further improves the convergence speed by using the method of aggregated momentum. Experimental results on several datasets show that the proposed methods effectively alleviate the local optimal oscillation problem and greatly improve the convergence speed and accuracy. A novel parameter updating algorithm is also provided in this paper for deep neural network.
7
Zang, Yu, Zhe Xue, Shilong Ou, Lingyang Chu, Junping Du e Yunfei Long. "Efficient Asynchronous Federated Learning with Prospective Momentum Aggregation and Fine-Grained Correction". Proceedings of the AAAI Conference on Artificial Intelligence 38, n. 15 (24 marzo 2024): 16642–50. http://dx.doi.org/10.1609/aaai.v38i15.29603.
Abstract (sommario):
Asynchronous federated learning (AFL) is a distributed machine learning technique that allows multiple devices to collaboratively train deep learning models without sharing local data. However, AFL suffers from low efficiency due to poor client model training quality and slow server model convergence speed, which are a result of the heterogeneous nature of both data and devices. To address these issues, we propose Efficient Asynchronous Federated Learning with Prospective Momentum Aggregation and Fine-Grained Correction (FedAC). Our framework consists of three key components. The first component is client weight evaluation based on temporal gradient, which evaluates the client weight based on the similarity between the client and server update directions. The second component is adaptive server update with prospective weighted momentum, which uses an asynchronous buffered update strategy and a prospective weighted momentum with adaptive learning rate to update the global model in server. The last component is client update with fine-grained gradient correction, which introduces a fine-grained gradient correction term to mitigate the client drift and correct the client stochastic gradient. We conduct experiments on real and synthetic datasets, and compare with existing federated learning methods. Experimental results demonstrate effective improvements in model training efficiency and AFL performance by our framework.
8
Liu, Miaomiao, Dan Yao, Zhigang Liu, Jingfeng Guo e Jing Chen. "An Improved Adam Optimization Algorithm Combining Adaptive Coefficients and Composite Gradients Based on Randomized Block Coordinate Descent". Computational Intelligence and Neuroscience 2023 (10 gennaio 2023): 1–14. http://dx.doi.org/10.1155/2023/4765891.
Abstract (sommario):
An improved Adam optimization algorithm combining adaptive coefficients and composite gradients based on randomized block coordinate descent is proposed to address issues of the Adam algorithm such as slow convergence, the tendency to miss the global optimal solution, and the ineffectiveness of processing high-dimensional vectors. The adaptive coefficient is used to adjust the gradient deviation value and correct the search direction firstly. Then, the predicted gradient is introduced, and the current gradient and the first-order momentum are combined to form a composite gradient to improve the global optimization ability. Finally, the random block coordinate method is used to determine the gradient update mode, which reduces the computational overhead. Simulation experiments on two standard datasets for classification show that the convergence speed and accuracy of the proposed algorithm are higher than those of the six gradient descent methods, and the CPU and memory utilization are significantly reduced. In addition, based on logging data, the BP neural networks optimized by six algorithms, respectively, are used to predict reservoir porosity. Results show that the proposed method has lower system overhead, higher accuracy, and stronger stability, and the absolute error of more than 86% data is within 0.1%, which further verifies its effectiveness.
9
Jiang, Shuoran, Qingcai Chen, Youcheng Pan, Yang Xiang, Yukang Lin, Xiangping Wu, Chuanyi Liu e Xiaobao Song. "ZO-AdaMU Optimizer: Adapting Perturbation by the Momentum and Uncertainty in Zeroth-Order Optimization". Proceedings of the AAAI Conference on Artificial Intelligence 38, n. 16 (24 marzo 2024): 18363–71. http://dx.doi.org/10.1609/aaai.v38i16.29796.
Abstract (sommario):
Lowering the memory requirement in full-parameter training on large models has become a hot research area. MeZO fine-tunes the large language models (LLMs) by just forward passes in a zeroth-order SGD optimizer (ZO-SGD), demonstrating excellent performance with the same GPU memory usage as inference. However, the simulated perturbation stochastic approximation for gradient estimate in MeZO leads to severe oscillations and incurs a substantial time overhead. Moreover, without momentum regularization, MeZO shows severe over-fitting problems. Lastly, the perturbation-irrelevant momentum on ZO-SGD does not improve the convergence rate. This study proposes ZO-AdaMU to resolve the above problems by adapting the simulated perturbation with momentum in its stochastic approximation. Unlike existing adaptive momentum methods, we relocate momentum on simulated perturbation in stochastic gradient approximation. Our convergence analysis and experiments prove this is a better way to improve convergence stability and rate in ZO-SGD. Extensive experiments demonstrate that ZO-AdaMU yields better generalization for LLMs fine-tuning across various NLP tasks than MeZO and its momentum variants.
10
Sineglazov, Victor, e Anatoly Kot. "Design of hybrid neural networks of the ensemble structure". Eastern-European Journal of Enterprise Technologies 1, n. 4 (109) (26 febbraio 2021): 31–45. http://dx.doi.org/10.15587/1729-4061.2021.225301.
Abstract (sommario):
This paper considers the structural-parametric synthesis (SPS) of neural networks (NNs) of deep learning, in particular convolutional neural networks (CNNs), which are used in image processing. It has been shown that modern neural networks may possess a variety of topologies. That is ensured by using unique blocks that determine their essential features, namely, the compression and excitation unit, the attention module convolution unit, the channel attention module, the spatial attention module, the residual unit, the ResNeXt block. This, first of all, is due to the need to increase their efficiency in the processing of images. Due to the large architectural space of parameters, including the type of unique block, the location in the structure of the convolutional neural network, its connections with other blocks, layers, computing costs grow nonlinearly. To minimize computational costs while maintaining the specified accuracy this work set tasks of both the generation of possible topology and structural-parametric synthesis of convolutional neural networks. To resolve them, the use of a genetic algorithm (GA) has been proposed. Parameter configuration was implemented using a genetic algorithm and modern gradient methods (GM). For example, stochastic gradient descent with momentum, accelerated Nesterov gradient, adaptive gradient algorithm, distribution of the root of the mean square of the gradient, assessment of adaptive momentum, adaptive Nesterov momentum. It is assumed to use such networks in the intelligent medical diagnostic system (IMDS), for determining the activity of tuberculosis. To improve the accuracy of solving the classification problem in the processing of images, the ensemble structure of hybrid convolutional neural networks (HCNNs) has been proposed in the current work. The parallel structure of the ensemble with the merged layer was used. Algorithms of optimal choice and integration of features in the construction of the ensemble have been developed
Tesi sul tema "Adaptive gradient methods with momentum":
1
Barakat, Anas. "Contributions to non-convex stochastic optimization and reinforcement learning". Electronic Thesis or Diss., Institut polytechnique de Paris, 2021. http://www.theses.fr/2021IPPAT030.
Abstract (sommario):
Cette thèse est centrée autour de l'analyse de convergence de certains algorithmes d'approximation stochastiques utilisés en machine learning appliqués à l'optimisation et à l'apprentissage par renforcement. La première partie de la thèse est dédiée à un célèbre algorithme en apprentissage profond appelé ADAM, utilisé pour entraîner des réseaux de neurones. Cette célèbre variante de la descente de gradient stochastique est plus généralement utilisée pour la recherche d'un minimiseur local d'une fonction. En supposant que la fonction objective est différentiable et non convexe, nous établissons la convergence des itérées au temps long vers l'ensemble des points critiques sous une hypothèse de stabilité dans le régime des pas constants. Ensuite, nous introduisons une nouvelle variante de l'algorithme ADAM à pas décroissants. Nous montrons alors sous certaines hypothèses réalistes que les itérées sont presque sûrement bornées et convergent presque sûrement vers des points critiques de la fonction objective. Enfin, nous analysons les fluctuations de l'algorithme par le truchement d'un théorème central limite conditionnel. Dans la deuxième partie de cette thèse, dans le régime des pas décroissants, nous généralisons nos résultats de convergence et de fluctuations à une procédure d'optimisation stochastique unifiant plusieurs variantes de descente de gradient stochastique comme la méthode de la boule pesante, l'algorithme stochastique de Nesterov accéléré ou encore le célèbre algorithme ADAM, parmi d'autres. Nous concluons cette partie par un résultat d'évitement de pièges qui établit la non convergence de l'algorithme général vers des points critiques indésirables comme les maxima locaux ou les points-selles. Ici, le principal ingrédient est un nouveau résultat indépendant d'évitement de pièges pour un contexte non-autonome. Enfin, la dernière partie de cette thèse qui est indépendante des deux premières parties est dédiée à l'analyse d'un algorithme d'approximation stochastique pour l'apprentissage par renforcement. Dans cette dernière partie, dans le cadre des processus décisionnels de Markov avec critère de récompense gamma-pondéré, nous proposons une analyse d'un algorithme acteur-critique en ligne intégrant un réseau cible et avec approximation de fonction linéraire. Notre algorithme utilise trois échelles de temps distinctes: une échelle pour l'acteur et deux autres pour la critique. Au lieu d'utiliser l'algorithme de différence temporelle (TD) standard à une échelle de temps, nous utilisons une version de l'algorithme TD à deux échelles de temps intégrant un réseau cible inspiré des algorithmes acteur-critique utilisés en pratique. Tout d'abord, nous établissons des résultats de convergence pour la critique et l'acteur sous échantillonnage Markovien. Ensuite, nous menons une analyse à temps fini montrant l'impact de l'utilisation d'un réseau cible sur les méthodes acteur-critiqueThis thesis is focused on the convergence analysis of some popular stochastic approximation methods in use in the machine learning community with applications to optimization and reinforcement learning.The first part of the thesis is devoted to a popular algorithm in deep learning called ADAM used for training neural networks. This variant of stochastic gradient descent is more generally useful for finding a local minimizer of a function. Assuming that the objective function is differentiable and non-convex, we establish the convergence of the iterates in the long run to the set of critical points under a stability condition in the constant stepsize regime. Then, we introduce a novel decreasing stepsize version of ADAM. Under mild assumptions, it is shown that the iterates are almost surely bounded and converge almost surely to critical points of the objective function. Finally, we analyze the fluctuations of the algorithm by means of a conditional central limit theorem.In the second part of the thesis, in the vanishing stepsizes regime, we generalize our convergence and fluctuations results to a stochastic optimization procedure unifying several variants of the stochastic gradient descent such as, among others, the stochastic heavy ball method, the Stochastic Nesterov Accelerated Gradient algorithm, and the widely used ADAM algorithm. We conclude this second part by an avoidance of traps result establishing the non-convergence of the general algorithm to undesired critical points, such as local maxima or saddle points. Here, the main ingredient is a new avoidance of traps result for non-autonomous settings, which is of independent interest.Finally, the last part of this thesis which is independent from the two previous parts, is concerned with the analysis of a stochastic approximation algorithm for reinforcement learning. In this last part, we propose an analysis of an online target-based actor-critic algorithm with linear function approximation in the discounted reward setting. Our algorithm uses three different timescales: one for the actor and two for the critic. Instead of using the standard single timescale temporal difference (TD) learning algorithm as a critic, we use a two timescales target-based version of TD learning closely inspired from practical actor-critic algorithms implementing target networks. First, we establish asymptotic convergence results for both the critic and the actor under Markovian sampling. Then, we provide a finite-time analysis showing the impact of incorporating a target network into actor-critic methods
2
Vaudrey, Michael Allen. "Adaptive Control Methods for Non-Linear Self-Excited Systems". Diss., Virginia Tech, 2001. http://hdl.handle.net/10919/28884.
Abstract (sommario):
Self-excited systems are open loop unstable plants having a nonlinearity that prevents an exponentially increasing time response. The resulting limit cycle is induced by any slight disturbance that causes the response of the system to grow to the saturation level of the nonlinearity. Because there is no external disturbance, control of these self-excited systems requires that the open loop system dynamics are altered so that any unstable open loop poles are stabilized in the closed loop.
This work examines a variety of adaptive control approaches for controlling a thermoacoustic instability, a physical self-excited system. Initially, a static feedback controller loopshaping design and associated system identification method is presented. This design approach is shown to effectively stabilize an unstable Rijke tube combustor while preventing the creation of additional controller induced instabilities. The loopshaping design method is then used in conjunction with a trained artificial neural network to demonstrate stabilizing control in the presence of changing plant dynamics over a wide variety of operating conditions. However, because the ANN is designed specifically for a single combustor/actuator arrangement, its limited portability is a distinct disadvantage.
Filtered-X least mean squares (LMS) adaptive feedback control approaches are examined when applied to both stable and unstable plants. An identification method for approximating the relevant plant dynamics to be modeled is proposed and shown to effectively stabilize the self-excited system in simulations and experiments. The adaptive feedback controller is further analyzed for robust performance when applied to the stable, disturbance rejection control problem. It is shown that robust stability cannot be guaranteed because arbitrarily small errors in the plant model can generate gradient divergence and unstable feedback loops.
Finally, a time-averaged-gradient (TAG) algorithm is investigated for use in controlling self-excited systems such as the thermoacoustic instability. The TAG algorithm is shown to be very effective in stabilizing the unstable dynamics using a variety of controller parameterizations, without the need for plant estimation information from the system to be controlled.Ph. D.
3
O'Neal, Jerome W. "The use of preconditioned iterative linear solvers in interior-point methods and related topics". Diss., Available online, Georgia Institute of Technology, 2005, 2005. http://etd.gatech.edu/theses/available/etd-06242005-162854/.
Abstract (sommario):
Thesis (Ph. D.)--Industrial and Systems Engineering, Georgia Institute of Technology, 2006.Parker, R. Gary, Committee Member ; Shapiro, Alexander, Committee Member ; Nemirovski, Arkadi, Committee Member ; Green, William, Committee Member ; Monteiro, Renato, Committee Chair.
4
Santos, Rodríguez Cristian de. "Backanalysis methodology based on multiple optimization techniques for geotechnical problems". Doctoral thesis, Universitat Politècnica de Catalunya, 2015. http://hdl.handle.net/10803/334179.
Abstract (sommario):
Nowadays, thanks to the increase of computers capability to solve huge and complex problems, and also thanks to the endless effort of the geotechnical community to define better and more sophisticated constitutive models, the challenge to predict and simulate soil behavior has been eased. However, due to the increase in that sophistication, the number of parameters that define the problem has also increased. Moreover, frequently, some of those parameters do not have a real geotechnical meaning as they just come from mathematical expressions, which makes them difficult to identify. As a consequence, more effort has to be placed on parameters identification in order to fully define the problem.
This thesis aims to provide a methodology to facilitate the identification of parameters of soil constitutive models by backanalysis. The best parameters are defines as those that minimize an objective function based on the differences between measurements and computed values. Different optimization techniques have been used in this study, from the most traditional ones, such as the gradient based methods, to the newest ones, such as adaptive genetic algorithms and hybrid methods. From this study, several recommendations have been put forward in order to take the most advantage of each type of optimization technique. Along with that, an extensive analysis has been carried out to determine the influence on soil parameters identification of what to measure, where to measure and when to measure in the context of tunneling. The Finite Element code Plaxis has been used as a tool for the direct analysis. A FORTRAN code has been developed to automate the entire backanalysis procedure. The Hardening Soil Model (HSM) has been adopted to simulate the soil behavior. Several soil parameters of the HSM implemented in Plaxis, such as E_50^ref, E_ur^ref, c and f, have been identified for different geotechnical scenarios. First, a synthetic tunnel case study has been used to analyze all the different approaches that have been proposed in this thesis. Then, two complex real cases of a tunnel construction (Barcelona Metro Line 9) and a large excavation (Girona High-Speed Railway Station) have been presented to illustrate the potential of the methodology. Special focus on the influence of construction procedures and instruments error structure has been placed for the tunnel backanalysis, whereas in the station backanalysis, more effort has been devoted to the potential of the concept of adaptive design by backanalysis. Moreover, another real case, involving a less conventional geotechnical problem, such as Mars surface exploratory rovers, has been also presented to test the backanalysis methodology and the reliability of the Wong & Reece wheel-terrain model; widely adopted by the terramechanics community, but nonetheless, still not fully accepted when analyzing lightweight rovers as the ones that have been used in recent Mars exploratory missions.Actualmente, gracias al aumento de la capacidad de los ordenadores para resolver problemas grandes y complejos, y gracias también al gran esfuerzo de la comunidad geotécnica de definir mejores y más sofisticados modelos constitutivos, se ha abordado el reto de predecir y simular el comportamiento del terreno. Sin embargo, debido al aumento de esa sofisticación, también ha aumentado el número de parámetros que definen el problema. Además, frecuentemente, muchos de esos parámetros no tienen un sentido geotécnico real dado que vienen directamente de expresiones puramente matemáticas, lo cual dificulta su identificación. Como consecuencia, es necesario un mayor esfuerzo en la identificación de los parámetros para poder definir apropiadamente el problema. Esta tesis pretende proporcionar una metodología que facilite la identificación mediante el análisis inverso de los parámetros de modelos constitutivos del terreno. Los mejores parámetros se definen como aquellos que minimizan una función objetivo basada en la diferencia entre medidas y valores calculados. Diferentes técnicas de optimización han sido utilizadas en este estudio, desde las más tradicionales, como los métodos basados en el gradiente, hasta las más modernas, como los algoritmos genéticos adaptativos y los métodos híbridos. De este estudio, se han extraído varias recomendaciones para sacar el mayor provecho de cada una de las técnicas de optimización. Además, se ha llevado a cabo un análisis extensivo para determinar la influencia sobre qué medir, dónde medir y cuándo medir en el contexto de la excavación de un túnel. El código de Elementos Finitos Plaxis ha sido utilizado como herramienta de cálculo del problema directo. El desarrollo de un código FORTRAN ha sido necesario para automatizar todo el procedimiento de Análisis Inverso. El modelo constitutivo de Hardening Soil ha sido adoptado para simular el comportamiento del terreno. Varios parámetros del modelo constitutivo de Hardening implementado en Plaxis, como E_50^ref, E_ur^ref, c y f, han sido identificados para diferentes escenarios geotécnicos. Primero, se ha utilizado un caso sintético de un túnel donde se han analizado todas las distintas técnicas que han sido propuestas en esta tesis. Después, dos casos reales complejos de una construcción de un túnel (Línea 9 del Metro de Barcelona) y una gran excavación (Estación de Girona del Tren de Alta Velocidad) se han presentado para ilustrar el potencial de la metodología. Un enfoque especial en la influencia del procedimiento constructivo y la estructura del error de las medidas se le ha dado al análisis inverso del túnel, mientras que en el análisis inverso de la estación el esfuerzo se ha centrado más en el concepto del diseño adaptativo mediante el análisis inverso. Además, otro caso real, algo menos convencional en términos geotécnicos, como es la exploración de la superficie de Marte mediante robots, ha sido presentado para examinar la metodología y la fiabilidad del modelo de interacción suelo-rueda de Wong y Reece; extensamente adoptado por la comunidad que trabajo en Terramecánica, pero aún no totalmente aceptada para robots ligeros como los que se han utilizado recientemente en las misiones de exploración de Marte.
5
Vu, Do Huy Cuong. "Méthodes numériques pour les écoulements et le transport en milieu poreux". Thesis, Paris 11, 2014. http://www.theses.fr/2014PA112348/document.
Abstract (sommario):
Cette thèse porte sur la modélisation de l’écoulement et du transport en milieu poreux ;nous effectuons des simulations numériques et démontrons des résultats de convergence d’algorithmes.Au Chapitre 1, nous appliquons des méthodes de volumes finis pour la simulation d’écoulements à densité variable en milieu poreux ; il vient à résoudre une équation de convection diffusion parabolique pour la concentration couplée à une équation elliptique en pression.Nous nous appuyons sur la méthode des volumes finis standard pour le calcul des solutions de deux problèmes spécifiques : une interface en rotation entre eau salée et eau douce et le problème de Henry. Nous appliquons ensuite la méthode de volumes finis généralisés SUSHI pour la simulation des mêmes problèmes ainsi que celle d’un problème de bassin salé en dimension trois d’espace. Nous nous appuyons sur des maillages adaptatifs, basés sur des éléments de volume carrés ou cubiques.Au Chapitre 2, nous nous appuyons de nouveau sur la méthode de volumes finis généralisés SUSHI pour la discrétisation de l’équation de Richards, une équation elliptique parabolique pour le calcul d’écoulements en milieu poreux. Le terme de diffusion peut être anisotrope et hétérogène. Cette classe de méthodes localement conservatrices s’applique àune grande variété de mailles polyédriques non structurées qui peuvent ne pas se raccorder.La discrétisation en temps est totalement implicite. Nous obtenons un résultat de convergence basé sur des estimations a priori et sur l’application du théorème de compacité de Fréchet-Kolmogorov. Nous présentons aussi des tests numériques.Au Chapitre 3, nous discrétisons le problème de Signorini par un schéma de type gradient,qui s’écrit à l’aide d’une formulation variationnelle discrète et est basé sur des approximations indépendantes des fonctions et des gradients. On montre l’existence et l’unicité de la solution discrète ainsi que sa convergence vers la solution faible du problème continu. Nous présentons ensuite un schéma numérique basé sur la méthode SUSHI.Au Chapitre 4, nous appliquons un schéma semi-implicite en temps combiné avec la méthode SUSHI pour la résolution numérique d’un problème d’écoulements à densité variable ;il s’agit de résoudre des équations paraboliques de convection-diffusion pour la densité de soluté et le transport de la température ainsi que pour la pression. Nous simulons l’avance d’un front d’eau douce assez chaude et le transport de chaleur dans un aquifère captif qui est initialement chargé d’eau froide salée. Nous utilisons des maillages adaptatifs, basés sur des éléments de volume carrésThis thesis bears on the modelling of groundwater flow and transport in porous media; we perform numerical simulations by means of finite volume methods and prove convergence results. In Chapter 1, we first apply a semi-implicit standard finite volume method and then the generalized finite volume method SUSHI for the numerical simulation of density driven flows in porous media; we solve a nonlinear convection-diffusion parabolic equation for the concentration coupled with an elliptic equation for the pressure. We apply the standard finite volume method to compute the solutions of a problem involving a rotating interface between salt and fresh water and of Henry's problem. We then apply the SUSHI scheme to the same problems as well as to a three dimensional saltpool problem. We use adaptive meshes, based upon square volume elements in space dimension two and cubic volume elements in space dimension three. In Chapter 2, we apply the generalized finite volume method SUSHI to the discretization of Richards equation, an elliptic-parabolic equation modeling groundwater flow, where the diffusion term can be anisotropic and heterogeneous. This class of locally conservative methods can be applied to a wide range of unstructured possibly non-matching polyhedral meshes in arbitrary space dimension. As is needed for Richards equation, the time discretization is fully implicit. We obtain a convergence result based upon a priori estimates and the application of the Fréchet-Kolmogorov compactness theorem. We implement the scheme and present numerical tests. In Chapter 3, we study a gradient scheme for the Signorini problem. Gradient schemes are nonconforming methods written in discrete variational formulation which are based on independent approximations of the functions and the gradients. We prove the existence and uniqueness of the discrete solution as well as its convergence to the weak solution of the Signorini problem. Finally we introduce a numerical scheme based upon the SUSHI discretization and present numerical results. In Chapter 4, we apply a semi-implicit scheme in time together with a generalized finite volume method for the numerical solution of density driven flows in porous media; it comes to solve nonlinear convection-diffusion parabolic equations for the solute and temperature transport as well as for the pressure. We compute the solutions for a specific problem which describes the advance of a warm fresh water front coupled to heat transfer in a confined aquifer which is initially charged with cold salt water. We use adaptive meshes, based upon square volume elements in space dimension two
6
"Adaptive Curvature for Stochastic Optimization". Master's thesis, 2019. http://hdl.handle.net/2286/R.I.53675.
Abstract (sommario):
abstract: This thesis presents a family of adaptive curvature methods for gradient-based stochastic optimization. In particular, a general algorithmic framework is introduced along with a practical implementation that yields an efficient, adaptive curvature gradient descent algorithm. To this end, a theoretical and practical link between curvature matrix estimation and shrinkage methods for covariance matrices is established. The use of shrinkage improves estimation accuracy of the curvature matrix when data samples are scarce. This thesis also introduce several insights that result in data- and computation-efficient update equations. Empirical results suggest that the proposed method compares favorably with existing second-order techniques based on the Fisher or Gauss-Newton and with adaptive stochastic gradient descent methods on both supervised and reinforcement learning tasks.Dissertation/Thesis
Masters Thesis Computer Science 2019
Libri sul tema "Adaptive gradient methods with momentum":
1
King, John. Application of the conjugate gradient method to the COMMIX-1B three dimensional momentum equation. 1987.
Capitoli di libri sul tema "Adaptive gradient methods with momentum":
1
Alexander, S. Thomas. "Gradient Adaptive Lattice Methods". In Adaptive Signal Processing, 99–110. New York, NY: Springer New York, 1986. http://dx.doi.org/10.1007/978-1-4612-4978-8_7.
2
Tsoi, Ah Chung, e Ah Chung Tsoi. "Gradient based learning methods". In Adaptive Processing of Sequences and Data Structures, 27–62. Berlin, Heidelberg: Springer Berlin Heidelberg, 1998. http://dx.doi.org/10.1007/bfb0053994.
3
Zhai, Tingting, Hao Wang, Frédéric Koriche e Yang Gao. "Online Feature Selection by Adaptive Sub-gradient Methods". In Machine Learning and Knowledge Discovery in Databases, 430–46. Cham: Springer International Publishing, 2019. http://dx.doi.org/10.1007/978-3-030-10928-8_26.
4
Konnov, Igor V. "Adaptive Step-Size for Non-monotone Gradient Type Methods". In Encyclopedia of Optimization, 1–4. Cham: Springer International Publishing, 2023. http://dx.doi.org/10.1007/978-3-030-54621-2_792-1.
5
Zhang, Xin, Can Cui, Wen-Jian Cai, Hui Cai e Gang Jing. "A Gradient-Based Online Adaptive Air Balancing Method". In Principle, Design and Optimization of Air Balancing Methods for the Multi-zone Ventilation Systems in Low Carbon Green Buildings, 35–53. Singapore: Springer Nature Singapore, 2022. http://dx.doi.org/10.1007/978-981-19-7091-7_3.
6
Rehman, M. Z., e N. M. Nawi. "The Effect of Adaptive Momentum in Improving the Accuracy of Gradient Descent Back Propagation Algorithm on Classification Problems". In Software Engineering and Computer Systems, 380–90. Berlin, Heidelberg: Springer Berlin Heidelberg, 2011. http://dx.doi.org/10.1007/978-3-642-22170-5_33.
7
Abbasi, Hamidreza, Ali Akbar Safavi e Maryam Salimifard. "Type-2 Fuzzy Wavelet Neural Network Controller Design Based on an Adaptive Gradient Descent Method for Nonlinear Dynamic Systems". In Applied Methods and Techniques for Mechatronic Systems, 229–47. Berlin, Heidelberg: Springer Berlin Heidelberg, 2013. http://dx.doi.org/10.1007/978-3-642-36385-6_12.
8
Hong, Dian, e Deng Chen. "Gradient-Based Adversarial Example Generation with Root Mean Square Propagation". In Artificial Intelligence and Human-Computer Interaction. IOS Press, 2024. http://dx.doi.org/10.3233/faia240141.
Abstract (sommario):
Nowadays, the security of neural networks has attracted more and more attention. Adversarial examples are one of the problems that affect the security of neural networks. The gradient-based attack method is a typical attack method, and the Momentum Iterative Fast Gradient Sign Method (MI-FGSM) is a typical attack algorithm among the gradient-based attack algorithms. However, this method may suffer from the problems of excessive gradient growth and low efficiency. In this paper, we propose a gradient-based attack algorithm RMS-FGSM based on Root Mean Square Propagation (RMSProp). RMS-FGSM algorithm avoids excessive gradient growth by Exponential Weighted Moving Average method and adaptive learning rate when gradient updates. Experiments on MNIST and CIFAR-100 and several models show that the attack success rate of our approach is higher than the baseline methods. Above all, our generated adversarial examples have a smaller perturbation under the same attack success rate.
9
"Gradient Methods Using Momentum". In Optimization for Data Analysis, 55–74. Cambridge University Press, 2022. http://dx.doi.org/10.1017/9781009004282.005.
10
"Adaptive Gradient Methods". In Inference and Learning from Data, 599–641. Cambridge University Press, 2022. http://dx.doi.org/10.1017/9781009218146.018.
Atti di convegni sul tema "Adaptive gradient methods with momentum":
1
Chen, Jinghui, Dongruo Zhou, Yiqi Tang, Ziyan Yang, Yuan Cao e Quanquan Gu. "Closing the Generalization Gap of Adaptive Gradient Methods in Training Deep Neural Networks". In Twenty-Ninth International Joint Conference on Artificial Intelligence and Seventeenth Pacific Rim International Conference on Artificial Intelligence {IJCAI-PRICAI-20}. California: International Joint Conferences on Artificial Intelligence Organization, 2020. http://dx.doi.org/10.24963/ijcai.2020/452.
Abstract (sommario):
Adaptive gradient methods, which adopt historical gradient information to automatically adjust the learning rate, despite the nice property of fast convergence, have been observed to generalize worse than stochastic gradient descent (SGD) with momentum in training deep neural networks. This leaves how to close the generalization gap of adaptive gradient methods an open problem. In this work, we show that adaptive gradient methods such as Adam, Amsgrad, are sometimes "over adapted". We design a new algorithm, called Partially adaptive momentum estimation method, which unifies the Adam/Amsgrad with SGD by introducing a partial adaptive parameter $p$, to achieve the best from both worlds. We also prove the convergence rate of our proposed algorithm to a stationary point in the stochastic nonconvex optimization setting. Experiments on standard benchmarks show that our proposed algorithm can maintain fast convergence rate as Adam/Amsgrad while generalizing as well as SGD in training deep neural networks. These results would suggest practitioners pick up adaptive gradient methods once again for faster training of deep neural networks.
2
Zhou, Beitong, Jun Liu, Weigao Sun, Ruijuan Chen, Claire Tomlin e Ye Yuan. "pbSGD: Powered Stochastic Gradient Descent Methods for Accelerated Non-Convex Optimization". In Twenty-Ninth International Joint Conference on Artificial Intelligence and Seventeenth Pacific Rim International Conference on Artificial Intelligence {IJCAI-PRICAI-20}. California: International Joint Conferences on Artificial Intelligence Organization, 2020. http://dx.doi.org/10.24963/ijcai.2020/451.
Abstract (sommario):
We propose a novel technique for improving the stochastic gradient descent (SGD) method to train deep networks, which we term pbSGD. The proposed pbSGD method simply raises the stochastic gradient to a certain power elementwise during iterations and introduces only one additional parameter, namely, the power exponent (when it equals to 1, pbSGD reduces to SGD). We further propose pbSGD with momentum, which we term pbSGDM. The main results of this paper present comprehensive experiments on popular deep learning models and benchmark datasets. Empirical results show that the proposed pbSGD and pbSGDM obtain faster initial training speed than adaptive gradient methods, comparable generalization ability with SGD, and improved robustness to hyper-parameter selection and vanishing gradients. pbSGD is essentially a gradient modifier via a nonlinear transformation. As such, it is orthogonal and complementary to other techniques for accelerating gradient-based optimization such as learning rate schedules. Finally, we show convergence rate analysis for both pbSGD and pbSGDM methods. The theoretical rates of convergence match the best known theoretical rates of convergence for SGD and SGDM methods on nonconvex functions.
3
Li, Sumu, Dequan Li e Yuheng Zhang. "Incorporating Nesterov’s Momentum into Distributed Adaptive Gradient Method for Online Optimization". In 2021 China Automation Congress (CAC). IEEE, 2021. http://dx.doi.org/10.1109/cac53003.2021.9727247.
4
Weng, Bowen, Huaqing Xiong, Yingbin Liang e Wei Zhang. "Analysis of Q-learning with Adaptation and Momentum Restart for Gradient Descent". In Twenty-Ninth International Joint Conference on Artificial Intelligence and Seventeenth Pacific Rim International Conference on Artificial Intelligence {IJCAI-PRICAI-20}. California: International Joint Conferences on Artificial Intelligence Organization, 2020. http://dx.doi.org/10.24963/ijcai.2020/422.
Abstract (sommario):
Existing convergence analyses of Q-learning mostly focus on the vanilla stochastic gradient descent (SGD) type of updates. Despite the Adaptive Moment Estimation (Adam) has been commonly used for practical Q-learning algorithms, there has not been any convergence guarantee provided for Q-learning with such type of updates. In this paper, we first characterize the convergence rate for Q-AMSGrad, which is the Q-learning algorithm with AMSGrad update (a commonly adopted alternative of Adam for theoretical analysis). To further improve the performance, we propose to incorporate the momentum restart scheme to Q-AMSGrad, resulting in the so-called Q-AMSGradR algorithm. The convergence rate of Q-AMSGradR is also established. Our experiments on a linear quadratic regulator problem demonstrate that the two proposed Q-learning algorithms outperform the vanilla Q-learning with SGD updates. The two algorithms also exhibit significantly better performance than the DQN learning method over a batch of Atari 2600 games.
5
Wu, Jui-Yu, e Pei-Ci Liu. "Identifying a Default of Credit Card Clients by using a LSTM Method: A Case Study". In 8th International Conference on Artificial Intelligence (ARIN 2022). Academy and Industry Research Collaboration Center (AIRCC), 2022. http://dx.doi.org/10.5121/csit.2022.121012.
Abstract (sommario):
Detecting fraudulent transactions is critical and challenging for financial banks and institutes. This study used a deep learning technique, which is a long short-term memory (LSTM) method, for identifying a default of credit card clients (an imbalanced dataset). To evaluate the performance of optimizers for the LSTM approach, this study employed three optimizers based on gradient methods, such as adaptive moment estimation (Adam), stochastic gradient descent with momentum (Sgdm) and root mean square propagation (Rmsprop). This study used 10-fold cross-validation. Moreover, this study compared the best numerical results of the LSTM method with those of supervised machine learning classifiers, which are back-propagation neural network (BPNN) with a gradient descent algorithm (GDA) and a scaled conjugate gradient algorithm (SCGA). Numerical results indicate that the LSTM-Adam and the BPNN-SCGA classifiers have identical performance, and that selecting an appropriate classification threshold value is important for an imbalanced dataset. Based on the numerical results, the LSTM-Adam classifier can be considered for dealing with credit scoring problems, which are binary classification problems.
6
Xu, Chenyuan, Kosuke Haruki, Taiji Suzuki, Masahiro Ozawa, Kazuki Uematsu e Ryuji Sakai. "Data-Parallel Momentum Diagonal Empirical Fisher (DP-MDEF):Adaptive Gradient Method is Affected by Hessian Approximation and Multi-Class Data". In 2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA). IEEE, 2022. http://dx.doi.org/10.1109/icmla55696.2022.00221.
7
Olivett, A., P. Corrao e M. A. Karami. "Flow Control and Separation Delay in Morphing Wing Aircraft Using Traveling Wave Actuation". In ASME 2020 Conference on Smart Materials, Adaptive Structures and Intelligent Systems. American Society of Mechanical Engineers, 2020. http://dx.doi.org/10.1115/smasis2020-2355.
Abstract (sommario):
Abstract This study examines the biomimicry of wave propagation, a mode of locomotion in aquatic life for the use-case of morphing aircraft surfaces for boundary layer control. Such motion is theorized to inject momentum into the flow on the upper surface of airfoils, and as a consequence, creates a forcible pressure gradient thereby increasing lift. It is thought that this method can be used to control flow separation and reduce likelihood of stall at high angles of attack. The motivation for such a mechanism is especially relevant for aircraft requiring abrupt maneuvers, and especially at high angles of attack as a safety measure against stalling. The actuation mechanism consists of lightweight piezoelectric ceramic transducers placed beneath the upper surface of an airfoil. An open-loop system controls surface morphing. A two-dimensional Fourier Transform technique is used to estimate traveling to standing wave ratio, which is verified analytically using Euler Bernoulli beam theory, and experimentally using a prototype wing. Propagating wave control is tuned and verified using a series of scanning laser vibrometry tests. A custom two-dimensional NACA 0018 airfoil tests the concept in a low-speed wind tunnel with approximate Reynolds Number of 50,000. Both traveling waves and the changes in lift and drag will be experimentally characterized.
8
Masood, Sarfaraz, M. N. Doja e Pravin Chandra. "Analysis of weight initialization methods for gradient descent with momentum". In 2015 International Conference on Soft Computing Techniques and Implementations (ICSCTI). IEEE, 2015. http://dx.doi.org/10.1109/icscti.2015.7489618.
9
Sun, Huikang, Lize Gu e Bin Sun. "Adathm: Adaptive Gradient Method Based on Estimates of Third-Order Moments". In 2019 IEEE Fourth International Conference on Data Science in Cyberspace (DSC). IEEE, 2019. http://dx.doi.org/10.1109/dsc.2019.00061.
10
Ren, Xuchun, e Sharif Rahman. "Robust Design Optimization by Adaptive-Sparse Polynomial Dimensional Decomposition". In ASME 2016 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference. American Society of Mechanical Engineers, 2016. http://dx.doi.org/10.1115/detc2016-59691.
Abstract (sommario):
This work proposes a new methodology for robust design optimization (RDO) of complex engineering systems. The method, capable of solving large-scale RDO problems, involves (1) an adaptive-sparse polynomial dimensional decomposition (AS-PDD) for stochastic moment analysis of a high-dimensional stochastic response, (2) a novel integration of score functions and AS-PDD for design sensitivity analysis, and (3) a multi-point design process, facilitating standard gradient-based optimization algorithms. Closed-form formulae are developed for first two moments and their design sensitivities. The method allow that both the stochastic moments and their design sensitivities can be concurrently determined from a single stochastic simulation or analysis. Precisely for this reason, the multi-point framework of the proposed method affords the ability of solving industrial-scale problems with large design spaces. The robust shape optimization of a three-hole bracket was accomplished, demonstrating the efficiency of the new method to solve industry-scale RDO problems.
Rapporti di organizzazioni sul tema "Adaptive gradient methods with momentum":
1
Sett, Dominic, Florian Waldschmidt, Alvaro Rojas-Ferreira, Saut Sagala, Teresa Arce Mojica, Preeti Koirala, Patrick Sanady et al. Climate and disaster risk analytics tool for adaptive social protection. United Nations University - Institute for Environment and Human Security, marzo 2022. http://dx.doi.org/10.53324/wnsg2302.
Abstract (sommario):
Adaptive Social Protection (ASP) as discussed in this report is an approach to enhance the well-being of communities at risk. As an integrated approach, ASP builds on the interface of Disaster Risk Management (DRM), Climate Change Adaptation (CCA) and Social Protection (SP) to address interconnected risks by building resilience, thereby overcoming the shortcomings of traditionally sectoral approaches. The design of meaningful ASP measures needs to be informed by specific information on risk, risk drivers and impacts on communities at risk. In contrast, a limited understanding of risk and its drivers can potentially lead to maladaptation practices. Therefore, multidimensional risk assessments are vital for the successful implementation of ASP. Although many sectoral tools to assess risks exist, available integrated risk assessment methods across sectors are still inadequate in the context of ASP, presenting an important research and implementation gap. ASP is now gaining international momentum, making the timely development of a comprehensive risk analytics tool even more important, including in Indonesia, where nationwide implementation of ASP is currently under way. OBJECTIVE: To address this gap, this study explores the feasibility of a climate and disaster risk analytics tool for ASP (CADRAT-ASP), combining sectoral risk assessment in the context of ASP with a more comprehensive risk analytics approach. Risk analytics improve the understanding of risks by locating and quantifying the potential impacts of disasters. For example, the Economics of Climate Adaptation (ECA) framework quantifies probable current and expected future impacts of extreme events and determines the monetary cost and benefits of specific risk management and adaptation measures. Using the ECA framework, this report examines the viability and practicality of applying a quantitative risk analytics approach for non-financial and non-tangible assets that were identified as central to ASP. This quantitative approach helps to identify cost-effective interventions to support risk-informed decision making for ASP. Therefore, we used Nusa Tenggara, Indonesia, as a case study, to identify potential entry points and examples for the further development and application of such an approach. METHODS & RESULTS: The report presents an analysis of central risks and related impacts on communities in the context of ASP. In addition, central social protection dimensions (SPD) necessary for the successful implementation of ASP and respective data needs from a theoretical perspective are identified. The application of the quantitative ECA framework is tested for tropical storms in the context of ASP, providing an operational perspective on technical feasibility. Finally, recommendations on further research for the potential application of a suitable ASP risk analytics tool in Indonesia are proposed. Results show that the ECA framework and its quantitative modelling platform CLIMADA successfully quantified the impact of tropical storms on four SPDs. These SPDs (income, access to health, access to education and mobility) were selected based on the results from the Hazard, Exposure and Vulnerability Assessment (HEVA) conducted to support the development of an ASP roadmap for the Republic of Indonesia (UNU-EHS 2022, forthcoming). The SPDs were modelled using remote sensing, gridded data and available global indices. The results illustrate the value of the outcome to inform decision making and a better allocation of resources to deliver ASP to the case study area. RECOMMENDATIONS: This report highlights strong potential for the application of the ECA framework in the ASP context. The impact of extreme weather events on four social protection dimensions, ranging from access to health care and income to education and mobility, were successfully quantified. In addition, further developments of CADRAT-ASP can be envisaged to improve modelling results and uptake of this tool in ASP implementation. Recommendations are provided for four central themes: mainstreaming the CADRAT approach into ASP, data and information needs for the application of CADRAT-ASP, methodological advancements of the ECA framework to support ASP and use of CADRAT-ASP for improved resilience-building. Specific recommendations are given, including the integration of additional hazards, such as flood, drought or heatwaves, for a more comprehensive outlook on potential risks. This would provide a broader overview and allow for multi-hazard risk planning. In addition, high-resolution local data and stakeholder involvement can increase both ownership and the relevance of SPDs. Further recommendations include the development of a database and the inclusion of climate and socioeconomic scenarios in analyses.