149 Hits in 7.0 sec

Multi-Domain Learning by Meta-Learning: Taking Optimal Steps in Multi-Domain Loss Landscapes by Inner-Loop Learning [article]

Anthony Sicilia, Xingchen Zhao, Davneet Minhas, Erin O'Connor, Howard Aizenstein, William Klunk, Dana Tudorascu, Seong Jae Hwang
2021 arXiv   pre-print
Specifically, we take inner-loop gradient steps to dynamically estimate posterior distributions over the hyperparameters of our loss function.  ...  To this end, we consider a weighted loss function and extend it to an effective procedure by employing techniques from the recently active area of learning-to-learn (meta-learning).  ...  INTRODUCTION In this paper, we consider the problem of Multi-Domain Learning (MDL) in which the goal is to take labeled data from some collection of domains {D i } i and minimize the risk on all of these  ... 
arXiv:2102.13147v1 fatcat:bc7ql5gudrag7nw6ff2oewkxf4

Meta-Learning in Neural Networks: A Survey [article]

Timothy Hospedales, Antreas Antoniou, Paul Micaelli, Amos Storkey
2020 arXiv   pre-print
The field of meta-learning, or learning-to-learn, has seen a dramatic rise in interest in recent years.  ...  This survey describes the contemporary meta-learning landscape.  ...  (ii) Reducing the inevitable gradient degradation problems whose severity increases with the number of inner loop optimization steps.  ... 
arXiv:2004.05439v2 fatcat:3r23tsxxkfbgzamow5miglkrye

Meta-Learning in Neural Networks: A Survey

Timothy M Hospedales, Antreas Antoniou, Paul Micaelli, Amos J. Storkey
2021 IEEE Transactions on Pattern Analysis and Machine Intelligence  
In this survey we describe the contemporary meta-learning landscape.  ...  We first discuss definitions of meta-learning and position it with respect to related fields, such as transfer learning, multi-task learning, and hyperparameter optimization.  ...  (ii) Reducing the inevitable gradient degradation problems whose severity increases with the number of inner loop optimization steps.  ... 
doi:10.1109/tpami.2021.3079209 pmid:33974543 fatcat:wkzeodki4fbcnjlcczn4mr6kry

MetalGAN: Multi-Domain Label-Less Image Synthesis Using cGANs and Meta-Learning [article]

Tomaso Fontanini, Eleonora Iotti, Luca Donati, Andrea Prati
2020 arXiv   pre-print
This is achieved by combining a conditional Generative Adversarial Network (cGAN) for image generation and a Meta-Learning algorithm for domain switch, and we called our approach MetalGAN.  ...  In fact, a single multi-domain network would allow greater flexibility and robustness in the image synthesis task than other approaches.  ...  Intuitively, such a representation permits the model to easily move towards the optimal one in few steps of the inner training loop.  ... 
arXiv:1912.02494v2 fatcat:boweo35gwrflljbc3dlgbg7r44

Meta-learning based Alternating Minimization Algorithm for Non-convex Optimization [article]

Jingyuan Xia, Shengxi Li, Jun-Jie Huang, Imad Jaimoukha, Deniz Gunduz
2022 arXiv   pre-print
To tackle these issues, we propose a meta-learning based alternating minimization (MLAM) method, which aims to minimize a partial of the global losses over iterations instead of carrying minimization on  ...  In this paper, we propose a novel solution for non-convex problems of multiple variables, especially for those typically solved by an alternating minimization (AM) strategy that splits the original optimization  ...  This allows for variable updated at the inner loops being guided by global loss information from the outer loops, thus achieving a global scope optimization. 2) Ground-level Meta-learning: The general  ... 
arXiv:2009.04899v7 fatcat:ddw4js7kljgqnpm2xe2pd56uea

Meta-Learning Initializations for Image Segmentation [article]

Sean M. Hendryx, Andrew B. Leach, Paul D. Hein, Clayton T. Morrison
2020 arXiv   pre-print
We also construct a small benchmark dataset, FP-k, for the empirical study of how meta-learning systems perform in both few- and many-shot settings.  ...  We show state of the art results on the FSS-1000 dataset by meta-training EfficientLab with FOMAML and using Bayesian optimization to infer the optimal test-time adaptation routine hyperparameters.  ...  Hyperparameter value Meta-batch size 5 Meta-steps 50000 Initial meta-learning rate 0.1 Final meta-learning rate 1.e − 5 Inner batch size 8 Inner steps 5 Inner learning rate 0.005 Final  ... 
arXiv:1912.06290v4 fatcat:igdnxj2z7zgjjf75mvzeisfcgu

Bootstrapped Meta-Learning [article]

Sebastian Flennerhag and Yannick Schroecker and Tom Zahavy and Hado van Hasselt and David Silver and Satinder Singh
2022 arXiv   pre-print
We achieve a new state-of-the art for model-free agents on the Atari ALE benchmark and demonstrate that it yields both performance and efficiency gains in multi-task meta-learning.  ...  Meta-learning empowers artificial intelligence to increase its efficiency by learning how to learn. Unlocking this potential involves overcoming a challenging meta-optimisation problem.  ...  This work was funded by DeepMind.  ... 
arXiv:2109.04504v2 fatcat:qbxvdo44xbhjdnzndjsp5frn4q

Meta-learning with differentiable closed-form solvers [article]

Luca Bertinetto, João F. Henriques, Philip H.S. Torr, Andrea Vedaldi
2019 arXiv   pre-print
While normally the cost of the matrix operations involved in such a process would be significant, by using the Woodbury identity we can make the small number of examples work to our advantage.  ...  In this paper, we propose to use these fast convergent methods as the main adaptation mechanism for few-shot learning.  ...  Since eq. (7) takes a similar form to ridge regression, we can use it for meta-learning in the same way as in section 3.2, with the difference that a small number of steps (eq.  ... 
arXiv:1805.08136v3 fatcat:xw4gx75hubf3fkawph73wn2ayu

Meta-Learning Update Rules for Unsupervised Representation Learning [article]

Luke Metz, Niru Maheswaranathan, Brian Cheung, Jascha Sohl-Dickstein
2019 arXiv   pre-print
In this work, we propose instead to directly target later desired tasks by meta-learning an unsupervised learning rule which leads to representations useful for those tasks.  ...  We show that the meta-learned update rule produces useful features and sometimes outperforms existing unsupervised learning techniques.  ...  The inner loop of our meta-learning process trains this base model via iterative application of our learned update rule.  ... 
arXiv:1804.00222v3 fatcat:edrxrao4svfxvo7kablwnszkcy

Meta Learning Black-Box Population-Based Optimizers [article]

Hugo Siqueira Gomes, Benjamin Léger, Christian Gagné
2021 arXiv   pre-print
We suggest a general modeling of population-based algorithms that result in Learning-to-Optimize POMDP (LTO-POMDP), a meta-learning framework based on a specific partially observable Markov decision process  ...  This paper addresses this issue by proposing the use of meta-learning to infer population-based black-box optimizers that can automatically adapt to specific classes of problems.  ...  A base (inner) loop is defined by the optimization rollouts of a parametrized optimizer on a single task and a meta (or outer) loop that corresponds to successive updates of its parameters based on the  ... 
arXiv:2103.03526v1 fatcat:ckngsz36afcyhmsyscjc4zs2zm

Multimodality in Meta-Learning: A Comprehensive Survey [article]

Yao Ma, Shilin Zhao, Weixiao Wang, Yaoman Li, Irwin King
2022 arXiv   pre-print
This survey provides a comprehensive overview of the multimodality-based meta-learning landscape in terms of the methodologies and applications.  ...  We then propose a new taxonomy to discuss typical meta-learning algorithms in multimodal tasks systematically. We investigate the contributions of related papers and summarize them by our taxonomy.  ...  The main idea is to wrap up the data augmentation strategy in optimization steps of the inner meta-learning process [7] , with the augmentation parameterized and learned by the outer optimization in the  ... 
arXiv:2109.13576v2 fatcat:khofp6ldxrcafa7iyk7g6pxchi

Large-Scale Meta-Learning with Continual Trajectory Shifting [article]

Jaewoong Shin and Hae Beom Lee and Boqing Gong and Sung Ju Hwang
2022 arXiv   pre-print
inner-gradient steps.  ...  In this paper, we first show that allowing the meta-learners to take a larger number of inner gradient steps better captures the structure of heterogeneous and large-scale task distributions, thus results  ...  In our case, the number of inner-gradient steps k used to compute each meta-gradient determines the complexity of the meta-training loss landscape.  ... 
arXiv:2102.07215v3 fatcat:caq72vxclrbrlnodxdsxsf55hq

Meta-learning by the baldwin effect

Chrisantha Fernando, Jakub Sygnowski, Simon Osindero, Jane Wang, Tom Schaul, Denis Teplyashin, Pablo Sprechmann, Alexander Pritzel, Andrei Rusu
2018 Proceedings of the Genetic and Evolutionary Computation Conference Companion on - GECCO '18  
hyperparameters, and permits effectively any number of gradient updates in the inner loop.  ...  Furthermore it can genetically accommodate strong learning biases on the same set of problems as a recent machine learning algorithm called MAML "Model Agnostic Meta-Learning" which uses second-order gradients  ...  It requires a differentiable learning procedure to backpropagate into the reference parameter values, and even then it is limited in the number of gradient steps in the inner learning loop that can be  ... 
doi:10.1145/3205651.3205763 dblp:conf/gecco/FernandoSOWSTSP18 fatcat:gdtvorekirgsvhhs5acq3dufaq

Meta-Learning Acquisition Functions for Transfer Learning in Bayesian Optimization [article]

Michael Volpp, Lukas P. Fröhlich, Kirsten Fischer, Andreas Doerr, Stefan Falkner, Frank Hutter, Christian Daniel
2020 arXiv   pre-print
Transferring knowledge across tasks to improve data-efficiency is one of the open key challenges in the field of global black-box optimization.  ...  We propose a novel transfer learning method to obtain customized optimizers within the well-established framework of Bayesian optimization, allowing our algorithm to utilize the proven generalization capabilities  ...  To approximate the gradients of the PPO loss function, we record a batch of episodes in the inner loop, i.e., a set of (s t , a t , r t )-tuples, by rolling out the current policy π θi .  ... 
arXiv:1904.02642v5 fatcat:fqyqmppyjneuxld4ex53cjjhu4

Model Based Meta Learning of Critics for Policy Gradients [article]

Sarah Bechtle, Ludovic Righetti, Franziska Meier
2022 arXiv   pre-print
In this paper we present a framework to meta-learn the critic for gradient-based policy learning.  ...  However, learning representations that generalize quickly to new scenarios is still an open research problem in reinforcement learning.  ...  Given the task loss our framework differentiates through the inner loop to update φ.  ... 
arXiv:2204.02210v1 fatcat:maanjwrcyngidmhbb7levmac7q
« Previous Showing results 1 — 15 out of 149 results