87,191 Hits in 2.0 sec

Meta-Q-Learning [article]

Rasool Fakoor, Pratik Chaudhari, Stefano Soatto, Alexander J. Smola
2020 arXiv   pre-print
This paper introduces Meta-Q-Learning (MQL), a new off-policy algorithm for meta-Reinforcement Learning (meta-RL). MQL builds upon three simple ideas.  ...  First, we show that Q-learning is competitive with state-of-the-art meta-RL algorithms if given access to a context variable that is a representation of the past trajectory.  ...  Our second contribution is an off-policy meta-RL algorithm named Meta-Q-Learning (MQL) that builds upon the above result.  ... 
arXiv:1910.00125v2 fatcat:vv7o6dhce5grzfrcrqiqao42pq

Revisiting Meta-Learning as Supervised Learning [article]

Wei-Lun Chao, Han-Jia Ye, De-Chuan Zhan, Mark Campbell, Kilian Q. Weinberger
2020 arXiv   pre-print
This view not only unifies meta-learning into an intuitive and practical framework but also allows us to transfer insights from supervised learning directly to improve meta-learning.  ...  Recent years have witnessed an abundance of new publications and approaches on meta-learning.  ...  Introduction Meta-learning, or learning to learn, is the sub-field of machine learning occupied with the search for the best An extended version of the paper titled "A Meta Understanding of Meta-Learning  ... 
arXiv:2002.00573v1 fatcat:f5fcejzwmrgbjp4pn64fvqp43a

Generating meta-learning tasks to evolve parametric loss for classification learning [article]

Zhaoyang Hai, Xiabi Liu, Yuchen Ren, Nouman Q. Soomro
2021 arXiv   pre-print
In existing meta-learning approaches, learning tasks for training meta-models are usually collected from public datasets, which brings the difficulty of obtaining a sufficient number of meta-learning tasks  ...  In this paper, we propose a meta-learning approach based on randomly generated meta-learning tasks to obtain a parametric loss for classification learning based on big data.  ...  Meta-learning for few-shot learning In the field of supervised learning, existing meta-learning methods are mainly developed for few-shot learning.  ... 
arXiv:2111.10583v1 fatcat:zgbhmbuewrdahbmygu6gazczv4

Meta-Learning GNN Initializations for Low-Resource Molecular Property Prediction [article]

Cuong Q. Nguyen, Constantine Kreatsoulas, Kim M. Branson
2020 arXiv   pre-print
Meanwhile advances in meta-learning have enabled state-of-the-art performances in few-shot learning benchmarks, naturally prompting the question: Can meta-learning improve deep learning performance in  ...  In this work, we assess the transferability of graph neural networks initializations learned by the Model-Agnostic Meta-Learning (MAML) algorithm - and its variants FO-MAML and ANIL - for chemical properties  ...  Meta-Learning The same GGNN architecture as the baselines is used for all three meta-learning algorithms.  ... 
arXiv:2003.05996v2 fatcat:enugsgd27bbc7jvlfuxixkfedy

ST^2: Small-data Text Style Transfer via Multi-task Meta-Learning [article]

Xiwen Chen, Kenny Q. Zhu
2020 arXiv   pre-print
In this work, we develop a meta-learning framework to transfer between any kind of text styles, including personal writing styles that are more fine-grained, share less content and have much smaller training  ...  We define the shared model with parameters θ as a meta-learner. The data for each task is divided in to a support set D s and a query set D q .  ...  Compared with other model-based meta-learning methods, modelagnostic meta-learning algorithm (MAML) utilizes only gradient information (Finn et al., 2017) .  ... 
arXiv:2004.11742v1 fatcat:j3tbgwdy4bhojmeu4sbu6zkope

Q-HeteLearn: A Progressive Learning approach for Classifying Meta-Paths in Heterogeneous Information Networks

Sadhana Kodali
2020 International Journal of Emerging Trends in Engineering Research  
The concept of Q-HeteLearn which is a Progressive Learning technique is introduced to improve the swift traversal of the objects in the meta-paths and to classify them.  ...  In this paper a novel approach called Q-HeteLearn a Progressive Learning method is introduced to classify the objects by traversing the meta-paths in the Heterogeneous Information Networks.  ...  But a meta-path expands as new objects add to it and new relationships exist. Therefore, the Deep Q-Learning strategy is also used to compare Q-HeteLearn.  ... 
doi:10.30534/ijeter/2020/35832020 fatcat:m6xvtws7mzhyzb526w4eme57da

MICK: A Meta-Learning Framework for Few-shot Relation Classification with Little Training Data [article]

Xiaoqing Geng, Xiwen Chen, Kenny Q. Zhu
2020 arXiv   pre-print
We also propose a few-shot learning framework for relation classification, which is particularly powerful when the training data is very small.  ...  Given an encoded query instance Q = (E q , r q ) where r q is to be predicted, class matching aims to match r q with some relation class r i ∈ R.  ...  Meta-learning is a popular method for few-shot learning and is widely investigated in computer vision (CV).  ... 
arXiv:2004.14164v1 fatcat:u65ygi6ahjekrbdx35pomhtqqe

Fingerprinting diverse nanoporous materials for optimal hydrogen storage conditions using meta-learning

Yangzesheng Sun, Robert F. DeJaco, Zhao Li, Dai Tang, Stephan Glante, David S. Sholl, Coray M. Colina, Randall Q. Snurr, Matthias Thommes, Martin Hartmann, J. Ilja Siepmann
2021 Science Advances  
Using data obtained from high-throughput Monte Carlo simulations for zeolites, metal-organic frameworks, and hyper–cross-linked polymers, we develop a meta-learning model that jointly predicts the adsorption  ...  Meta-learning gives higher accuracy and improved generalization compared to fitting a model separately to each material and allows us to identify the optimal hydrogen storage temperature with the highest  ...  Meta-learning for predicting gas adsorption loading, q, in NPMs. (A) A meta-learning problem setup.  ... 
doi:10.1126/sciadv.abg3983 pmid:34290094 fatcat:xfmb7jtfsbb7tarezbiabpafim

Meta-game Equilibrium for Multi-agent Reinforcement Learning [chapter]

Yang Gao, Joshua Zhexue Huang, Hongqiang Rong, Zhi-Hua Zhou
2004 Lecture Notes in Computer Science  
This paper proposes a multi-agent Q-learning algorithm called meta-game-Q learning that is developed from the meta-game equilibrium concept.  ...  We use the repeated prisoner's dilemma example to empirically demonstrate that the algorithm converges to meta-game equilibrium.  ...  A template for meta-game-Q reinforcement learning is presented in Table 2 .  ... 
doi:10.1007/978-3-540-30549-1_81 fatcat:7z2k6nlsxngu5igw7kl4xc6o5a

Meta-Learning Divergences of Variational Inference [article]

Ruqi Zhang, Yingzhen Li, Christopher De Sa, Sam Devlin, Cheng Zhang
2021 arXiv   pre-print
In this paper we propose a meta-learning algorithm to learn the divergence metric suited for the task of interest, automating the design of VI methods.  ...  In addition, we learn the initialization of the variational parameters without additional cost when our method is deployed in the few-shot learning scenarios.  ...  During meta-testing, a new task is sampled from p(T ), and the learned divergence D η is used to optimize the variational distribution q φ .  ... 
arXiv:2007.02912v2 fatcat:4zzyeegb3bb6hefcezwyr7hyeq

Adaptive-Control-Oriented Meta-Learning for Nonlinear Systems [article]

Spencer M. Richards, Navid Azizan, Jean-Jacques Slotine, Marco Pavone
2021 arXiv   pre-print
Our key insight is that we can better prepare the controller for deployment with control-oriented meta-learning of features in closed-loop simulation, rather than regression-oriented meta-learning of features  ...  Specifically, we meta-learn the adaptive controller with closed-loop tracking simulation as the base-learner and the average tracking error as the meta-objective.  ...  Meta-Learning Meta-learning is the tool we use to inject the downstream adaptive control application into offline learning from data.  ... 
arXiv:2103.04490v2 fatcat:uf7fzau2qjc6fk2oco3tjp73ku

Learning State-Dependent Losses for Inverse Dynamics Learning [article]

Kristen Morse, Neha Das, Yixin Lin, Austin S. Wang, Akshara Rai, Franziska Meier
2020 arXiv   pre-print
In this work, we propose to apply meta-learning to learn structured, state-dependent loss functions during a meta-training phase.  ...  We then replace standard losses with our learned losses during online adaptation tasks.  ...  Meta-Learning Dynamics Models There has also been research in meta-learning with a focus on meta-learning dynamics models, either with fast finetuning [26] or with online learning of a memory learning  ... 
arXiv:2003.04947v3 fatcat:c4l3qa4zqndedkyfijgzztoiai

Meta-Learning by Adjusting Priors Based on Extended PAC-Bayes Theory [article]

Ron Amit, Ron Meir
2019 arXiv   pre-print
We present a framework for meta-learning that is based on generalization error bounds, allowing us to extend various PAC-Bayes bounds to meta-learning.  ...  In meta-learning an agent extracts knowledge from observed tasks, aiming to facilitate learning of novel future tasks.  ...  Meta-Learning Problem Formulation The meta-learning problem formulation follows Pentina & Lampert (2014) .  ... 
arXiv:1711.01244v8 fatcat:uvbegjw6erezjebbinjrrejzpe

Meta-Learning without Memorization [article]

Mingzhang Yin, George Tucker, Mingyuan Zhou, Sergey Levine, Chelsea Finn
2020 arXiv   pre-print
In some domains, this makes meta-learning entirely inapplicable.  ...  Meta-learning has emerged as a promising technique for leveraging data from previous tasks to enable efficient learning of new tasks.  ...  Complete memorization in meta-learning is when the learned model ignores the task training data such that I(ŷ * ; D|x * , θ) = 0 (i.e., q(ŷ * |x * , θ, D) = q(ŷ * |x * , θ) = E D |x * [q(ŷ * |x * , θ,  ... 
arXiv:1912.03820v3 fatcat:dsunnq42zvdsxa372bdtkn56yi

Off-Policy Meta-Reinforcement Learning with Belief-based Task Inference

Takahisa Imagawa, Takuya Hiraoka, Yoshimasa Tsuruoka
2022 IEEE Access  
The agent learns a belief model over the task embedding space and trains a belief-conditional policy and Q-function.  ...  To alleviate this problem, we propose a novel off-policy meta-RL method, embedding learning and uncertainty evaluation (ELUE).  ...  LEARNING BELIEF-CONDITIONAL POLICY AND Q-FUNCTION In this section, we introduce how the ELUE agent learns a belief-conditional policy and Q-function in meta-training.  ... 
doi:10.1109/access.2022.3170582 fatcat:nkkav52wenbhrontkzmewehw7u
« Previous Showing results 1 — 15 out of 87,191 results