Distributed Evolution Strategies Using TPUs for Meta-Learning

被引:0
|
作者
Sheng, Alex [1 ]
He, Jun Yi Derek [2 ]
机构
[1] NYU, Tandon Sch Engn, New York, NY 10003 USA
[2] Woodlands Coll Pk High Sch, Acad Sci & Technol, The Woodlands, TX USA
关键词
meta-learning; evolution strategies; tpu;
D O I
10.1109/ssci47803.2020.9308334
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Meta-learning traditionally relies on backpropagation through entire tasks to iteratively improve a model's learning dynamics. However, this approach is computationally intractable when scaled to complex tasks. We propose a distributed evolutionary meta-learning strategy using Tensor Processing Units (TPUs) that is highly parallel and scalable to arbitrarily long tasks with no increase in memory cost. Using a Prototypical Network trained with evolution strategies on the Omniglot dataset, we achieved an accuracy of 98.4% on a 5-shot classification problem. Our algorithm used as much as 40 times less memory than automatic differentiation to compute the gradient, with the resulting model achieving accuracy within 1.3% of a backpropagation-trained equivalent (99.6%). We observed better classification accuracy as high as 99.1% with larger population configurations. We further experimentally validate the stability and performance of ES-ProtoNet across a variety of training conditions (varying population size, model size, number of workers, shot, way, ES hyperparameters, etc.). Our contributions are twofold: we provide the first assessment of evolutionary meta-learning in a supervised setting, and create a general framework for distributed evolution strategies on TPUs.
引用
收藏
页码:721 / 728
页数:8
相关论文
共 50 条
  • [1] Meta-learning and the evolution of cognition
    Veit, Walter
    Browning, Heather
    BEHAVIORAL AND BRAIN SCIENCES, 2024, 47
  • [2] CORBA infrastructure for distributed learning and meta-learning
    Werges, SC
    Naylor, DL
    KNOWLEDGE-BASED SYSTEMS, 2002, 15 (1-2) : 139 - 144
  • [3] Distributed Meta-Learning with Networked Agents
    Kayaalp, Mert
    Vlaski, Stefan
    Sayed, Ali H.
    29TH EUROPEAN SIGNAL PROCESSING CONFERENCE (EUSIPCO 2021), 2021, : 1361 - 1365
  • [4] Learning how to learn: Meta-learning strategies for the challenges of learning pharmacology
    Alton, Suzanne
    NURSE EDUCATION TODAY, 2016, 38 : 2 - 4
  • [5] Meta-Learning Effective Exploration Strategies for Contextual Bandits
    Sharaf, Amr
    Daume, Hal, III
    THIRTY-FIFTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, THIRTY-THIRD CONFERENCE ON INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE AND THE ELEVENTH SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE, 2021, 35 : 9541 - 9548
  • [6] EVALUATING PREDICTION STRATEGIES IN AN ENHANCED META-LEARNING FRAMEWORK
    Cacoveanu, Silviu
    Vidrighin, Camelia
    Potolea, Rodica
    ICEIS 2010: PROCEEDINGS OF THE 12TH INTERNATIONAL CONFERENCE ON ENTERPRISE INFORMATION SYSTEMS, VOL 2: ARTIFICIAL INTELLIGENCE AND DECISION SUPPORT SYSTEMS, 2010, : 148 - 156
  • [7] Online Meta-Learning via Learning with Layer-Distributed Memory
    Babu, Sudarshan
    Savarese, Pedro
    Maire, Michael
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 34 (NEURIPS 2021), 2021, 34
  • [8] A distributed meta-learning system for Chinese entity relation extraction
    Li, Lishuang
    Zhang, Jing
    Jin, Liuke
    Guo, Rui
    Huang, Degen
    NEUROCOMPUTING, 2015, 149 : 1135 - 1142
  • [9] LEARNING AMD META-LEARNING USING A BLACK BOX
    WEINBERG, GM
    CYBERNETICA, 1971, 14 (02): : 132 - &
  • [10] Learning Meta-Learning (LML) dataset: Survey data of meta-learning parameters
    Corraya, Sonia
    Al Mamun, Shamim
    Kaiser, M. Shamim
    DATA IN BRIEF, 2023, 51