Neural representational geometry underlies few-shot concept learning

被引:22
|
作者
Sorscher, Ben [1 ]
Ganguli, Surya [1 ,2 ]
Sompolinsky, Haim [3 ,4 ,5 ]
机构
[1] Stanford Univ, Dept Appl Phys, Stanford, CA 94305 USA
[2] Stanford Univ, Stanford Inst Human Ctr Artificial Intelligence, Stanford, CA 94305 USA
[3] Harvard Univ, Ctr Brain Sci, Cambridge, MA 02138 USA
[4] Hebrew Univ Jerusalem, Racah Inst Phys, IL-9190501 Jerusalem, Israel
[5] Hebrew Univ Jerusalem, Edmond & Lily Safra Ctr Brain Sci, IL-9190501 Jerusalem, Israel
关键词
few-shot learning; neural networks; ventral visual stream; population coding; OBJECT REPRESENTATIONS; HIERARCHICAL-MODELS; RECOGNITION; NETWORK; LEVEL;
D O I
10.1073/pnas.2200800119
中图分类号
O [数理科学和化学]; P [天文学、地球科学]; Q [生物科学]; N [自然科学总论];
学科分类号
07 ; 0710 ; 09 ;
摘要
Understanding the neural basis of the remarkable human cognitive capacity to learn novel concepts from just one or a few sensory experiences constitutes a fundamental problem. We propose a simple, biologically plausible, mathematically tractable, and computationally powerful neural mechanism for few-shot learning of naturalistic concepts. We posit that the concepts that can be learned from few examples are defined by tightly circumscribed manifolds in the neural firing-rate space of higher-order sensory areas. We further posit that a single plastic downstream readout neuron learns to discriminate new concepts based on few examples using a simple plasticity rule. We demonstrate the computational power of our proposal by showing that it can achieve high few-shot learning accuracy on natural visual concepts using both macaque inferotemporal cortex representations and deep neural network (DNN) models of these representations and can even learn novel visual concepts specified only through linguistic descriptors. Moreover, we develop a mathematical theory of few-shot learning that links neurophysiology to predictions about behavioral outcomes by delineating several fundamental and measurable geometric properties of neural representations that can accurately predict the few-shot learning performance of naturalistic concepts across all our numerical simulations. This theory reveals, for instance, that high-dimensional manifolds enhance the ability to learn new concepts from few examples. Intriguingly, we observe striking mismatches between the geometry of manifolds in the primate visual pathway and in trained DNNs. We discuss testable predictions of our theory for psychophysics and neurophysiological experiments.
引用
收藏
页数:12
相关论文
共 50 条
  • [1] Learning about few-shot concept learning
    Ananya Rastogi
    [J]. Nature Computational Science, 2022, 2 : 698 - 698
  • [2] Learning about few-shot concept learning
    Rastogi, Ananya
    [J]. NATURE COMPUTATIONAL SCIENCE, 2022, 2 (11): : 698 - 698
  • [3] Neural Snowball for Few-Shot Relation Learning
    Gao, Tianyu
    Han, Xu
    Xie, Ruobing
    Liu, Zhiyuan
    Lin, Fen
    Lin, Leyu
    Sun, Maosong
    [J]. THIRTY-FOURTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, THE THIRTY-SECOND INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE CONFERENCE AND THE TENTH AAAI SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE, 2020, 34 : 7772 - 7779
  • [4] Blessing of dimensionality at the edge and geometry of few-shot learning
    Tyukin, Ivan Y.
    Gorban, Alexander N.
    McEwan, Alistair A.
    Meshkinfamfard, Sepehr
    Tang, Lixin
    [J]. INFORMATION SCIENCES, 2021, 564 : 124 - 143
  • [5] Few-Shot Novel Concept Learning for Semantic Parsing
    Dan, Soham
    Bastani, Osbert
    Roth, Dan
    [J]. FINDINGS OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS, EMNLP 2021, 2021, : 2064 - 2075
  • [6] Few-Shot Few-Shot Learning and the role of Spatial Attention
    Lifchitz, Yann
    Avrithis, Yannis
    Picard, Sylvaine
    [J]. 2020 25TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION (ICPR), 2021, : 2693 - 2700
  • [7] Hybrid Graph Neural Networks for Few-Shot Learning
    Yu, Tianyuan
    He, Sen
    Song, Yi-Zhe
    Xiang, Tao
    [J]. THIRTY-SIXTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE / THIRTY-FOURTH CONFERENCE ON INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE / THE TWELVETH SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE, 2022, : 3179 - 3187
  • [8] Lexicon Learning for Few-Shot Neural Sequence Modeling
    Akyurek, Ekin
    Andreas, Jacob
    [J]. 59TH ANNUAL MEETING OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS AND THE 11TH INTERNATIONAL JOINT CONFERENCE ON NATURAL LANGUAGE PROCESSING (ACL-IJCNLP 2021), VOL 1, 2021, : 4934 - 4946
  • [9] Hierarchical Graph Neural Networks for Few-Shot Learning
    Chen, Cen
    Li, Kenli
    Wei, Wei
    Zhou, Joey Tianyi
    Zeng, Zeng
    [J]. IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, 2022, 32 (01) : 240 - 252
  • [10] Fuzzy Graph Neural Network for Few-Shot Learning
    Wei, Tong
    Hou, Junlin
    Feng, Rui
    [J]. 2020 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN), 2020,