Modelling continual learning in humans with Hebbian context gating and exponentially decaying task signals

@article{Flesch2022ModellingCL,
  title={Modelling continual learning in humans with Hebbian context gating and exponentially decaying task signals},
  author={Timo Flesch and D{\'a}vid G. Nagy and Andrew M. Saxe and Christopher Summerfield},
  journal={ArXiv},
  year={2022},
  volume={abs/2203.11560}
}
Humans can learn several tasks in succession with minimal mutual interference but perform more poorly when trained on multiple tasks at once. The opposite is true for standard deep neural networks. Here, we propose novel computational constraints for artificial neural networks, inspired by earlier work on gating in the primate prefrontal cortex, that capture the cost of interleaved training and allow the network to learn two tasks in sequence without forgetting. We augment standard stochastic… 

Figures from this paper

Avoiding Catastrophe: Active Dendrites Enable Multi-Task Learning in Dynamic Environments
TLDR
This article proposes a novel artificial neural network architecture that incorporates active dendrites and sparse representations into the standard deep learning framework and marks the first time a single architecture has achieved competitive results in both multi-task and continual learning settings.

References

SHOWING 1-10 OF 64 REFERENCES
A Neural Network Model of Continual Learning with Cognitive Control. arXiv:220204773 [cs, q-bio
  • 2022
Modelling continual learning in humans with Hebbian context gating and exponentially decaying task signals
TLDR
Novel computational constraints for artificial neural networks are proposed, inspired by earlier work on gating in the primate prefrontal cortex, that capture the cost of interleaved training and allow the network to learn two tasks in sequence without forgetting.
Geometry of sequence working memory in macaque prefrontal cortex
TLDR
A regular geometrical organization was discovered: the high-dimensional neural state space during the delay could be decomposed into a sum of low-dimensional subspaces, each storing the spatial location at a given ordinal rank, which could be generalized to novel sequences and explain monkey behavior.
A Neural Network Model of Continual Learning with Cognitive Control
TLDR
It is shown that neural networks equipped with a mechanism for cognitive control do not exhibit catastrophic forgetting when trials are blocked, and an advantage of blocking over interleaving when there is a bias for active maintenance in the control signal, implying a tradeoff between maintenance and the strength of control.
Texture-like representation of objects in human visual cortex
TLDR
The results suggest that the role of human category-selective visual cortex is not to explicitly encode objects but rather to provide a basis set of texture-like features that can be infinitely reconfigured to flexibly learn and identify new object categories.
Multi-task representations in human cortex transform along a sensory-to-motor hierarchy
TLDR
Compression-then-expansion organization in models emerged exclusively in a training regime where internal representations are highly optimized for sensory-to-motor transformation, and not under generic signal propagation.
Interleaved practice enhances memory and problem-solving ability in undergraduate physics
TLDR
In a domain that entails considerable amounts of problem-solving, replacing conventionally arranged homework with interleaved homework can (despite perceptions to the contrary) foster longer lasting and more generalizable learning.
Continual Learning in the Teacher-Student Setup: Impact of Task Similarity
TLDR
This work extends previous analytical work on two-layer networks in the teacher-student setup to multiple teachers and finds a complex interplay between both types of similarity, initial transfer/forgetting rates, maximum transfer/ forgetting, and long-term transfer/Forgetting.
Shared mechanisms underlie the control of working memory and attention.
TLDR
It is shown that shared neural mechanisms underlie the selection of items from working memory and attention to sensory stimuli and suggest that prefrontal cortex controls cognition by dynamically transforming representations to control what and when cognitive computations are engaged.
...
...