Stable Lifelong Learning: Spiking neurons as a solution to instability in plastic neural networks

Schmidgall, Samuel; Hays, Joe

doi:10.1145/3517343.3517345

Cited by 5 publications

(3 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The use of structured connectomes could prevent highly dynamic weights from deteriorating behavior. This differs from a more general architecture where the behavior would change much more dramatically with small changes in weights (Schmidgall & Hays (2022b)). These perturbations become more dramatic when weights are changing simultaneously and independently.…”

Section: Discussion and Future Workmentioning

confidence: 92%

Biological connectomes as a representation for the architecture of artificial neural networks

Schmidgall¹,

Schuman²,

Parsa³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

Grand efforts in neuroscience are working toward mapping the connectomes of many new species, including the near completion of the Drosophila melanogaster.It is important to ask whether these models could benefit artificial intelligence. In this work we ask two fundamental questions: (1) where and when biological connectomes can provide use in machine learning, (2) which design principles are necessary for extracting a good representation of the connectome. Toward this end, we translate the motor circuit of the C. Elegans nematode into artificial neural networks at varying levels of biophysical realism and evaluate the outcome of training these networks on motor and non-motor behavioral tasks. We demonstrate that biophysical realism need not be upheld to attain the advantages of using biological circuits. We also establish that, even if the exact wiring diagram is not retained, the architectural statistics provide a valuable prior. Finally, we show that while the C. Elegans locomotion circuit provides a powerful inductive bias on locomotion problems, its structure may hinder performance on tasks unrelated to locomotion such as visual classification problems.

show abstract

Section: Discussion and Future Workmentioning

confidence: 92%

Biological connectomes as a representation for the architecture of artificial neural networks

Schmidgall¹,

Schuman²,

Parsa³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…There have also been many previous contributions toward neuromodulated plasticity in non-spiking Artificial Neural Networks (ANNs) (62)(63)(64)(65)(66). However, plastic ANNs have been demonstrated to struggle maintaining functional stability across time due to their continuous nature which causes synapses to be in a constant state of change (41). The effect of this instability was shown to not disturb the performance as significantly in plastic SNNs as it did on plastic ANNs.…”

Section: Discussionmentioning

confidence: 99%

“…Learning how to learn online with synaptic plasticity through gradient descent. In learning applications with networks of spiking neurons, synaptic plasticity rules have historically been optimized through black-box optimization techniques such as evolutionary strategies (39)(40)(41), genetic algorithms (42,43), or Bayesian optimization (44,45). This is because spiking dynamics are inherently non-differentiable, and non-differentiable computations prevent gradient descent from being harnessed for optimization.…”

Section: Learning In Network With Plastic Synapsesmentioning

confidence: 99%

Learning to learn online with neuromodulated synaptic plasticity in spiking neural networks

Schmidgall

Hays

2022

Preprint

Self Cite

View full text Add to dashboard Cite

We propose that in order to harness our understanding of neuroscience toward machine learning, we must first have powerful tools for training brain-like models of learning. Although substantial progress has been made toward understanding the dynamics of learning in the brain, neuroscience-derived models of learning have yet to demonstrate the same performance capabilities as methods in deep learning such as gradient descent. Inspired by the successes of machine learning using gradient descent, we demonstrate that models of neuromodulated synaptic plasticity from neuroscience can be trained in Spiking Neural Networks (SNNs) with a framework of learning to learn through gradient descent to address challenging online learning problems. This framework opens a new path toward developing neuroscience inspired online learning algorithms.

show abstract

Platform-Based Design of Embedded Neuromorphic Systems

Varshika,

Das

2023

Embedded Machine Learning for Cyber-Physical, IoT, and Edge Computing

View full text Add to dashboard Cite

Stable Lifelong Learning: Spiking neurons as a solution to instability in plastic neural networks

Cited by 5 publications

References 15 publications

Biological connectomes as a representation for the architecture of artificial neural networks

Biological connectomes as a representation for the architecture of artificial neural networks

Learning to learn online with neuromodulated synaptic plasticity in spiking neural networks

Platform-Based Design of Embedded Neuromorphic Systems

Contact Info

Product

Resources

About