Stimulus-to-stimulus learning in RNNs with cortical inductive biases

TL;DR

This paper introduces a biologically plausible model of how the brain learns associations between stimuli using realistic neural structures and learning rules.

Contribution

A novel recurrent neural network model that incorporates cortical inductive biases to implement stimulus substitution learning.

Findings

The model can learn complex conditioning phenomena using biologically plausible learning rules.

Hebbian learning rules fail to generalize associations without task-specific tuning.

Multi-compartment neurons enable robust and efficient learning under mixed representations.

Abstract

Animals learn to predict external contingencies from experience through a process of conditioning. A natural mechanism for conditioning is stimulus substitution, whereby the neuronal response to the CS becomes increasingly identical to that of the US. We propose a recurrent neural network model of stimulus substitution which leverages two forms of inductive bias pervasive in the cortex: representational inductive bias in the form of mixed stimulus representations, and architectural inductive bias in the form of two-compartment pyramidal neurons that have been shown to serve as a fundamental unit of cortical associative learning. The properties of these neurons allow for a biologically plausible learning rule that implements stimulus substitution, utilizing only information available locally at the synapses. We show that the model generates a wide array of conditioning phenomena, and can learn large numbers of associations with an amount of training commensurate with animal experiments, without relying on parameter fine-tuning for each individual experimental task. In contrast, we show that commonly used Hebbian rules fail to learn generic stimulus-stimulus associations with mixed selectivity, and require task-specific parameter fine-tuning. Our framework highlights the importance of multi-compartment neuronal processing in the cortex, and showcases how it might confer cortical animals the evolutionary edge.

Animals learn to anticipate important events by forming associations between neutral cues (like a bell) and meaningful outcomes (like food). This process, known as conditioning, is fundamental to survival. Traditional, Hebbian models of synaptic plasticity ("fire together-wire together") are able to recapitulate these behavioral phenomena at the neuronal level, yet they rely on the simplifying assumption that individual neuronal populations are responsible for a specific association. This assumption does not hold under the current established view of mixed representations, particularly in the cerebral cortex. To address this limitation, we develop a biologically plausible synaptic plasticity model that implements predictive learning within single pyramidal neurons in the cortex. Our model is able to account for a host of conditioning phenomena, even when individual neurons respond to multiple stimuli. Compared to Hebbian rules, we show that our learning rule is robust to hyperparameter and experimental design changes, as it utilizes biologically plausible self-supervision. Overall, our work helps explain how the structure of pyramidal neurons in the mammalian cortex may allow cortical animals to more efficiently pack associations in the cortex, leading to optimized cognition under biologically imposed constraints.

Figures