Memory Capacity of Heterogeneous Hopfield networks

[Preprint]

The classic Hopfield network, as a model of associative memory, assumes homogeneous connections and coding levels (i.e., activation rates of neurons in memory patterns). However, biological networks exhibit heterogeneous connection structures and coding levels. I derive an analytical formula for the memory capacity of such heterogeneous networks.

In general heterogeneous random networks, I demonstrate that maximizing memory capacity requires a correlation between the number of inward connections and the coding levels of neurons, presenting a normative prediction that is amenable to experimental testing.

This work bridges theoretical neuroscience and biological plausibility and offers insights into how structural and functional neural heterogeneity impacts memory storage.

Novelty Signals Across Brain Areas and Their Adaptation

[Article]

This project explores how primates detect novel objects by examining the neural mechanisms underlying computations of recency (how recently a stimulus was experienced) and sensory surprise (unexpected violations of predictions about incoming stimuli), using high-channel electrophysiology in primates.

We found that novelty detection is tightly linked to recency and sensory surprise computations across multiple brain regions.

Additionally, we observed that neurons exhibit varied learning and forgetting timescales related to novelty, with some adapting and forgetting quickly, while others learn and retain information over longer periods. A similar diversity of timescales exists across brain regions. This variability supports flexible behavior and learning.

Basal Forebrain Encodes Novelty and Salience Signals

[Article]

This project investigates the role of neurons in the basal forebrain (BF) in processing salience-related information, such as novelty, surprise, reward, and reward prediction. We identified two types of neurons, distinguished by their firing rate profiles and responses to stimuli.

Phasic Bursting Neurons: These neurons respond rapidly to salient external events, including unexpected rewards and novel objects.

Ramping Neurons: These neurons anticipate the timing of salient events, displaying ramping activity that is sensitive to the subject’s confidence in event timing.

The study concludes that these two neuronal types support complementary cognitive processes: one facilitates immediate responses to external stimuli, while the other contributes to anticipation and internal timing.