# Continual familiarity decoding from recurrent connections in spiking networks

## TL;DR

This paper introduces a spiking neural network model that can recognize familiar stimuli using brain-like mechanisms, outperforming traditional networks in certain tasks.

## Contribution

The novel contribution is a spiking network with unsupervised learning that encodes and decodes familiarity memory using both frequency and temporal coding.

## Key findings

- Familiarity can be decoded from spiking activity using both spike count and synchrony.
- Temporal coding improves performance under sparse input conditions.
- The model outperforms LSTMs in continual familiarity detection tasks.

## Abstract

Familiarity memory enables recognition of previously encountered inputs as familiar without recalling detailed stimuli information, which supports adaptive behavior across various timescales. We present a spiking neural network model with lateral connectivity shaped by unsupervised spike-timing-dependent plasticity (STDP) that encodes familiarity via local plasticity events. We show that familiarity can be decoded from network activity using both frequency (spike count) and temporal (spike synchrony) characteristics of spike trains. Temporal coding demonstrates enhanced performance under sparse input conditions, consistent with the principles of sparse coding observed in the brain. We also show how connectivity structure supports each decoding strategy, revealing different plasticity regimes. Our approach outperforms LSTM in temporal generalizability on the continual familiarity detection task, with input stimuli being naturally encoded in the recurrent connectivity without a separate training stage.

The ability to recognize a familiar stimulus without recalling its specific details is known as familiarity memory, and is fundamental to how animals and artificial agents learn and adapt to the environment. Our study explores how familiarity memory can be represented in a recurrent spiking network − a biologically-inspired computational model. Our model continuously encodes the familiarity of incoming stimuli in recurrent interneuronal connections via a local learning mechanism called Hebbian plasticity. We show that the structure of recurrent connectivity defines the firing activity of a network, and the familiarity information can be decoded from such activity using frequency (spike count) and temporal (spike synchrony) metrics. We demonstrate that a recurrent spiking network with unsupervised local learning recognizes familiar inputs across different timescales more accurately than an artificial neural network (ANN). Our findings advance the understanding of familiarity memory and suggest that spiking networks can be useful in real-world tasks, such as continual learning in dynamic environments.