A deep learning-inspired model of the hippocampus as storage device of the brain extended dataset

TL;DR

This paper proposes a neural network-inspired model of the hippocampus as a storage system for the brain's dataset, emphasizing its role in memory and feature learning, inspired by machine learning principles.

Contribution

It introduces a novel perspective that the hippocampus functions primarily as a dataset storage device, based on insights from neural network learning processes.

Findings

Hippocampus stores the brain's dataset for repeated learning.

High-level features are encoded in cortical neurons.

The model bridges neural network principles with biological memory architecture.

Abstract

The standard model of memory consolidation foresees that memories are initially recorded in the hippocampus, while features that capture higher-level generalisations of data are created in the cortex, where they are stored for a possibly indefinite period of time. Computer scientists have sought inspiration from nature to build machines that exhibit some of the remarkable properties present in biological systems. One of the results of this effort is represented by artificial neural networks, a class of algorithms that represent the state of the art in many artificial intelligence applications. In this work, we reverse the inspiration flow and use the experience obtained from neural networks to gain insight into the design of brain architecture and the functioning of memory. Our starting observation is that neural networks learn from data and need to be exposed to each data record many times during learning: this requires the storage of the entire dataset in computer memory. Our thesis is that the same holds true for the brain and the main role of the hippocampus is to store the "brain dataset", from which high-level features are learned and encoded in cortical neurons.