Hippocampus as a generative circuit for predictive coding of future sequences (unable to locate Whit
Abstract
The interaction between hippocampus and cortex is key to the formation of long-term memories and abstract representation learning. T,he sequences of inputs experienced during the day is stored in the hippocampus and is replayed to the cortex during sleep as part of, the process of memory consolidation (Hayes, 2021). Based on anatomical wiring and transmission delays in the hippocampus, we propo,se to build a self-supervised predictive recurrent autoencoder neural network that can predict sequences of inputs. This framework e,xtends predictive coding in the time axis and incorporates predictive features in Bayes filters for temporal prediction. The researc,h goals of the proposal are 1) to use deep learning to create a predictive recurrent autoencoder of the hippocampus that can learn s,equences and to analyze the capacity and capabilities of the model; 2) to create predictive model of the hippocampus using the same,architecture as the deep learning model but with biologically plausible learning algorithms; and 3) to generalize the hippocampus mo,del to stacked autoencoder model for the neocortex. The consequences of this research are twofold: First, the proposed predictive r,ecurrent autoencoder neural network model can enhance existing deep learning networks and transformer models with long-term memory s,ystem; second, the predictive recurrent autoencoder with biologically plausible learning algorithms makes strong predictions for the, functions of neurons in different parts of the hippocampus that can be empirically tested.This research has the potential for high,impact by enhancing existing AI systems and others that the Navy is developing. Our proposed predictive recurrent autoencoder netwo,rk models are like co-processors for learning and generating sequences.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 06, 2022
- Source ID
- N000142312069
Entities
People
- Terrence J. Sejnowski
Organizations
- Office of Naval Research
- United States Navy
- University of California, San Diego