# Implementing engrams from a machine learning perspective: the relevance of a latent space

**Authors:** J Marco de Lucas  
**arXiv:** [2407.16616](https://arxiv.org/abs/2407.16616)  
**Date:** 2024-07-23  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2407.16616v1/

## Key points

- Engrams are neural representations of memories
- Engrams can be implemented biologically as autoencoders over recurrent neural networks
- Autoencoders consist of basic excitatory/inhibitory motifs and utilize credit assignment based on a simple homeostatic criterion
- Importance of the latent space within autoencoders is investigated
- Differences in connectome structures across species may be linked to cognitive capacities
- Human cognition may be constrained by brain structure, limiting cognitive abilities
- Machine learning systems offer an augmented perspective on understanding and processing information
- Need to explore and leverage enhanced capability for advancing understanding of cognitive processes

## Summary

In the study "Implementing engrams from a machine learning perspective: the relevance of a latent space" by J Marco de Lucas, the concept of engrams in the brain is explored through the lens of machine learning. The researchers propose that engrams, which are neural representations of memories, can be implemented biologically as autoencoders over recurrent neural networks. These autoencoders consist of basic excitatory/inhibitory motifs and utilize credit assignment based on a simple homeostatic criterion. The focus of this study is on investigating the importance of the latent space within these autoencoders. By analyzing differences in connectome structures across species, they suggest a potential link between these variances and cognitive capacities. A key insight highlighted in this research is that human cognition may be constrained by our brain's structure, limiting our cognitive abilities. In contrast, machine learning systems do not face this limitation, offering an augmented perspective on understanding and processing information. The study emphasizes the need to explore and leverage this enhanced capability for advancing our understanding of cognitive processes. Overall, this study provides valuable insights into how engrams can be understood and implemented from a machine learning standpoint, shedding light on the potential implications for cognitive research and artificial intelligence development.