https://github.com/saran-nns/neurogenesis
Investigate the necessary conditions to add new neurons in the brain inspired neural network models
https://github.com/saran-nns/neurogenesis
Last synced: about 2 months ago
JSON representation
Investigate the necessary conditions to add new neurons in the brain inspired neural network models
- Host: GitHub
- URL: https://github.com/saran-nns/neurogenesis
- Owner: Saran-nns
- License: mit
- Created: 2022-02-23T09:06:28.000Z (over 3 years ago)
- Default Branch: main
- Last Pushed: 2024-03-25T22:03:06.000Z (about 1 year ago)
- Last Synced: 2025-02-09T15:38:48.889Z (4 months ago)
- Language: Python
- Homepage:
- Size: 241 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Neurogenesis
This project will investigate the necessary conditions to add new neurons in the brain inspired neural network models.
Relevant articles:
1. New neurons and new memories: how does adult hippocampal neurogenesis affect learning and memory? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2886712/
2. Prefrontal–hippocampal interactions in episodic memory https://www.nature.com/articles/nrn.2017.74#:~:text=Key%20Points,interact%20bidirectionally%20through%20oscillatory%20synchrony.
PFC and hippocampus communicate bidirectionally through oscillatory synchrony.
3. MEMORY CAPACITY OF NEURAL NETWORKS WITH THRESHOLD AND RELU ACTIVATIONS https://arxiv.org/pdf/2001.06938.pdf
4. Optimal modularity and memory capacity of neural reservoirs https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6497001/
5. Current understanding of adult neurogenesis in the mammalian brain: how does adult neurogenesis decrease with age?https://inflammregen.biomedcentral.com/articles/10.1186/s41232-020-00122-x
The attenuation of neurogenesis in the mouse SGZ has been reported to result in the impairment of new memory formation. These new neurons are also important for the formation of spatial memories. Furthermore, new integrated neurons in the DG have the function of organizing past memories and alleviating the stress response
6. Decomposing the local arrow of time in interacting systems https://arxiv.org/pdf/2112.14721.pdf
Study entropy production and the arrow of time. Treating each variable independently leads to overestimate. Time revesibility?
7. Older circuits get richer: neurogenesis timeline shapes hubness and neurogenetic profiles of the adult human brain https://www.biorxiv.org/content/10.1101/2022.04.01.486541v2
8.
# Goals
1. modelling neurogenesis in deep self organizing recurrent networks and online learning
2. Alleviate catastrophic forgetting phenomena in continuos learning
3. Distributed representation of patterns and memory retrieval: catastrophic interference
4. Investigate inductive biases and multi-task learning
# TODO
1. Why we need neurogenesis in SORN?
a. Network size vs Number of active synapses vs Network performance in sequence learning
b. Investigate and identify document how and why catastrophic forgetting occuring.
b. Observe dynamic memory and adaptation strategies. https://www.nature.com/articles/s41467-021-25858-z
2. Add new neurons at random time intervals and repeat step 1.
3.