Free energy landscape of Dense Associative Memory
The paper gives a large-deviations route to the free energy functional for dense associative memories.
Sumedha and Abhishek Singh apply the method to recover known Hopfield-model results, then extend it to dense memories with polynomial interactions and Log-Sum-Exponential activation. For finite pattern counts, they derive temperature-dependent free energy functionals. In the extensive limit, they compute the disorder-averaged ground-state energy and identify an exact full-retrieval threshold for the LSE model. ArXiv · AI/CL/LG's note
Sumedha and Abhishek Singh apply the method to recover known Hopfield-model results, then extend it to dense memories with polynomial interactions and Log-Sum-Exponential activation. For finite pattern counts, they derive temperature-dependent free energy functionals. In the extensive limit, they compute the disorder-averaged ground-state energy and identify an exact full-retrieval threshold for the LSE model. ArXiv · AI/CL/LG's note
score 4