Complexity in neural networks: the good with the bad

In the first part of the seminar I will describe neural networks from a statistical-mechanics perspective. Focusing on networks trained for retrieval tasks, I will discuss their complexity and show how it is related to network’s performance: on the one hand, complexity is a desirable feature as it (potentially) increases the amount of information stored in the network, on the other hand, beyond a critical threshold, it dramatically impairs retrieval. 

Finally, I will introduce an algorithm for information storage, inspired by neurophysiological mechanisms occurring during mammal’s sleep, which yields to an “optimal complexity”.  

Zoom details:

https://us02web.zoom.us/j/87951525991?pwd=U2Q2RWJDSDlXeXpsK0RBVkRSY21FZz09

Meeting ID: 879 5152 5991

Passcode: 537385