Logo

Harvard ML theory talk - Surya Ganguli - Shared screen with speaker view
Cunlu Zhou
41:47
Is there a scaling complexity of measurement for MAP to be optimal?
Cong Peng
42:48
How does OLS look like in the graph?
Kyle Poe
01:12:37
To what extent can we extrapolate these analytical linear network dynamics to the local behavior of nonlinear networks? If so, can we develop an "atlas" of linear regimes to generalize the global behavior of nonlinear networks?
Venkatesh Medabalimi
01:13:13
feels something like loss landscape having troughs inside troughs inside troughs
Venkatesh Medabalimi
01:14:24
mines just more of a comment about multi scale structure..
Kenneth Chiu
01:34:46
Would there some benefit to more than two types of cells?
Jakub
01:52:57
Isn’t this equivalent to standard Boltzmann machines? I mean the probabilistic difference to hopfield
Kenneth Chiu
01:55:20
Apologies if you already stated this, but how do you add or program-in a "memory" to the BEC?
Hacene Boukari
01:57:29
Is it a 3-D Ising Model??
Kenneth Chiu
01:58:38
To clarify, I mean a specific memory, i.e., add a new minimum.