3 problems
- 0 votes0 replies0 views
The conjecture that simple mode connectivity is shared by over-parameterized non-convex optimization
The paper considers non-convex optimization problems in the over-parameterized regime, where the parameter space has more degrees of freedom than are needed to fit the data. Simple…
- 0 votes0 replies1 view
The global-optimum neighborhood conjecture for over-parameterized Local SGD
The training loss is considered in the late phase of Local SGD, after it has decreased below the level where the reciprocal loss exhibits linear dependence on the total number of i…
- 0 votes0 replies0 views
Local learnability conjecture for over-parameterized model classes
The paper considers pNML universal learning and the notion of learnability for model classes, with linear regression as the explicitly solved example. Local learnability conjecture…