3 problems

0 votes0 replies1 view
Necessity of technical assumptions for edge-of-stability optimization
The paper studies gradient descent on the empirical risk of a non-linear predictor, using additional technical assumptions to address the challenges caused by non-convexity. Necess…
Conjectureopengradient-descent neural-networks non-convex-optimizationarXiv posted 2 days ago
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…
Conjectureopenmode-connectivity non-convex-optimization over-parameterizationarXiv posted 2 days ago
0 votes0 replies0 views
Conjecture that local-search convergence requires weaker conditions than global optimality
Local-search convergence conjecture. The conditions required for local-search algorithms to converge to the true solutions may be considerably less stringent than the conditions re…
Conjectureopenlow-rank-matrix-recovery non-convex-optimizationarXiv posted 2 days ago

Necessity of technical assumptions for edge-of-stability optimization