1 problem

0 votes0 replies1 view
Conjecture that the tail-decay bias bound does not require a logarithmic factor
Let $s$ and $K$ be the parameters appearing in the tail-decay upper bound, and let the bound involve the factor $\log(s+K)$ . Tail-decay bias conjecture. Theorem is not tight with t…
Conjectureopenoverparameterized-learning risk-bounds stochastic-gradient-descentarXiv posted 2 days ago

Conjecture that the tail-decay bias bound does not require a logarithmic factor