8 problems

In FUVAL, let $alpha_i$ denote the learned loss-value parameter associated with the $i$th sampled term, and consider single-element sampling, in which one term is sampled at each i…

Let $n$ be the sample size, let $\cF$ be a function class, and let $\mathbb P$ be a probability distribution. Under the paper's random-design setting, write $\widehat f_n$ for empi…

Let $n$ be the sample size, let $\cF$ be a function class, and let $\mathbb P$ be a probability distribution. Write $\widehat f_n$ for the empirical risk minimizer,…

Catalyst performance conjecture. Catalyst approaches would perform better for other function classes and parameter regimes.

Overparametrization hypothesis. Highly overparametrized models should be easy to optimize despite the empirical risk being generally highly non-convex. This is an informal hypothes…

The matrix $\mathcal{B}$ is a positive definite matrix indexed by multi-indices in $\mathbb{Z}_+^d$ with total degree at most $l$, and the paper derives a lower bound on…

Consider the distributed empirical risk minimization (ERM) setting, in which a fixed finite collection of objectives is optimized, and let FedAc denote the federated accelerated st…

Theorem concerns conditions ensuring that the relevant system of equations has a unique and bounded solution. The paper does not specify the technical conditions in the conjectural…