6 problems

Min-norm interpolators and estimators are considered in an overparameterized regime, where the number of parameters may exceed the number of observations and interpolation is possi…

Projected $\ell_1$-norm concentration conjecture. There is a constant $c$ such that

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

Consider a statistical learning model with a non-convex objective, parameter-space dimension equal to the number of optimization variables, and sample size viewed as the number of…

Deep learning refers here to highly overparameterized models trained by gradient methods, with overparametrization meaning that the model has more effective degrees of freedom than…

Large overparameterized neural networks are models with substantially more parameters than are needed to fit the training data. Local search methods include stochastic gradient des…