Local minima of the empirical risk in high dimension: General theorems and convex examples
This addresses the challenge of optimization landscape analysis in high-dimensional statistics and machine learning, with incremental contributions to rigorous asymptotics for convex losses.
The paper tackles the problem of understanding local minima in high-dimensional empirical risk minimization for models like neural networks, deriving a bound on the expected number of local minima and providing sharp asymptotics for estimation and prediction error in convex cases.
We consider a general model for high-dimensional empirical risk minimization whereby the data $\mathbf{x}_i$ are $d$-dimensional isotropic Gaussian vectors, the model is parametrized by $\mathbfΘ\in\mathbb{R}^{d\times k}$, and the loss depends on the data via the projection $\mathbfΘ^\mathsf{T}\mathbf{x}_i$. This setting covers as special cases classical statistics methods (e.g. multinomial regression and other generalized linear models), but also two-layer fully connected neural networks with $k$ hidden neurons. We use the Kac-Rice formula from Gaussian process theory to derive a bound on the expected number of local minima of this empirical risk, under the proportional asymptotics in which $n,d\to\infty$, with $n\asymp d$. Via Markov's inequality, this bound allows to determine the positions of these minimizers (with exponential deviation bounds) and hence derive sharp asymptotics on the estimation and prediction error. In this paper, we apply our characterization to convex losses, where high-dimensional asymptotics were not (in general) rigorously established for $k\ge 2$. We show that our approach is tight and allows to prove previously conjectured results. In addition, we characterize the spectrum of the Hessian at the minimizer. A companion paper applies our general result to non-convex examples.