Learning through atypical "phase transitions" in overparameterized neural networks

Abstract

Current deep neural networks are highly overparameterized (up to billions of connection weights) and nonlinear. Yet they can fit data almost perfectly through variants of gradient descent algorithms and achieve unexpected levels of prediction accuracy without overfitting. These are formidable results that defy predictions of statistical learning and pose conceptual challenges for non-convex optimization. In this paper, we use methods from statistical physics of disordered systems to analytically study the computational fallout of overparameterization in non-convex binary neural network models, trained on data generated from a structurally simpler but "hidden" network. As the number of connection weights increases, we follow the changes of the geometrical structure of different minima of the error loss function and relate them to learning and generalization performance. A first transition happens at the so-called interpolation point, when solutions begin to exist (perfect fitting becomes possible). This transition reflects the properties of typical solutions, which however are in sharp minima and hard to sample. After a gap, a second transition occurs, with the discontinuous appearance of a different kind of "atypical" structures: wide regions of the weight space that are particularly solution-dense and have good generalization properties. The two kinds of solutions coexist, with the typical ones being exponentially more numerous, but empirically we find that efficient algorithms sample the atypical, rare ones. This suggests that the atypical phase transition is the relevant one for learning. The results of numerical tests with realistic networks on observables suggested by the theory are consistent with this scenario.

28 pages, 14 figures

References in corpus (6)

• Many body localization and thermalization in quantum statistical mechanics
• Efficient supervised learning in networks with binary synapses
• Origin of the computational hardness for learning with binary synapses
• Unveiling the structure of wide flat minima in neural networks
• Generalization learning in a perceptron with binary synapses
• Wide flat minima and optimal generalization in classifying high-dimensional Gaussian mixtures

Cited by in corpus (9)

• A statistical mechanics framework for Bayesian deep neural networks beyond the infinite-width limit
• Phase transitions in the mini-batch size for sparse and dense two-layer neural networks
• Exact full-RSB SAT/UNSAT transition in infinitely wide two-layer neural networks
• Eight challenges in developing theory of intelligence
• Daydreaming Hopfield Networks and their surprising effectiveness on correlated data
• Inversion dynamics of class manifolds in deep learning reveals tradeoffs underlying generalisation
• Stochastic Gradient Descent-like relaxation is equivalent to Metropolis dynamics in discrete optimization and inference problems
• Supervised Learning Guarantee for Quantum AdaBoost
• Pseudo-likelihood produces associative memories able to generalize, even for asymmetric couplings