Deep learning: a statistical viewpoint

Abstract

The remarkable practical success of deep learning has revealed some major surprises from a theoretical perspective. In particular, simple gradient methods easily find near-optimal solutions to non-convex optimization problems, and despite giving a near-perfect fit to training data without any explicit effort to control model complexity, these methods exhibit excellent predictive accuracy. We conjecture that specific principles underlie these phenomena: that overparametrization allows gradient methods to find interpolating solutions, that these methods implicitly impose regularization, and that overparametrization leads to benign overfitting, that is, accurate predictions despite overfitting training data. In this article, we survey recent progress in statistical learning theory that provides examples illustrating these principles in simpler settings. We first review classical uniform convergence results and why they fall short of explaining aspects of the behaviour of deep learning methods. We give examples of implicit regularization in simple settings, where gradient methods lead to minimal norm functions that perfectly fit the training data. Then we review prediction methods that exhibit benign overfitting, focusing on regression problems with quadratic loss. For these methods, we can decompose the prediction rule into a simple component that is useful for prediction and a spiky component that is useful for overfitting but, in a favourable setting, does not harm prediction accuracy. We focus specifically on the linear regime for neural networks, where the network can be approximated by a linear model. In this regime, we demonstrate the success of gradient flow, and we consider benign overfitting with two-layer networks, giving an exact asymptotic analysis that precisely demonstrates the impact of overparametrization. We conclude by highlighting the key challenges that arise in extending these insights to realistic deep learning settings.

AI based pre-screening of large bowel cancer via weakly supervised learning of colorectal biopsy histology images

“…Despite the tile-level false positives, the quantitative measures have shown excellent predictive accuracy, robust performance and generalization on both cohorts. This may be associated with interpolation and generalization capabilities of overparameterized deep learning systems leading to benign overfitting, as demonstrated by Bartlett et al in their latest findings (23).…”

“…This may be associated with interpolation and generalization capabilities of overparameterized deep learning systems leading to benign overfitting , as demonstrated by Bartlett et al . in their latest findings (23).…”

Security Evaluation of Financial and Insurance and Ruin Probability Analysis Integrating Deep Learning Models

“…The activation of one neuron can generate a data analysis result, and many neurons are connected to form a complete NN model and output the analysis result of the complete data. The NN technology was first produced in the 1950s and 1960s, and the simplest NN technology is the perceptron, which is essentially a feedforward NN structure, and this NN structure is also relatively common [ 9 ]. The perceptron consists of an input layer, an output layer, and a hidden layer.…”

Nearest Class-Center Simplification through Intermediate Layers

“…Several recent works have investigated the nature of modern Deep Neural Networks (DNNs) past the point of zero training error (Belkin, 2021;Nakkiran et al, 2020;Bartlett et al, 2021;Power et al, 2022). The stage at which the training error reaches zero is called the Interpolation Threshold (IT), since at this point, the learned network function interpolates between training samples.…”