Learning a Single Neuron with Adversarial Label Noise via Gradient Descent
This addresses a fundamental learning theory problem for researchers, providing efficient algorithms with provable guarantees in noisy settings, though it is incremental as it builds on prior work with specific improvements.
The paper tackles the problem of learning a single neuron with adversarial label noise using gradient descent, achieving efficient constant-factor approximations for distributions like isotropic log-concave ones, with sample complexities of Õ(d/ε) for logistic activation and Õ(d polylog(1/ε)) for ReLU, which are tight or improve prior bounds.
We study the fundamental problem of learning a single neuron, i.e., a function of the form $\mathbf{x}\mapstoσ(\mathbf{w}\cdot\mathbf{x})$ for monotone activations $σ:\mathbb{R}\mapsto\mathbb{R}$, with respect to the $L_2^2$-loss in the presence of adversarial label noise. Specifically, we are given labeled examples from a distribution $D$ on $(\mathbf{x}, y)\in\mathbb{R}^d \times \mathbb{R}$ such that there exists $\mathbf{w}^\ast\in\mathbb{R}^d$ achieving $F(\mathbf{w}^\ast)=ε$, where $F(\mathbf{w})=\mathbf{E}_{(\mathbf{x},y)\sim D}[(σ(\mathbf{w}\cdot \mathbf{x})-y)^2]$. The goal of the learner is to output a hypothesis vector $\mathbf{w}$ such that $F(\mathbb{w})=C\, ε$ with high probability, where $C>1$ is a universal constant. As our main contribution, we give efficient constant-factor approximate learners for a broad class of distributions (including log-concave distributions) and activation functions. Concretely, for the class of isotropic log-concave distributions, we obtain the following important corollaries: For the logistic activation, we obtain the first polynomial-time constant factor approximation (even under the Gaussian distribution). Our algorithm has sample complexity $\widetilde{O}(d/ε)$, which is tight within polylogarithmic factors. For the ReLU activation, we give an efficient algorithm with sample complexity $\tilde{O}(d\, \polylog(1/ε))$. Prior to our work, the best known constant-factor approximate learner had sample complexity $\tildeΩ(d/ε)$. In both of these settings, our algorithms are simple, performing gradient-descent on the (regularized) $L_2^2$-loss. The correctness of our algorithms relies on novel structural results that we establish, showing that (essentially all) stationary points of the underlying non-convex loss are approximately optimal.