We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Learning a Single Neuron with Adversarial Label Noise via Gradient Descent

Abstract: We study the fundamental problem of learning a single neuron, i.e., a function of the form $\mathbf{x}\mapsto\sigma(\mathbf{w}\cdot\mathbf{x})$ for monotone activations $\sigma:\mathbb{R}\mapsto\mathbb{R}$, 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 $F(\mathbf{w}^\ast)=\epsilon$, where $F(\mathbf{w})=\mathbf{E}_{(\mathbf{x},y)\sim D}[(\sigma(\mathbf{w}\cdot \mathbf{x})-y)^2]$. The goal of the learner is to output a hypothesis vector $\mathbf{w}$ such that $F(\mathbb{w})=C\, \epsilon$ 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 $\widetilde{O}(d/\epsilon)$, which is tight within polylogarithmic factors.
For the ReLU activation, we give an efficient algorithm with sample complexity $\tilde{O}(d\, \polylog(1/\epsilon))$. Prior to our work, the best known constant-factor approximate learner had sample complexity $\tilde{\Omega}(d/\epsilon)$.
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.
Subjects: Machine Learning (cs.LG); Data Structures and Algorithms (cs.DS); Statistics Theory (math.ST); Machine Learning (stat.ML)
Cite as: arXiv:2206.08918 [cs.LG]
  (or arXiv:2206.08918v1 [cs.LG] for this version)

Submission history

From: Nikos Zarifis [view email]
[v1] Fri, 17 Jun 2022 17:55:43 GMT (538kb,D)

Link back to: arXiv, form interface, contact.