AI Paper Atlas

Research map / Learning theory and algorithms / Sample complexity

Learning Distributions Generated by One-Layer ReLU Networks

Shanshan Wu, Alexandros G Dimakis, Sujay Sanghavi

NeurIPS 2019 · Learning theory and algorithms › Sample complexity · approach: Statistical analysis of GAN estimators · 7 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Assume the bias vector is non-negativeSingle-layer ReLU learning algorithm · Rectified Gaussian parameter estimationsample complexity: \widetilde{O}(1/\epsilon^2) versus at least exponential in \norm{b}^2_{\infty}Improved
Estimate row norms and pairwise angles separatelySingle-layer ReLU learning algorithm · Rectified Gaussian parameter estimationsample complexity: \widetilde{O}(1/\epsilon^2)Improved
Increase the number of samplesSingle-layer ReLU learning algorithm · Synthetic rectified Gaussian data (d=k=5, κ=1)parameter estimation error and KL divergenceImproved
Increase the dimensionSingle-layer ReLU learning algorithm · Synthetic rectified Gaussian data (n=5×10^5, κ=1)KL divergenceHurt
Increase the condition numberSingle-layer ReLU learning algorithm · Synthetic rectified Gaussian data (n=5×10^5, d=5)KL divergenceHurt
Increase the number of samplesTwo-layer ReLU learning algorithm · Two-layer ReLU generative model (d=10, p=k=5)probability of success recovering A*: 0.30 -> 0.78 -> 0.99 (n=50, 100, 150)Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy