Provable Guarantees for Neural Networks via Gradient Feature Learning

Authors: Zhenmei Shi, Junyi Wei, Yingyu Liang

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical Our paper is purely theoretical in nature, and thus we do not anticipate an immediate negative ethical impact.
Researcher Affiliation Academia Zhenmei Shi , Junyi Wei , Yingyu Liang University of Wisconsin, Madison zhmeishi@cs.wisc.edu,jwei53@wisc.edu,yliang@cs.wisc.edu
Pseudocode Yes Algorithm 1 Network Training via Gradient Descent
Open Source Code No The paper is theoretical and does not mention providing open-source code for its framework or any associated implementation.
Open Datasets No The paper analyzes theoretical data distributions (e.g., mixtures of Gaussians, parity functions) rather than using named publicly available datasets for empirical training. No specific access information to a public dataset is provided.
Dataset Splits No The paper is theoretical and works with mathematical models of data distributions rather than empirical datasets, so there are no specified training, validation, or test splits in the experimental sense.
Hardware Specification No The paper is purely theoretical and does not conduct empirical experiments, so it does not mention any hardware specifications.
Software Dependencies No The paper is purely theoretical and does not conduct empirical experiments, so it does not list any software dependencies with version numbers.
Experiment Setup No The paper discusses "proper hyper-parameter values" within its theoretical framework (e.g., in Theorem 3.12), and provides conditions for these values, but it does not specify concrete numerical values for an experimental setup.