Robust Model Selection and Nearly-Proper Learning for GMMs

Authors: Allen Liu, Jerry Li, Ankur Moitra

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical In this work, we study the problem of robust model selection for univariate Gaussian mixture models (GMMs). Given poly(k/ϵ) samples from a distribution that is ϵ-close in TV distance to a GMM with k components, we can construct a GMM with e O(k) components that approximates the distribution to within e O(ϵ) in poly(k/ϵ) time. Thus we are able to approximately determine the minimum number of components needed to fit the distribution within a logarithmic factor. Prior to our work, the only known algorithms for learning arbitrary univariate GMMs either output significantly more than k components (e.g. k/ϵ2 components for kernel density estimates) or run in time exponential in k.
Researcher Affiliation Collaboration Allen Liu MIT Cambridge, MA 02139 cliu568@mit.edu Jerry Li Microsoft Research Redmond, WA 98052 jerrl@microsoft.com Ankur Moitra MIT Cambridge, MA 02139 moitra@mit.edu
Pseudocode Yes We show how to eliminate the need for pdf access below and present our full algorithm in complete detail in Section C. ... In Section C, we provide Algorithm 1: Learning GMMs, which details the steps of our approach.
Open Source Code No The paper is theoretical and focuses on algorithms and proofs; it does not mention the release of any source code.
Open Datasets No The paper is theoretical and does not use or provide access to specific datasets for training or evaluation. The checklist states 'N/A' for 'Did you include the code, data, and instructions needed to reproduce the main experimental results'.
Dataset Splits No The paper is theoretical and does not report on experiments, therefore no dataset splits for training, validation, or testing are mentioned.
Hardware Specification No The paper is theoretical and does not report on experiments, thus no hardware specifications are mentioned.
Software Dependencies No The paper is theoretical and does not mention specific software dependencies with version numbers for implementation or experimental setup.
Experiment Setup No The paper is theoretical and does not report on experiments, thus no experimental setup details such as hyperparameters or training settings are provided.