Contrastive Moments: Unsupervised Halfspace Learning in Polynomial Time

Authors: Xinyuan Cao, Santosh Vempala

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

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experiments While our primary goal is to establish polynomial bounds on the sample and time complexity, our algorithms are natural and easy to implement. We study the efficiency and performance of Algorithm 1 on data drawn from affine product distributions with margin. Here we consider three special cases of logconcave distribution: Gaussian, uniform in an interval and exponential. We include four experiments. In all results, we measure the performance of the algorithm using the sin of the angle between the true normal vector u and the predicted vector ˆu, i.e., sin θ(u, ˆu), which bounds the TV distance between the underlying distribution and the predicted one after isotropic transformation.
Researcher Affiliation Academia Xinyuan Cao Georgia Tech xcao78@gatech.edu Santosh S. Vempala Georgia Tech vempala@gatech.edu
Pseudocode Yes Algorithm 1 Unsupervised Halfspace Learning with Contrastive Moments
Open Source Code No The paper does not provide an explicit statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets No The paper describes generating data from 'Gaussian, uniform in an interval and exponential' distributions for its experiments, but it does not refer to or provide access information for any pre-existing publicly available datasets.
Dataset Splits No The paper does not provide specific details on training, validation, or test dataset splits. It mentions using 'iid samples' and 'sample size N' but no explicit partitioning strategy.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, processor types, or memory used for running the experiments.
Software Dependencies No The paper does not specify any software dependencies, such as libraries or frameworks, with their version numbers.
Experiment Setup Yes For the parameters, we choose α1 = α3 = 0.1, α2 = 0.2.