Universality laws for Gaussian mixtures in generalized linear models

Authors: Yatin Dandi, Ludovic Stephan, Florent Krzakala, Bruno Loureiro, Lenka Zdeborová

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

Reproducibility Variable Result LLM Response
Research Type Experimental Figure 2: Illustration of the universality scenario described in Fig.1. Logistic (left) & ridge (right) regression test (up) and training (bottom) errors are shown versus the sample complexity α = n/d for an odd vs. even binary classification task on two data models: Blue dots data generated from a conditional GAN [41] trained on the fashion-MNIST dataset [45] and pre-processed with a random features map x 7 tanh(Wx) with Gaussian weights W R1176 784; Red dots are the 10clusters Gaussian mixture model with means and covariances matching each fashion-MNIST cluster conditioned on labels (ℓ2 regularization is λ = 10 4). Details on the simulations are discussed in Appendix D.
Researcher Affiliation Academia Yatin Dandi, Ludovic Stephan, Florent Krzakala Idephics, EPFL, Switzerland Bruno Loureiro DI, Ecole Normale Superieure, Paris, France Lenka Zdeborova SPOC, EPFL, Switzerland
Pseudocode No The paper does not contain any clearly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code No The paper does not contain any explicit statements about making its source code available or links to a code repository for its methodology.
Open Datasets Yes Blue dots data generated from a conditional GAN [41] trained on the fashion-MNIST dataset [45] and pre-processed with a random features map x 7 tanh(Wx) with Gaussian weights W R1176 784; Red dots are the 10clusters Gaussian mixture model with means and covariances matching each fashion-MNIST cluster conditioned on labels (ℓ2 regularization is λ = 10 4). Details on the simulations are discussed in Appendix D.
Dataset Splits No The paper mentions 'training data' but does not specify the explicit train/validation/test split percentages, sample counts, or refer to any predefined splits with citations for reproducibility.
Hardware Specification No The paper does not specify any hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper does not provide specific version numbers for any software components or libraries used in the experiments.
Experiment Setup Yes Figure 2: Illustration of the universality scenario described in Fig.1. Logistic (left) & ridge (right) regression test (up) and training (bottom) errors are shown versus the sample complexity α = n/d for an odd vs. even binary classification task on two data models: Blue dots data generated from a conditional GAN [41] trained on the fashion-MNIST dataset [45] and pre-processed with a random features map x 7 tanh(Wx) with Gaussian weights W R1176 784; Red dots are the 10clusters Gaussian mixture model with means and covariances matching each fashion-MNIST cluster conditioned on labels (ℓ2 regularization is λ = 10 4). Details on the simulations are discussed in Appendix D.