Log-Likelihood Ratio Minimizing Flows: Towards Robust and Quantifiable Neural Distribution Alignment

Authors: Ben Usman, Avneesh Sud, Nick Dufour, Kate Saenko

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we present experiments that verify that minimizing the proposed LRMF objective (3) with Gaussian, Real NVP, and FFJORD density estimators does indeed result in dataset alignment.
Researcher Affiliation Collaboration Ben Usman 1,2 usmn@bu.edu Avneesh Sud 2 asud@google.com Nick Dufour 2 ndufour@google.com Kate Saenko 1,3 saenko@bu.edu Boston University 1 Google AI 2 MIT-IBM Watson AI Lab 3
Pseudocode No The paper presents mathematical definitions and derivations but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes We provide Jupyter notebooks with code in JAX [6] and Tensor Flow Probability (TFP) [7].
Open Datasets Yes We also trained a Real NVP LRMF to map latent codes of USPS digits to latent codes of MNIST.
Dataset Splits No This enables automatic model validation and hyperparameter tuning on the held-out set.
Hardware Specification No The paper does not specify any particular hardware (e.g., GPU models, CPU types, or cloud instance specifications) used for running the experiments.
Software Dependencies No We provide Jupyter notebooks with code in JAX [6] and Tensor Flow Probability (TFP) [7].
Experiment Setup No We used original hyperparameters and network architectures from Real NVP [8] and FFJORD [11], the exact values are given in the supplementary.