Unsupervised Risk Estimation Using Only Conditional Independence Structure

Authors: Jacob Steinhardt, Percy S. Liang

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To better understand the behavior of our algorithms, we perform experiments on a version of the MNIST data set that is modified to ensure that the 3-view assumption holds. We use Algorithm 1 to perform unsupervised risk estimation for a model trained on a = 0, testing on various values of a [0, 10].
Researcher Affiliation Academia Jacob Steinhardt Stanford University jsteinhardt@cs.stanford.edu Percy Liang Stanford University pliang@cs.stanford.edu
Pseudocode Yes Algorithm 1 Algorithm for estimating R(θ) from unlabeled data.
Open Source Code No No explicit statement about code availability or repository link found in the paper.
Open Datasets Yes To better understand the behavior of our algorithms, we perform experiments on a version of the MNIST data set that is modified to ensure that the 3-view assumption holds.
Dataset Splits No We trained the model with Ada Grad (Duchi et al., 2010) on 10,000 training examples, and used 10,000 test examples to estimate the risk. The term “validation error” is in the legend of Figure 4a, but no specific validation split or size is given.
Hardware Specification No The paper does not specify any hardware details like GPU/CPU models or specific computing infrastructure used for the experiments.
Software Dependencies No We trained the model with Ada Grad (Duchi et al., 2010) and locally minimize a weighted ℓ2-norm of the moment errors in (4) using L-BFGS. No version numbers for software are provided.
Experiment Setup Yes setting ρ = 10 in (8).