On the Fairness of Disentangled Representations
Authors: Francesco Locatello, Gabriele Abbati, Thomas Rainforth, Stefan Bauer, Bernhard Schölkopf, Olivier Bachem
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Analyzing the representations of more than 12 600 trained state-ofthe-art disentangled models, we observe that several disentanglement scores are consistently correlated with increased fairness, suggesting that disentanglement may be a useful property to encourage fairness when sensitive variables are not observed. |
| Researcher Affiliation | Collaboration | Francesco Locatello2,5, Gabriele Abbati3, Tom Rainforth4, Stefan Bauer5, Bernhard Schölkopf5, and Olivier Bachem1 1Google Research, Brain Team 2Dept. of Computer Science, ETH Zurich 3Dept. of Engineering Science, University of Oxford 4Dept. of Statistics, University of Oxford 5Max-Planck Institute for Intelligent Systems |
| Pseudocode | No | The paper describes various models and methods but does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | No | The paper refers to 'pre-trained models of [61]' and mentions that 'Details on architecture, hyperparameter, implementation of the methods can be found in Appendices B, C, G, and H of [61]', but it does not provide an explicit statement or link for the open-source code of the methodology described in this paper. |
| Open Datasets | Yes | Their analysis spans seven datasets: in four of them (d Sprites [36], Cars3D [78], Small NORB [59] and Shapes3D [49]), a deterministic function of the factors of variation is incorporated into the mixing process; they further introduce three additional variants of d Sprites, Noisy-d Sprites, Color-d Sprites, and Scream-d Sprites. |
| Dataset Splits | No | The paper mentions that the gradient boosting classifier was 'trained on 10 000 labeled examples' and refers to 'crossvalidated logistic regression', but it does not provide explicit percentages, counts, or a detailed methodology for dataset splits (training, validation, and test) needed to reproduce the data partitioning. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used to run the experiments. |
| Software Dependencies | No | The paper mentions 'Scikit-learn [72]' but does not provide specific version numbers for it or any other software dependencies, which are necessary for full reproducibility. |
| Experiment Setup | No | The paper states that the '12 600 pre-trained models of [61]... cover a large number of hyperparameters and random seeds' and refers to 'Appendices B, C, G, and H of [61]' for 'Details on architecture, hyperparameter, implementation of the methods'. However, it does not explicitly provide the specific hyperparameter values or training configurations for its own experiments or the gradient boosting classifier used. |