Invariant Representations with Stochastically Quantized Neural Networks
Authors: Mattia Cerrato, Marius Köppel, Roberto Esposito, Stefan Kramer
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experimentation proves that quantized, low-precision models are able to obtain group-invariant representations in a fair classification setting and avoid training set biases in image data. Our experimental setting focuses on analyzing the performance of the methodology presented in this paper in fair classification and invariant representation learning settings. Our experimentation aims to answer the following questions: Q1. Is the present methodology able to learn fair models which avoid discrimination? A1. Yes. We analyze our models accuracy and fairness by measuring the area under curve (AUC) and their disparate impact / disparate mistreatment. |
| Researcher Affiliation | Academia | Mattia Cerrato 1*, Marius Köppel 2*, Roberto Esposito 3, Stefan Kramer 1 1 Institute of Computer Science, Johannes Gutenberg-Universität Mainz, Germany 2 Institute for Nuclear Physics, Johannes Gutenberg-Universität Mainz, Germany 3 Computer Science Department, Università degli Studi di Torino, Italy |
| Pseudocode | No | The paper does not contain any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not include an unambiguous statement or a direct link to a source-code repository for the methodology described. It mentions "Weights & Biases platform for an implementation of Bayesian optimization and overall experiment tracking (Biewald 2020)", but this is a tool used, not their released code. |
| Open Datasets | Yes | COMPAS. This dataset was released by Pro Publica (Angwin et al. 2016)... Adult. This dataset is part of the UCI repository (Dua and Graff 2017)... Biased-MNIST. This is an image dataset based on the wellknown MNIST Handwritten Digits database in which the background has been modified so to display a color bias (Bahng et al. 2020). |
| Dataset Splits | Yes | We split all datasets into 3 internal and 3 external folds. On the 3 internal folds, we employ a Bayesian optimization technique to find the best hyperparameters for our model. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory specifications) used for running the experiments. It only generally refers to training models without mentioning the computational infrastructure. |
| Software Dependencies | No | The paper mentions using "Weights & Biases platform" but does not specify version numbers for this or any other software dependencies like programming languages, libraries, or frameworks used for implementation. |
| Experiment Setup | Yes | We set the maximum number of iterations to 200. The influence of the two loss functions is controlled via the parameter γ. We also note that the variational approximationbased model (VFAE) struggles to come to an accurate result, whereas it mostly takes fair decisions. The correlation between γ, which controls the strength of the Maximum Mean Discrepancy regularization in this model, and AUC/1-GPA is also quite low (0.173 and 0.168 respectively). |