A Unifying Perspective on Multi-Calibration: Game Dynamics for Multi-Objective Learning

Authors: Nika Haghtalab, Michael Jordan, Eric Zhao

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we study the empirical performance of multicalibration algorithms on the UCI Adult Income dataset [26], a real-world dataset for predicting individuals incomes based on the US Census.
Researcher Affiliation Academia Nika Haghtalab, Michael I. Jordan, and Eric Zhao University of California, Berkeley {nika,jordan,eric.zh}@berkeley.edu
Pseudocode Yes Algorithm 1 Non-Deterministic Multicalibration Algorithm (Theorem 4.1)
Open Source Code Yes The source code for these experiments is included in the repository https://github.com/ericzhao28/multicalibration.
Open Datasets Yes We conduct three sets of experiments to evaluate different batch multicalibration algorithms. The three sets of experiments we conduct correspond to three datasets: the UCI Adult Income dataset [26], a real-world dataset for predicting individuals incomes based on the US Census, the UCI Bank Marketing dataset [32], a dataset for predicting whether an individual will subscribe to a bank s term deposit, and the Dry Bean Dataset [27], a dataset for predicting a dry bean s variety.
Dataset Splits No The paper mentions 'random 80-20 train/test splits' but does not explicitly state a separate validation split or its size/percentage.
Hardware Specification Yes All experiments were performed on a 2021 Mac Book Pro, with a M1 Pro chip.
Software Dependencies No The paper does not explicitly provide specific software dependencies with version numbers (e.g., Python, PyTorch, or other libraries).
Experiment Setup Yes Learning rate decay is tuned on the training set by sweeping over [0.8, 0.85, 0.9, 0.95] for the learner and [0.9, 0.95, 0.98, 0.99] for the adversary.