Dealing With Misspecification In Fixed-Confidence Linear Top-m Identification

Authors: Clémence Réda, Andrea Tirinzoni, Rémy Degenne

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we evaluate our algorithm on both synthetic and real-world data, showing competitive performance with respect to existing baselines.
Researcher Affiliation Academia Clémence Réda Université de Paris, Neuro Diderot, Inserm, F-75019 Paris, France clemence.reda@inria.fr Andrea Tirinzoni Univ. Lille, Inria, CNRS, Centrale Lille, UMR 9189 CRISt AL, F-59000 Lille, France andrea.tirinzoni@inria.fr Rémy Degenne Univ. Lille, Inria, CNRS, Centrale Lille, UMR 9189 CRISt AL, F-59000 Lille, France remy.degenne@inria.fr
Pseudocode Yes Algorithm 1 MISLID Require: Set of models M, online learner L, stopping thresholds {βt,δ}t 1
Open Source Code Yes All the code and scripts are available at https://github.com/clreda/misspecified-top-m.
Open Datasets Yes We use the drug repurposing problem for epilepsy proposed by [35] to investigate the practicality of our method. and a linear representation is extracted for an instance of online recommendation of music artists to users (Last.fm dataset [6]).
Dataset Splits No The paper does not provide information about specific training, validation, and test dataset splits. It describes conducting repetitions of experiments rather than splitting a static dataset.
Hardware Specification No The paper mentions 'computational resources' in Appendix G but does not provide specific hardware details such as GPU or CPU models, or memory specifications.
Software Dependencies No The paper does not specify any software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9).
Experiment Setup Yes In all experiments, we consider δ = 5%. For each experiment, we report the number of arms (K), the dimension of features (d), the size of the answer (m), the misspecification (ε) and the gap between the mth and (m + 1)th best arms. [...] we use a heuristic value for the stopping rule βt,δ := ln((1 + ln(t + 1))/δ) unless otherwise specified.