Multitask Learning with No Regret: from Improved Confidence Bounds to Active Learning

Authors: Pier Giuseppe Sessa, Pierre Laforgue, Nicolò Cesa-Bianchi, Andreas Krause

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

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we empirically validate our bounds and algorithms on synthetic and real-world (drug discovery) data. ... 5 Experiments The goal of our experiments is to evaluate the effectiveness of the studied MT regression, and in particular of the improved confidence intervals obtained in Section 2, both in online learning and active learning setups. We utilize the following synthetic and real-world data.
Researcher Affiliation Academia Pier Giuseppe Sessa ETH Zürich piergiuseppe.sessa@inf.ethz.ch Pierre Laforgue Università degli Studi di Milano pierre.laforgue@unimi.it Nicolò Cesa-Bianchi Università degli Studi di Milano Politecnico di Milano nicolo.cesa-bianchi@unimi.it Andreas Krause ETH Zürich krausea@ethz.ch
Pseudocode Yes Algorithm 1 MT-UCB ... Algorithm 2 MT-AL
Open Source Code Yes 2code available at: https://github.com/sessap/multitask-noregret.
Open Datasets Yes Drug discovery MHC-I data [27]: The goal is to discover the peptides with maximal binding affinity to each Major Histocompatibility Complex class-I (MHC-I) allele. The dataset from [27] contains the standardized binding affinities (IC50 values) of different peptide candidates to the MHC-I alleles (tasks).
Dataset Splits No The paper mentions using synthetic data and MHC-I data from [27] but does not specify any training, validation, or test dataset splits, percentages, or methodology for partitioning the data.
Hardware Specification No The paper does not provide specific details about the hardware used for the experiments (e.g., CPU, GPU models, memory).
Software Dependencies No The paper does not specify any software dependencies with version numbers used for the experiments (e.g., programming languages, libraries, frameworks).
Experiment Setup Yes For the synthetic data, we report results for d = 4, N = 5, δ = 0.4... for MHC-I data we use = 0.3... E = {.1, .2, . . . , 1} instead of knowing the true .