Adaptive Selective Sampling for Online Prediction with Experts

Authors: Rui Castro, Fredrik Hellström, Tim van Erven

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

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we present numerical experiments empirically showing that the normalized regret of the label-efficient forecaster can asymptotically match known minimax rates for pool-based active learning, suggesting it can optimally adapt to benign settings. To further examine the performance of the label-efficient forecaster, we conduct a simulation study.
Researcher Affiliation Academia Rui M. Castro Eindhoven University of Technology, Eindhoven Artificial Intelligence Systems Institute (EAISI) rmcastro@tue.nl Fredrik Hellström University College London f.hellstrom@ucl.ac.uk Tim van Erven University of Amsterdam tim@timvanerven.nl
Pseudocode No The paper describes the algorithms in prose and mathematical equations, but does not include structured pseudocode or algorithm blocks.
Open Source Code Yes The full code, which can be executed in less than one hour on an M1 processor, is provided in the supplementary material.
Open Datasets No For the simulations, we use the specific choice ζ(x) = 2sign(x τ0)|x τ0|κ−1 , to generate sequences (Y1, . . . , Yn), based on a sequence of features (X1, . . . , Xn) sampled from the uniform distribution on [0, 1].
Dataset Splits No The paper describes a sequential prediction problem where data is generated for n=50000 rounds. It does not mention traditional train/validation/test dataset splits as it operates in an online setting rather than a batch setting with a fixed dataset.
Hardware Specification Yes The full code, which can be executed in less than one hour on an M1 processor, is provided in the supplementary material.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes In the simulations, we set τ0 = 1/2 and N = n + 1 { n is even}. This choice enforces that N is odd, ensuring the optimal classifier is one of the experts. Throughout, we set η = sqrt(8 ln(N)/n), which minimizes the regret bound (7).