When to Make and Break Commitments?

Authors: Alihan Hüyük, Zhaozhi Qian, Mihaela van der Schaar

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

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we empirically evaluate the performance of our algorithm in running clinical trials with subpopulation selection.
Researcher Affiliation Academia Alihan H uy uk University of Cambridge ah2075@cam.ac.uk Zhaozhi Qian University of Cambridge zq224@cam.ac.uk Mihaela van der Schaar University of Cambridge The Alan Turing Institute mv472@cam.ac.uk
Pseudocode Yes Our algorithm is called Bayes-OCP and is given in Algorithm 1.
Open Source Code Yes Moreover, the source code necessary to reproduce our main results in Table 3 is made publicly available at https://github.com/alihanhyk/optcommit and https://github.com/vanderschaarlab/optcommit.
Open Datasets No To this end, we randomly generated 1000 environments (repeated five times to obtain error bars) with true mean outcomes θXA, θXB sampled independently from N(0.1, 0.1). This indicates a simulated dataset, not a publicly available fixed dataset with access information.
Dataset Splits No The paper describes simulating environments and evaluating algorithms on these simulations, but it does not specify train/validation/test splits of a pre-existing dataset, nor does it provide a methodology for creating reproducible splits if the simulated data were to be saved and reused.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory, cloud instance types) used to run the experiments or simulations.
Software Dependencies No The paper does not specify versions for any software dependencies, libraries, or programming languages used in the experiments.
Experiment Setup Yes In our environments, there are two atomic-populations, X = {XA, XB}. Both atomicpopulations have equal propensities ηXA = ηXB = 1/2 and the meta-experimenter has the same positively-biased prior for the mean outcome associated with each atomic-population: θXA, θXB N(0.1, 0.1). Experiment designs targeting one or both of these atomic-populations all have the same time horizon τ = 600 and success criterion ρ(Dτ) = 1{P (xt,yt) Dτ yt/|Dτ| > α/ τ}, where α = F 1(95%). Rewards are given by RX = 1000η0.1 X and costs are given by CX = 1/η0.1 X. Bayes-OCP is initialized with β = 0.80.