Adaptive Experimental Design with Temporal Interference: A Maximum Likelihood Approach

Authors: Peter W. Glynn, Ramesh Johari, Mohammad Rasouli

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical Our paper provides an optimal experimental design within a benchmark theoretical model for settings with temporal interference (Section 2). The central challenge posed by temporal interference is the following: we are effectively allowed only one real-world run of the system, with only finitely many observations. On the other hand, we need to use this single run to estimate performance of both the systems induced by each of the treatment and control policies. We model the problem by viewing each policy as its own Markov chain on a common underlying state space. The experimental design problem is then to estimate the difference in the steady state reward under the treatment and control Markov chains, using only one run of the system, and without prior knowledge of any of the parameters of either policy or their rewards.
Researcher Affiliation Academia Stanford University, Stanford, CA, 94305 {glynn, rjohari, rasoulim}@stanford.edu
Pseudocode No The paper states 'The full pseudocode of Online ETI is in [10].' [10] refers to a companion technical report, not the current paper itself. Therefore, the pseudocode is not provided within this paper.
Open Source Code No The paper does not provide any statement or link regarding the availability of open-source code for the methodology described.
Open Datasets No The paper is theoretical and focuses on experimental design. It does not use or refer to any publicly available dataset for empirical evaluation.
Dataset Splits No The paper is theoretical and does not present empirical results from data, therefore no dataset split information (train, validation, test) is provided.
Hardware Specification No The paper is theoretical and does not report on conducted experiments, so no specific hardware details are mentioned.
Software Dependencies No The paper is theoretical and does not report on conducted experiments, so no specific software dependencies with version numbers are mentioned.
Experiment Setup No The paper is theoretical and does not report on conducted experiments, so no specific experimental setup details such as hyperparameters or training configurations are provided.