Towards Uniformly Superhuman Autonomy via Subdominance Minimization

Authors: Brian Ziebart, Sanjiban Choudhury, Xinyan Yan, Paul Vernaza

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We apply our approach on a computer cursor pointing task, producing behavior that is 78% superhuman, while minimizing demonstration suboptimality provides 50% superhuman behavior and only 72% even after selective data cleaning.
Researcher Affiliation Collaboration 1Computer Science, University of Illinois Chicago 2Aurora Innovation. Correspondence to: B. Ziebart <bziebart@uic.edu>.
Pseudocode Yes Algorithm 1 Update w and α from demonstration(s) ξ
Open Source Code No No statement regarding the availability of open-source code for the methodology is provided in the paper.
Open Datasets No We focus our experiments on human-generated demonstrations. We analyze pointing task data gathered from 20 non-motor impaired individuals each performing 300 pointing tasks.
Dataset Splits No We randomly split the dataset into a training set of 200 tasks and a testing set of 100 tasks.
Hardware Specification No No specific hardware details (e.g., CPU, GPU models, memory, or cloud instance types) used for running experiments are mentioned in the paper.
Software Dependencies No No specific software dependencies with version numbers are mentioned.
Experiment Setup No We optimize each αk using stochastic exponentiated gradient descent: αk αkeηt(fk( ξ) fk(ξ ) λαk) using an appropriately decaying learning rate ηt, as shown in Algorithm 1.