Autofocused oracles for model-based design

Authors: Clara Fannjiang, Jennifer Listgarten

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

Reproducibility Variable Result LLM Response
Research Type Experimental demonstrate the promise of autofocusing empirically. and demonstrate empirically that autofocusing holds promise for improving oracle-based design.
Researcher Affiliation Academia Clara Fannjiang and Jennifer Listgarten Department of Electrical Engineering & Computer Sciences University of California, Berkeley Berkeley, CA 94720 {clarafy,jennl}@berkeley.edu
Pseudocode Yes Pseudo-code for autofocusing can be found in the Supplementary Material (Algorithms 1 and 2). and See Algorithm 3 in the Supplementary Material for pseudocode of this procedure.
Open Source Code Yes Code for our experiments is available at https://github.com/clarafy/autofocused_oracles.
Open Datasets Yes we used a dataset comprising 21, 263 superconducting materials paired with their critical temperatures [44]
Dataset Splits No The paper mentions selecting 'training points' and evaluates 'best samples', but does not explicitly describe a separate validation split or how validation was performed for model tuning or early stopping.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory, or cloud instance types) used for running the experiments.
Software Dependencies No The paper mentions using 'gradient-boosted regression trees' and 'neural networks' but does not specify the software frameworks, libraries, or their version numbers (e.g., TensorFlow, PyTorch, scikit-learn versions) required to reproduce the experiments.
Experiment Setup Yes We outline our experiments here, with details deferred to the Supplementary Material S4. and In all cases, we used a full-rank multivariate normal for the search model, and flattened the importance weights used for autofocusing to wα i [24] with α = 0.2 to help control variance. and for our oracle, we used {(xi, yi)}n i=1 to train an ensemble of three neural networks that output both µβ(x) and σ2 β(x), to provide predictions of the form pβ(y | x) = N(µβ(x), σ2 β(x)) [46].