Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Amortized Active Generation of Pareto Sets
Authors: Daniel M Steinberg, Asiri Wijesinghe, Rafael Oliveira, Piotr Koniusz, Cheng Soon Ong, Edwin V. Bonilla
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results on synthetic benchmarks and protein design tasks demonstrate strong sample efficiency and effective preference incorporation. |
| Researcher Affiliation | Academia | 1CSIRO s Data61 2University of New South Wales 3Australian National University |
| Pseudocode | Yes | Algorithm 1 A-GPS optimization loop. |
| Open Source Code | Yes | For code implementing A-GPS, VSD and all of the experimental results, please see github.com/csiro-funml/variationalsearch. |
| Open Datasets | Yes | Empirical results on synthetic benchmarks and protein design tasks demonstrate strong sample efficiency and effective preference incorporation. ... Ehrlich synthetic landscape [37] with a Prot Bert [18] naturalness score. ... bi-grams optimization task from [38] ... simulation-based protein stability vs. solvent accessible surface area (SASA) task from [38]. |
| Dataset Splits | Yes | All methods use 64 training points, and then recommend B = 5 candidates for T = 10 rounds... All methods are given 128 training samples... We start with 512 random sequences... 512 training samples, T = 64, B = 16... Train prior with a 10% validation set... |
| Hardware Specification | Yes | All experiments were run on a Dell Power Edge XE9640 rack server cluster with NVIDIA H100 GPUs and 4th generation Intel Xeon CPUs. |
| Software Dependencies | No | The paper mentions 'Bo Torch [4]' and 'poli and poli-baselines libraries [20]' but does not provide specific version numbers for these software dependencies. |
| Experiment Setup | Yes | For all experiments we set β = 0.5 as the full KL regularization in Equation 15 can hamper exploitation in later rounds on some tasks. We refer the reader to Appendix D for full experimental details. ... Table 4: Synthetic test functions experimental settings. ... Table 6: Sequence experimental settings. |