Offline Model-Based Optimization by Learning to Rank

Authors: Rong-Xi Tan, Ke Xue, Shen-Huan Lyu, Haopu Shang, Yao Wang, Yaoyuan Wang, Fu Sheng, Chao Qian

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical results across diverse tasks demonstrate the superior performance of our proposed ranking-based method than twenty existing methods. Our implementation is available at https://github.com/ lamda-bbo/Offline-Ra M. In this section, we empirically compare the proposed method with a large variety of previous offline MBO methods on various tasks.
Researcher Affiliation Collaboration 1 National Key Laboratory for Novel Software Technology, Nanjing University, China 2 School of Artificial Intelligence, Nanjing University, China 3 Key Laboratory of Water Big Data Technology of Ministry of Water Resources, Hohai University, China 4 College of Computer Science and Software Engineering, Hohai University, China 5 Advanced Computing and Storage Lab, Huawei Technologies Co., Ltd., China
Pseudocode Yes Algorithm 1 Offline MBO by Learning to Rank
Open Source Code Yes Our implementation is available at https://github.com/ lamda-bbo/Offline-Ra M.
Open Datasets Yes We benchmark our method on Design-Bench tasks (Trabucco et al., 2022), including three continuous tasks and two discrete tasks.
Dataset Splits Yes We split the dataset into a training set and a validation set of the ratio 8 : 2. ... In Design-Bench, the training dataset is selected as the bottom performing x% in the entire collected dataset, (i.e., x = 40, 50, 60). ... We identify the excluded (100 x)% high-scoring data to comprise the OOD dataset for analysis...
Hardware Specification No No specific hardware details (like GPU/CPU models or cloud instance types) were provided in the paper. The paper mentions using PyTorch for implementation but does not specify the hardware used to run experiments.
Software Dependencies No The paper mentions using PyTorch (Paszke et al., 2019) and Adam (Kingma & Ba, 2015) but does not provide specific version numbers for these software components.
Experiment Setup Yes We set the size n of training dataset to 10, 000, and following LETOR 4.0 (Qin & Liu, 2013; Qin et al., 2010b), a prevalent benchmark for LTR, we set the list length m = 1000. ... The model is optimized using Adam (Kingma & Ba, 2015) with a learning rate of 3 10 4 and a weight decay coefficient of 1 10 5. After the model is trained... we set η = 1 10 3 and T = 200 for continuous tasks, and η = 1 10 1 and T = 100 for discrete tasks to search for the final design. We use Re LU as activation functions.