COFlowNet: Conservative Constraints on Flows Enable High-Quality Candidate Generation
Authors: Yudong Zhang, Xuan Yu, Xu Wang, Zhaoyang Sun, Chen Zhang, Pengkun Wang, Yang Wang
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on several widely-used datasets validate the effectiveness of COFlow Net in generating high-scored and diverse candidates. All implementations are available at https://github.com/yuxuan9982/COflownet. In this section, we evaluate the proposed COFlow Net on two tasks, Hypergrid and molecule design. During the evaluation, we mainly focus on two research questions: 1) How is the performance of candidates generated by COFlow Net? 2) How is the diversity of the generated candidates? To facilitate a more comprehensive evaluation, we select various metrics tailored to different tasks. |
| Researcher Affiliation | Academia | Yudong Zhang1, Xuan Yu1, Xu Wang1,2, , Zhaoyang Sun1, Chen Zhang1, Pengkun Wang1,2, Yang Wang1,2,3 1. University of Science and Technology of China (USTC), Hefei, China 2. Suzhou Institute for Advanced Research, USTC, Suzhou, China 3. State Key Laboratory of Precision and Intelligent Chemistry, USTC, Hefei, China {zyd2020@mail., yx2024@mail., wx309@}ustc.edu.cn {sunzhaoyang@mail., zhangchenzc@mail., pengkun@, angyan@}ustc.edu.cn |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. The methods are described using mathematical equations and textual explanations. |
| Open Source Code | Yes | All implementations are available at https://github.com/yuxuan9982/COflownet. |
| Open Datasets | Yes | We first evaluate the proposed COFlow Net on the hypergrid task from Bengio et al. (2021). ... A proxy model PD is trained on DL to serve as oracle, since we are unable to access the real oracle due to the expensive computation. ... The experiment is conducted on ML1M dataset, a subset of the Movie Lens dataset2. The dataset is available at https://grouplens.org/datasets/movielens/1m/ |
| Dataset Splits | No | The paper mentions constructing an offline dataset for Hypergrid ( |
| Hardware Specification | Yes | All the experiments are conducted on an NVIDIA Tesla A100 80GB. |
| Software Dependencies | No | The paper does not provide specific version numbers for any software dependencies. |
| Experiment Setup | No | The paper describes the task definitions, reward functions, and how offline data was constructed, but it does not specify concrete hyperparameters like learning rate, batch size, optimizer type, or the number of epochs used for training COFlowNet or its baselines. |