Smooth Tchebycheff Scalarization for Multi-Objective Optimization
Authors: Xi Lin, Xiaoyuan Zhang, Zhiyuan Yang, Fei Liu, Zhenkun Wang, Qingfu Zhang
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct various experiments on diverse multi-objective optimization problems. The results confirm the effectiveness of our proposed STCH scalarization. |
| Researcher Affiliation | Academia | 1City University of Hong Kong (email: xi.lin@my.cityu.edu.hk) 2Southern University of Science and Technology. Correspondence to: Qingfu Zhang <qingfu.zhang@cityu.edu.hk>. |
| Pseudocode | Yes | Algorithm 1 STCH for Multi-Objective Optimization |
| Open Source Code | Yes | 1Our source code is available at: github.com/Xi-L/STCH. |
| Open Datasets | Yes | NYUv2 (Silberman et al., 2012) is an indoor scene understanding dataset with 3 tasks on semantic segmentation, depth estimation, and surface normal prediction. [...] Office-31 (Saenko et al., 2010) is an image classification dataset across 3 domains (Amazon, DSLR, and Webcam). [...] QM9 (Ramakrishnan et al., 2014) is a molecular property prediction dataset with 11 tasks. |
| Dataset Splits | Yes | The data split from (Lin et al., 2022a) is utilized to split the data as 60%-20%-20% for training, validation, and testing. (Office-31) [...] The data split in Navon et al. (2022) is used to divide the dataset into 110, 000 for training, 10, 000 for validation, and 10, 000 for testing. (QM9) |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running the experiments (e.g., GPU models, CPU types, or memory specifications). |
| Software Dependencies | No | The paper mentions software like 'Lib MTL library (Lin & Zhang, 2023)' and 'Adam (Kingma & Ba, 2015)' but does not specify exact version numbers for these or other software components used in the experiments. |
| Experiment Setup | Yes | The model is trained for 200 epochs with Adam (Kingma & Ba, 2015), of which the learning rate is initially set to 10 4 with 10 5 weight decay and will be halved to 5 10 5 after 100 epochs. The batch size is set to 2. (NYUv2) [...] The learning rate is 10 4 with 10 5 weight decay. The batch size is 64 and the training epoch is 100. (Office-31) [...] The learning rate is 10 3 with the Reduce LROn Plateau scheduler. The batch size is 128 and the number of training epochs is 300. (QM9) |