Adaptive Stochastic Gradient Algorithm for Black-box Multi-Objective Learning

Authors: Feiyang Ye, Yueming Lyu, Xuehao Wang, Yu Zhang, Ivor Tsang

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, the proposed ASMG method achieves competitive performance on multiple numerical benchmark problems. Additionally, the state-of-the-art performance on the black-box multi-task learning problem demonstrates the effectiveness of the proposed ASMG method.
Researcher Affiliation Collaboration 1Department of Computer Science and Engineering, Southern University of Science and Technology 2Australian Artificial Intelligence Institute, University of Technology Sydney 3Centre for Frontier AI Research, Agency for Science, Technology and Research, Singapore 4Institute of High Performance Computing, Agency for Science, Technology and Research, Singapore 5Shanghai Artificial Intelligence Laboratory 6School of Computer Science and Engineering, Nanyang Technological University
Pseudocode Yes Algorithm 1 The ASMG Method
Open Source Code No The paper mentions using and implementing existing methods (CVXPY library, CMA-ES, ES, BES) and provides a link to the official CMA-ES implementation, but it does not state that the code for their proposed ASMG method is open-source or provide a link to it.
Open Datasets Yes We conduct experiments on two MTL benchmark datasets (Lin & Zhang, 2023), i.e., Office-31 (Saenko et al., 2010) and Office-home (Venkateswara et al., 2017).
Dataset Splits No The paper mentions using Office-31 and Office-home datasets and specifies training parameters like batch size and epochs, but it does not explicitly provide the specific training/validation/test dataset splits (e.g., percentages, counts, or explicit citations to specific standard splits) used for reproducibility.
Hardware Specification Yes The experiments are conducted on a single NVIDIA Ge Force RTX 3090 GPU.
Software Dependencies No Here we use the open-source CVXPY library (Diamond & Boyd, 2016) to solve it. The paper mentions CVXPY but does not specify its version number. No other specific software with version numbers is listed.
Experiment Setup Yes For all methods, we set µ0 = 0 and Σ0 = I as initialization. For all baseline methods except the zero-shot setting, we optimize the prompt with a batch size of 64 for 200 epochs, the population size N is set as 20 for Office-31 and 40 for Office-home... For ASMG and ASMG-EW methods, the step size is fixed as β = 0.5. The coefficient γ in the ASMG method is set as γt = 1/(t + 1).