Building Variable-Sized Models via Learngene Pool
Authors: Boyu Shi, Shiyu Xia, Xu Yang, Haokun Chen, Zhiqiang Kou, Xin Geng
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Exhaustive experiments have been implemented and the results validate the effectiveness of the proposed Learngene Pool compared with SN-Net. Main Results and Analysis Learngene Pool vs. SN-Net. We first conduct the comparison between the proposed Learngene Pool and SN-Net (Pan, Cai, and Zhuang 2023). |
| Researcher Affiliation | Academia | Boyu Shi, Shiyu Xia, Xu Yang*, Haokun Chen, Zhiqiang Kou, Xin Geng School of Computer Science and Engineering, Southeast University, Nanjing 210096, China Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China {shiboyu, shiyu xia, 101013120, chenhaokun, zhiqiang kou, xgeng}@seu.edu.cn |
| Pseudocode | Yes | Algorithm 1: Building and Finetuning the Learngene Pool |
| Open Source Code | No | No explicit statement about releasing the code for the methodology described in this paper is found, nor is a specific repository link provided for their own implementation. The paper only mentions using the official code for SN-Net baseline: https://github.com/ziplab/SN-Net. |
| Open Datasets | Yes | We conduct all experiments on Image Net-1K (Russakovsky et al. 2015a) dataset. Image Net-1K is a large-scale image dataset designed for the classification task with 1,000 categories. |
| Dataset Splits | Yes | It consists of a training set with 1.2 million images, and a validation set consisting of 50,000 images. |
| Hardware Specification | No | No specific hardware details (e.g., GPU models, CPU types, memory amounts, or cloud instance specifications) used for running experiments are provided in the paper. |
| Software Dependencies | No | No specific ancillary software details (e.g., library or solver names with version numbers) are provided in the paper. |
| Experiment Setup | Yes | The batch size is set to 128, and the initial learning rate is set to 5 10 4. All other hyperparameters remain consistent with the default setting of SN-Net (Pan, Cai, and Zhuang 2023). |