Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Learning Expandable and Adaptable Representations for Continual Learning
Authors: Ruilong Yu, Mingyan Liu, Fei Ye, Adrian G. Bors, Rongyao Hu, Jingling sun, shijie zhou
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results demonstrate that the proposed framework achieves state-of-the-art performance. Code is available at https://github.com/yrluestc/NeurIPS2025-LEAR. Our comprehensive evaluation compares LEAR against state-of-the-art approaches across three distinct domain sequences (Tables 1 and 2). |
| Researcher Affiliation | Academia | 1University of Electronic Science and Technology of China, Chengdu, China 2Harbin Institute of Technology, Shenzhen, China 3University of York, York, U.K. |
| Pseudocode | Yes | The detailed algorithm is summarized in Appendix-B from SM. |
| Open Source Code | Yes | Code is available at https://github.com/yrluestc/NeurIPS2025-LEAR. |
| Open Datasets | Yes | The datasets used in our experiment can be logically categorized into three primary fields according to [21]. Natural Domains include CIFAR-10 [25] (C10), Tiny Image Net [26] (TImg), CUB-200 [39], MNIST [27] and Image Net-R [17] (Img R), covering a range of tasks from basic image classification to fine-grained recognition and robustness testing across various visual styles. Aerial Domains comprise Euro SAT [16] and RESISC45 [7], focusing on satellite imagery for land cover classification and environmental monitoring. Medical Domains consist of Crop Diseases [31] (Disease) and Chest X [43], specialized for identifying plant diseases and diagnosing medical conditions through radiographic images, respectively. |
| Dataset Splits | Yes | In a class-incremental learning scenario [3], a data split procedure is performed to divide the training dataset DS i into Ci subsets {DS i,1, , DS i,Ci} according to the category, where each task Tj is associated with a training dataset DS i,j formed by samples from several adjacent classes... Detailed experimental configurations are provided in Appendix-C from SM. |
| Hardware Specification | No | Table 3: Comparison of Baselines in terms of parameter and computational efficiency (in CDM). Methods Train Params Iter/s GPU Avg GPU Max CPU Avg CPU Max ... Ours 42.54M 3.06 2926.20MB 2926.20MB 9525.63MB 16681.54MB. The paper mentions 'Detailed experimental configurations are provided in Appendix-C from SM.' but does not specify hardware models or types in the main text. |
| Software Dependencies | No | The paper mentions 'Detailed experimental configurations are provided in Appendix-C from SM.' but does not specify any software dependencies with version numbers in the main text. |
| Experiment Setup | No | λ1, λ2, λ3 are trade-off hyperparameters balancing different loss components. The model parameters {θg, θl, φf j , φc j} is updated using Eq. (13). The detailed algorithm is summarized in Appendix-B from SM. Detailed experimental configurations are provided in Appendix-C from SM. |