Learnware Specification via Label-Aware Neural Embedding
Authors: Wei Chen, Jun-Xiang Mao, Min-Ling Zhang
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate the effectiveness and superiority of the proposed LANE approach in the learnware paradigm. ... We evaluate the proposed LANE specification approach on two benchmark dataset of different data types: CIFAR100 (Krizhevsky 2009) for image and 20newsgroups (Joachims 1997) for text. ... We conduct comparative studies between the LANE approach and three compared methods: a naive baseline, MAX, alongside a related method, HMR (Wu, Liu, and Zhou 2019), and a specification method, RKME, using the benchmark dataset. ... Empirical results related to the accuracy of model reuse on CIFAR100 and 20newsgroups, based on different numbers of mixed tasks, are reported in Table 1 and Table 2, respectively. The results clearly demonstrate the effectiveness and superiority of our LANE approach in model reuse. |
| Researcher Affiliation | Collaboration | Wei Chen1, 2, Jun-Xiang Mao1, 2, 3, Min-Ling Zhang1, 2* 1 School of Computer Science and Engineering, Southeast University, Nanjing, China 2 Key Laboratory of Computer Network and Information Integration (Southeast University), Ministry of Education, China 3 Information Technology and Data Management Department of China Mobile Communications Group Zhejiang Co., Ltd |
| Pseudocode | Yes | Algorithm 1: The submitting stage of our LANE approach Algorithm 2: The deploying stage of our LANE approach |
| Open Source Code | No | The paper does not contain any explicit statement about the authors releasing their code or a direct link to a code repository for the methodology described (LANE). It mentions third-party resources for feature extractors but not for their own implementation. |
| Open Datasets | Yes | We evaluate the proposed LANE specification approach on two benchmark dataset of different data types: CIFAR100 (Krizhevsky 2009) for image and 20newsgroups (Joachims 1997) for text. |
| Dataset Splits | No | The paper describes how superclass datasets are used to form tasks and requirements, and that the 'number of requirements fixed at 50 for each mixed task.' However, it does not provide specific percentages or sample counts for standard training, validation, and test splits, nor does it refer to predefined splits with citations or a detailed splitting methodology. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory amounts used for running its experiments. It mentions using specific models for feature extraction (Res Net110, Sentence Transformer) but not the hardware these ran on. |
| Software Dependencies | No | The paper mentions using 'Res Net110', 'Sentence Transformer', 'Conv Net BN', and 'SVM method', and refers to 'solvers (Vandenberghe 2010)' for quadratic programming. However, it does not provide specific version numbers for any of these software components or libraries. |
| Experiment Setup | Yes | The parameter settings for the compared methods are as follows: The RKME method uses a Gaussian kernel function (γ = 0.01) and the size of the specification is set to 10; the example communication budget in the HMR method is set to 10. ... the number of requirements fixed at 50 for each mixed task. |