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 [1].
Stabilizing Sample Similarity in Representation via Mitigating Random Consistency
Authors: Jieting Wang, Zelong Zhang, Feijiang Li, Yuhua Qian, Xinyan Liang
ICML 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically, our method surpasses conventional loss functions across multiple benchmarks, achieving significant improvements in accuracy, F1 score, and class-structure differentiation. (Code is published in https://github.com/Feijiang Li/ICML2025PSED) |
| Researcher Affiliation | Academia | 1Institute of Big Data Science and Industry, Key Laboratory of Evolutionary Science Intelligence of Shanxi Province, Shanxi University, Taiyuan, China. Correspondence to: Yuhua Qian <EMAIL>. |
| Pseudocode | Yes | The specific algorithm process of the method used in this paper is shown in Algorithm 1, where RL is the representation layer and CL is the classification layer. |
| Open Source Code | Yes | (Code is published in https://github.com/Feijiang Li/ICML2025PSED) |
| Open Datasets | Yes | We provide a detailed description of the dataset and download links. For more detailed information, please refer to Tables 4 and 5. Table 5. Addresses of Datasets ID Data Address 1 https://archive.ics.uci.edu/dataset/186/wine+quality |
| Dataset Splits | Yes | To ensure consistency in the evaluation, each dataset is randomly divided into training, validation, and testing sets in a 5:2:3 ratio. |
| Hardware Specification | Yes | In this study, we conducted experiments using hardware configurations including Intel (R) Core (TM) i7-14700F CPU, 16GB RAM, and NVIDIA Ge Force RTX 4060 GPU. |
| Software Dependencies | Yes | The experiment was conducted on the Windows operating system, with Python 3.10 as the programming language and Py Torch 2.4 library for model development and training. |
| Experiment Setup | Yes | The training process uses the Adam optimizer, which is widely favored for its efficiency in adjusting learning rates. The learning rate of each dataset has been fine tuned to achieve optimal convergence performance. Training for up to 60 epochs provides ample time for model learning and reduces the risk of overfitting. We also customized hidden layers and regularization parameters for each dataset. This customization takes into account the unique characteristics and complexity of each dataset, ensuring that the model architecture is best suited for optimal performance. The batch size is set to 16. For feature extraction, Image Net employs the self-supervised learning model MOCO V3 (Chen et al., 2021), while all other datasets utilize VGG (Fernandez-Delgado et al., 2014). |