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..
Adaptive and Multi-scale Affinity Alignment for Hierarchical Contrastive Learning
Authors: Jiawei Huang, Minming Li, Hu Ding
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct a set of experiments on diverse benchmarks to show that AMA-alignment can effectively preserve hierarchical structure; moreover, AMA-alignment also outperforms existing contrastive methods on a range of downstream tasks. |
| Researcher Affiliation | Academia | 1School of Computer Science and Technology, University of Science and Technology of China 2Department of Computer Science, City University of Hong Kong EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1 Adaptive Multi-scale Affinity Alignment (AMA-alignment) for Hierarchical CL |
| Open Source Code | Yes | Anonymized code and data are included in the supplemental material. |
| Open Datasets | Yes | Datasets. We experiment on diverse datasets with inherent hierarchical or fine-grained semantics: (1) Deep Fashion [33], a fine-grained clothing dataset; (2) i Naturalist [56], with taxonomic labels; (3) CIFAR-100 [27], grouped into superclasses; and (4) Model Net40 [61], a 3D object classification dataset. (5) Bu Img [37], a breast cancer diagnosis ultrasound images dataset. |
| Dataset Splits | No | The paper references multiple datasets (Deep Fashion [33], i Naturalist [56], CIFAR-100 [27], Model Net40 [61], Bu Img [37], Image Net, Conceptual Captions 3M [47]), but it does not explicitly provide details about the training, validation, and test splits used for these datasets within the main text or appendices. |
| Hardware Specification | Yes | All models are implemented with Py Torch on a single NVIDIA RTX 6000 Ada GPU. |
| Software Dependencies | No | All models are implemented with Py Torch on a single NVIDIA RTX 6000 Ada GPU. The model is optimized using the Adam W optimizer [34]. |
| Experiment Setup | Yes | The models are trained using Res Net-50 [17] as the backbone and evaluated via linear probing (unless otherwise specified). Training is conducted for 200 epochs using Adam W optimizer [34], with weight decay 5 10 6, and learning rate initialized at 0.01. |