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..
Vector Quantization Prompting for Continual Learning
Authors: Li Jiao, Qiuxia LAI, YU LI, Qiang Xu
NeurIPS 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show that VQPrompt outperforms state-of-the-art continual learning methods across a variety of benchmarks under the challenging class-incremental setting. |
| Researcher Affiliation | Academia | Li Jiao1, Qiuxia Lai1 , Yu Li2, Qiang Xu3 1 Communication University of China 2 Harbin Institute of Technology, Shenzhen 3 The Chinese University of Hong Kong |
| Pseudocode | No | The paper does not contain any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | Yes | The code is available at https://github.com/jiaolifengmi/VQ-Prompt. |
| Open Datasets | Yes | Datasets. We consider three representative benchmarks for evaluating CIL. Image Net-R [14] includes 200-class images... Split CIFAR-100 randomly splits the original CIFAR-100 [25] into 10 disjoint tasks... Split CUB-200 is built on CUB-200-2011 [57], a fine-grained classification dataset... |
| Dataset Splits | Yes | Following Dual Prompt [61] and CODA-Prompt [50], we use 20% of the training data as validation data, and perform hyperparameters tuning on it. |
| Hardware Specification | Yes | Each experiment is run on a single NVIDIA Ge Force RTX 4090 GPU. |
| Software Dependencies | No | The paper mentions using Adam W optimizer with specific beta values but does not provide version numbers for key software components like Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | Our method is trained using an Adam W optimizer [35] with an initial learning rate of 0.0025 and a cosine decay schedule. The batch size is 128 for Split CIFAR-100 and Split CUB-200, and 64 for Image Net-R. The number of epochs is set to be 20 for training on all three datasets. The classifier bias mitigation process described in 4.3 requires ten epochs of training. |