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..
Backdoor Cleaning without External Guidance in MLLM Fine-tuning
Authors: Xuankun Rong, Wenke Huang, Jian Liang, Jinhe Bi, Xun Xiao, Yiming Li, Bo Du, Mang Ye
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments across various datasets, models, and diverse trigger types validate BYE s effectiveness: it achieves near-zero attack success rates while maintaining clean-task performance, offering a robust and generalizable solution against backdoor threats in MLLMs. |
| Researcher Affiliation | Collaboration | 1School of Computer Science, Wuhan University 2Munich Research Center, Huawei Technologies 3Nanyang Technological University |
| Pseudocode | Yes | Algorithm 1: Believe Your Eyes (BYE): Attention Entropy-Driven Backdoor Cleaning |
| Open Source Code | Yes | Our code is publicly available at: https://github.com/Xuankun Rong/BYE. |
| Open Datasets | Yes | Science QA [55], Icon QA [56], and RSVQA [54] are used for visual question answering (VQA), while Flickr30k [95] is used for image captioning. |
| Dataset Splits | Yes | Science QA. ...We use 6,218 training and 2,017 test samples. Icon QA. ...We follow the multiple-choice setting (10,000 train / 6,316 test). Flickr30k. ...We select a subset containing 10,000 training and 1,000 test images... RSVQA. ...We select 10,000 training and 10,004 test samples. |
| Hardware Specification | Yes | All models were fine-tuned using 4 NVIDIA RTX 4090 GPUs (48 GB each). |
| Software Dependencies | No | The paper mentions the use of Adam W optimizer but does not specify version numbers for other key software components like programming languages (e.g., Python), deep learning frameworks (e.g., PyTorch, TensorFlow), or CUDA versions. |
| Experiment Setup | Yes | For each dataset, models were trained for 3 epochs with a global batch size of 16. The learning rate was set to 2e-4 for LLa VA-1.5-7B and 4e-5 for Intern VL-2.5-8B. Unless otherwise specified, the optimizer used was Adam W with a linear learning rate decay schedule. Gradient accumulation was applied where necessary to maintain the effective global batch size. |