Wenxuan Shen

Wenxuan Shen

First year PhD student

Posts by Collection

publications

Multi-group Particle Swarm Optimization with K-means Gap Filling for Finding Gamma Knife Treatment Plans

Published in 2023 13th International Conference on Information Science and Technology (ICIST)

The sphere packing problem (SPP) is a geometric optimization problem which is NP-hard. It seeks to find the most efficient arrangement of non-overlapping spheres within a containing space to maximize the total volume or the fraction of the containing space occupied by the spheres. As an extension of SPP, Gamma knife treatment (GKT) is a significant technique in the field of stereotactic radiosurgery, which uses Gamma rays to perform single-shot high-dose focused irradiation on the lesion to cause necrosis or functional changes of diseased tissue to achieve the purpose of treatment. Therefore, the GKT plan problem should be addressed to minimize the total dose passing through the human body, but must be enough to kill most lesions, while minimizing the dose of normal tissue exposure. This paper proposes a multi-group particle swarm optimization (MGPSO) algorithm with K-means gap filling (KGF) strategy to handle this problem. The MGPSO is adopted to find preliminary locations for Gamma knife shots. The KGF is used to find uncovered areas and fill them. The effectiveness of the MGPSO algorithm with the KGF strategy is verified by experiments on circular lesion areas by comparing with other algorithms.

Recommended citation: D. -J. Zhan, W. -X. Shen, J. Hong, X. -X. Xu and Z. -H. Zhan, "Multi-group Particle Swarm Optimization with K-means Gap Filling for Finding Gamma Knife Treatment Plans," 2023 13th International Conference on Information Science and Technology (ICIST), Cairo, Egypt, 2023, pp. 151-156, doi: 10.1109/ICIST59754.2023.10367177.
Download Paper

CrowdSelect: Synthetic Instruction Data Selection with Multi-LLM Wisdom

Submitted to ARR 2026 October cycle

Distilling advanced Large Language Models’ instruction-following capabilities into smaller models using a selected subset has become a mainstream approach in model training. While existing synthetic instruction data selection strategies rely mainly on single-dimensional signals (i.e., reward scores, model perplexity), they fail to capture the complexity of instruction-following across diverse fields. Therefore, we investigate more diverse signals to capture comprehensive instruction-response pair characteristics and propose three foundational metrics that leverage Multi-LLM wisdom, informed by (1) diverse LLM responses and (2) reward model assessment. Building upon base metrics, we propose CrowdSelect, an integrated metric incorporating a clustering-based approach to maintain response diversity. Our comprehensive experiments demonstrate that our foundation metrics consistently improve performance across 4 base models on MT-bench and Arena-Hard. CrowdSelect, efficiently incorporating all metrics, achieves state-of-the-art performance in both Full and LoRA fine-tuning, showing improvements of 4.81% on Arena-Hard and 11.1% on MT-bench with Llama-3.2-3b-instruct. We hope our findings will bring valuable insights for future research in this direction.

Recommended citation: Li, Y., Yang, L., Shen, W., Zhou, P., Wan, Y., Lin, W., & Chen, D. (2025). CrowdSelect: Synthetic Instruction Data Selection with Multi-LLM Wisdom. arXiv preprint arXiv:2503.01836.
Download Paper

Are We on the Right Way for Assessing Document Retrieval-Augmented Generation?

Submitted to AAAI 2026

Retrieval-Augmented Generation (RAG) systems using Multimodal Large Language Models (MLLMs) show great promise for complex document understanding, yet their development is critically hampered by inadequate evaluation. Current benchmarks often focus on specific part of document RAG system and use synthetic data with incomplete ground truth and evidence labels, therefore failing to reflect real-world bottlenecks and challenges. To overcome these limitations, we introduce Double-Bench: a new large-scale, multilingual, and multimodal evaluation system that is able to produce fine-grained assessment to each component within document RAG systems. It comprises 3,276 documents (72,880 pages) and 5,168 single- and multi-hop queries across 6 languages and 4 document types with streamlined dynamic update support for potential data contamination issues. Queries are grounded in exhaustively scanned evidence pages and verified by human experts to ensure maximum quality and completeness. Our comprehensive experiments across 9 state-of-the-art embedding models, 4 MLLMs and 4 end-to-end document RAG frameworks demonstrate the gap between text and visual embedding models is narrowing, highlighting the need in building stronger document retrieval models. Our findings also reveal the over-confidence dilemma within current document RAG frameworks that tend to provide answer even without evidence support. We hope our fully open-source Double-Bench provide a rigorous foundation for future research in advanced document RAG systems.

Recommended citation: Shen, W., Wang, M., Wang, Y., Chen, D., Yang, J., Wan, Y., & Lin, W. (2025). Are We on the Right Way for Assessing Document Retrieval-Augmented Generation?. arXiv preprint arXiv:2508.03644.
Download Paper