Yuancheng Yang

Yuqi Zhang, Xiaoqian Zhang, Jiakai Wang et al.

Computational pathology (CPath) has significantly advanced the clinical practice of pathology. Despite the progress made, Multiple Instance Learning (MIL), a promising paradigm within CPath, continues to face challenges, particularly related to incomplete information utilization. Existing frameworks, such as those based on Convolutional Neural Networks (CNNs), attention, and selective scan space state sequential model (SSM), lack sufficient flexibility and scalability in fusing diverse features, and cannot effectively fuse diverse features. Additionally, current approaches do not adequately exploit order-related and order-independent features, resulting in suboptimal utilization of sequence information. To address these limitations, we propose a novel MIL framework called Mamba2MIL. Our framework utilizes the state space duality model (SSD) to model long sequences of patches of whole slide images (WSIs), which, combined with weighted feature selection, supports the fusion processing of more branching features and can be extended according to specific application needs. Moreover, we introduce a sequence transformation method tailored to varying WSI sizes, which enhances sequence-independent features while preserving local sequence information, thereby improving sequence information utilization. Extensive experiments demonstrate that Mamba2MIL surpasses state-of-the-art MIL methods. We conducted extensive experiments across multiple datasets, achieving improvements in nearly all performance metrics. Specifically, on the NSCLC dataset, Mamba2MIL achieves a binary tumor classification AUC of 0.9533 and an accuracy of 0.8794. On the BRACS dataset, it achieves a multiclass classification AUC of 0.7986 and an accuracy of 0.4981. The code is available at https://github.com/YuqiZhang-Buaa/Mamba2MIL.

Yuancheng Yang, Lin Yang, Xu Wang et al.

As applications of large language models (LLMs) become increasingly complex, the demand for robust complex instruction following capabilities is growing accordingly. We argue that a thorough understanding of the instruction itself, especially the latent reasoning structure embedded between the lines, is crucial for improving instruction following. Therefore we target complex instructions that involve implicit reasoning, intricate logical relations, and multi-constraint dependencies. We propose ImpRIF, a method to enhance LLMs' understanding of implicit reasoning instructions, thereby improving its ability to follow complex instructions. We formalize such instructions as verifiable reasoning graphs, enabling programmatic verification and graph-driven chain-of-thought reasoning. Based on this formulation, we synthesize large-scale single- and multi-turn data, propose fine-tuning with graph reasoning, and apply reinforcement learning to explicitly train models to reason along the graph. On five complex instruction following benchmarks, our models substantially outperform their base models. These results demonstrate that enhancing implicit reasoning capabilities can significantly improve complex instruction following.

Lin Yang, Yuancheng Yang, Xu Wang et al.

Large Language Models (LLMs) have demonstrated impressive capabilities across various specialist domains and have been integrated into high-stakes areas such as medicine. However, as existing medical-related benchmarks rarely stress-test the long-context memory, interference robustness, and safety defense required in practice. To bridge this gap, we introduce MedMT-Bench, a challenging medical multi-turn instruction following benchmark that simulates the entire diagnosis and treatment process. We construct the benchmark via scene-by-scene data synthesis refined by manual expert editing, yielding 400 test cases that are highly consistent with real-world application scenarios. Each test case has an average of 22 rounds (maximum of 52 rounds), covering 5 types of difficult instruction following issues. For evaluation, we propose an LLM-as-judge protocol with instance-level rubrics and atomic test points, validated against expert annotations with a human-LLM agreement of 91.94\%. We test 17 frontier models, all of which underperform on MedMT-Bench (overall accuracy below 60.00\%), with the best model reaching 59.75\%. MedMT-Bench can be an essential tool for driving future research towards safer and more reliable medical AI. The benchmark is available in https://openreview.net/attachment?id=aKyBCsPOHB&name=supplementary_material