Effective clinical history taking is a foundational yet underexplored component of clinical reasoning. While large language models (LLMs) have shown promise on static benchmarks, they often fall short in dynamic, multi-turn diagnostic settings that require iterative questioning and hypothesis refinement. To address this gap, we propose Note2Chat, a note-driven framework that trains LLMs to conduct structured history taking and diagnosis by learning from widely available medical notes. Instead of relying on scarce and sensitive dialogue data, we convert real-world medical notes into high-quality doctor-patient dialogues using a decision tree-guided generation and refinement pipeline. We then propose a three-stage fine-tuning strategy combining supervised learning, simulated data augmentation, and preference learning. Furthermore, we propose a novel single-turn reasoning paradigm that reframes history taking as a sequence of single-turn reasoning problems. This design enhances interpretability and enables local supervision, dynamic adaptation, and greater sample efficiency. Experimental results show that our method substantially improves clinical reasoning, achieving gains of +16.9 F1 and +21.0 Top-1 diagnostic accuracy over GPT-4o.

Medical phrase grounding is crucial for identifying relevant regions in medical images based on phrase queries, facilitating accurate image analysis and diagnosis. However, current methods rely on manual extraction of key phrases from medical reports, reducing efficiency and increasing the workload for clinicians. Additionally, the lack of model confidence estimation limits clinical trust and usability. In this paper, we introduce a novel task-Medical Report Grounding (MRG)-which aims to directly identify diagnostic phrases and their corresponding grounding boxes from medical reports in an end-to-end manner. To address this challenge, we propose uMedGround, a a robust and reliable framework that leverages a multimodal large language model to predict diagnostic phrases by embedding a unique token, < $\mathtt {BOX}$BOX >, into the vocabulary to enhance detection capabilities. A vision encoder-decoder processes the embedded token and input image to generate grounding boxes. Critically, uMedGround incorporates an uncertainty-aware prediction model, significantly improving the robustness and reliability of grounding predictions. Experimental results demonstrate that uMedGround outperforms state-of-the-art medical phrase grounding methods and fine-tuned large visual-language models, validating its effectiveness and reliability. This study represents a pioneering exploration of the MRG task, marking the first-ever endeavor in this domain. Additionally, we demonstrate the applicability of uMedGround in medical visual question answering and class-based localization tasks, where it highlights visual evidence aligned with key diagnostic phrases, supporting clinicians in interpreting various types of textual inputs, including free-text reports, visual question answering queries, and class labels.

Automated radiology reporting holds immense clinical potential in alleviating the burdensome workload of radiologists and mitigating diagnostic bias. Recently, retrieval-based report generation methods have garnered increasing attention. These methods predefine a set of candidate queries and compose reports by searching for sentences in an off-the-shelf sentence gallery that best match these candidate queries. However, due to the long-tail distribution of the training data, these models tend to learn frequently occurring sentences and topics, overlooking the rare topics. Regrettably, in many cases, the descriptions of rare topics often indicate critical findings that should be mentioned in the report. To address this problem, we introduce a Topicwise Separable Sentence Retrieval (Teaser) for medical report generation. To ensure comprehensive learning of both common and rare topics, we categorize queries into common and rare types to learn differentiated topics, and then propose Topic Contrastive Loss to effectively align topics and queries in the latent space. Moreover, we integrate an Abstractor module following the extraction of visual features, which aids the topic decoder in gaining a deeper understanding of the visual observational intent. Experiments on the MIMIC-CXR and IU X-ray datasets demonstrate that Teaser surpasses state-of-the-art models, while also validating its capability to effectively represent rare topics and establish more dependable correspondences between queries and topics. The code is available at https://github.com/CindyZJT/Teaser.git.

Medical Vision-Language Pretraining (MedVLP) shows promise in learning generalizable and transferable visual representations from paired and unpaired medical images and reports. MedVLP can provide useful features to downstream tasks and facilitate adapting task-specific models to new setups using fewer examples. However, existing MedVLP methods often differ in terms of datasets, preprocessing, and finetuning implementations. This pose great challenges in evaluating how well a MedVLP method generalizes to various clinically-relevant tasks due to the lack of unified, standardized, and comprehensive benchmark. To fill this gap, we propose BenchX, a unified benchmark framework that enables head-to-head comparison and systematical analysis between MedVLP methods using public chest X-ray datasets. Specifically, BenchX is composed of three components: 1) Comprehensive datasets covering nine datasets and four medical tasks; 2) Benchmark suites to standardize data preprocessing, train-test splits, and parameter selection; 3) Unified finetuning protocols that accommodate heterogeneous MedVLP methods for consistent task adaptation in classification, segmentation, and report generation, respectively. Utilizing BenchX, we establish baselines for nine state-of-the-art MedVLP methods and found that the performance of some early MedVLP methods can be enhanced to surpass more recent ones, prompting a revisiting of the developments and conclusions from prior works in MedVLP. Our code are available at https://github.com/yangzhou12/BenchX.

We consider the task of generating designs directly from natural language descriptions, and consider floor plan generation as the initial research area.Language conditional generative models have recently been very successful in generating high-quality artistic images.However, designs must satisfy different constraints that are not present in generating artistic images, particularly spatial and relational constraints.We make multiple contributions to initiate research on this task.First, we introduce a novel dataset, Tell2Design (T2D), which contains more than 80k floor plan designs associated with natural language instructions.Second, we propose a Sequence-to-Sequence model that can serve as a strong baseline for future research.Third, we benchmark this task with several text-conditional image generation models.We conclude by conducting human evaluations on the generated samples and providing an analysis of human performance.We hope our contributions will propel the research on language-guided design generation forward 1 .

Document-level relation extraction (RE) aims at extracting relations among entities expressed across multiple sentences, which can be viewed as a multi-label classification problem. In a typical document, most entity pairs do not express any pre-defined relation and are labeled as "none" or "no relation". For good document-level RE performance, it is crucial to distinguish such none class instances (entity pairs) from those of pre-defined classes (relations). However, most existing methods only estimate the probability of pre-defined relations independently without considering the probability of "no relation". This ignores the context of entity pairs and the label correlations between the none class and pre-defined classes, leading to sub-optimal predictions. To address this problem, we propose a new multi-label loss that encourages large margins of label confidence scores between each pre-defined class and the none class, which enables captured label correlations and context-dependent thresholding for label prediction. To gain further robustness against positive-negative imbalance and mislabeled data that could appear in real-world RE datasets, we propose a margin regularization and a margin shifting technique. Experimental results demonstrate that our method significantly outperforms existing multi-label losses for document-level RE and works well in other multi-label tasks such as emotion classification when none class instances are available for training.

Robust tensor factorization is a fundamental problem in machine learning and computer vision, which aims at decomposing tensors into low-rank and sparse components. However, existing methods either suffer from limited modeling power in preserving low-rank structures, or have difficulties in determining the target tensor rank and the trade-off between the low-rank and sparse components. To address these problems, we propose a fully Bayesian treatment of robust tensor factorization along with a generalized sparsity-inducing prior. By adapting the recently proposed low-tubal-rank model in a generative manner, our method is effective in preserving low-rank structures. Moreover, benefiting from the proposed prior and the Bayesian framework, the proposed method can automatically determine the tensor rank while inferring the trade-off between the low-rank and sparse components. For model estimation, we develop a variational inference algorithm, and further improve its efficiency by reformulating the variational updates in the frequency domain. Experimental results on both synthetic and real-world datasets demonstrate the effectiveness of the proposed method in multi-rank determination as well as its superiority in image denoising and background modeling over state-of-the-art approaches.

Subspace learning for tensors attracts increasing interest in recent years, leading to the development of multilinear extensions of principal component analysis (PCA) and probabilistic PCA (PPCA). Existing multilinear PPCAs are based on the Tucker or CANDECOMP/PARAFAC (CP) models. Although both kinds of multilinear PPCAs have shown their effectiveness in dealing with tensors, they also have their own limitations. Tucker-based multilinear PPCAs have a restrictive subspace representation and suffer from rotational ambiguity, while CP-based ones are more prone to overfitting. To address these problems, we propose probabilistic rank-one tensor analysis (PROTA), a CP-based multilinear PPCA. PROTA has a more flexible subspace representation than Tucker-based PPCAs, and avoids rotational ambiguity. To alleviate overfitting for CP-based PPCAs, we propose two simple and effective regularization strategies, named as concurrent regularizations (CRs). By adjusting the noise variance or the moments of latent features, our strategies concurrently and coherently penalize the entire subspace. This relaxes unnecessary scale restrictions and gains more flexibility in regularizing CP-based PPCAs. To take full advantage of the probabilistic framework, we further propose a Bayesian treatment of PROTA, which achieves both automatic feature determination and robustness against overfitting. Experiments on synthetic and real-world datasets demonstrate the superiority of PROTA in subspace estimation and classification, as well as the effectiveness of CRs in alleviating overfitting.

Linear discriminant analysis (LDA) is a classical supervised subspace learning technique that has wide applications. However, it is designed for vector only, which cannot exploit the tensor structures and may lead to suboptimal results when dealing with tensorial data. To address this problem, several multilinear LDA (MLDA) methods have been proposed to learn the subspaces from tensors. By exploiting the tensor structures, they achieve compact subspace representations, reduced parameter sizes, and improved robustness against the small sample size problem. However, existing MLDA methods do not take data uncertainty into account, fail to converge properly, or have to introduce additional tuning parameters for good convergence properties. In this paper, we therefore solve these limitations by proposing a probabilistic MLDA method for matrix inputs. Specifically, we propose a new generative model to incorporate structural information into the probabilistic framework, where each observed matrix is represented as a linear combination of collective and individual rank-one matrices. This provides our method with both the expressiveness of capturing discriminative features and nondiscriminative noise, and the capability of exploiting the 2-D tensor structures. To overcome the convergence problem of existing MLDAs, we develop an EM-type algorithm for parameter estimation, which has closed-form solutions with convergence guarantees. Experimental results on real-world datasets show the superiority of the proposed method to other probabilistic and MLDA variants.

Canonical Correlation Analysis (CCA) is a classical technique for two-view correlation analysis, while Probabilistic CCA (PCCA) provides a generative and more general viewpoint for this task. Recently, PCCA has been extended to bilinear cases for dealing with two-view matrices in order to preserve and exploit the matrix structures in PCCA. However, existing bilinear PCCAs impose restrictive model assumptions for matrix structure preservation, sacrificing generative correctness or model flexibility. To overcome these drawbacks, we propose BPCCA, a new bilinear extension of PCCA, by introducing a hybrid joint model. Our new model preserves matrix structures indirectly via hybrid vector-based and matrix-based concatenations. This enables BPCCA to gain more model flexibility in capturing two-view correlations and obtain close-form solutions in parameter estimation. Experimental results on two real-world applications demonstrate the superior performance of BPCCA over competing methods.

We propose a novel algorithm for supervised dimensionality reduction named manifold partition discriminant analysis (MPDA). It aims to find a linear embedding space where the within-class similarity is achieved along the direction that is consistent with the local variation of the data manifold, while nearby data belonging to different classes are well separated. By partitioning the data manifold into a number of linear subspaces and utilizing the first-order Taylor expansion, MPDA explicitly parameterizes the connections of tangent spaces and represents the data manifold in a piecewise manner. While graph Laplacian methods capture only the pairwise interaction between data points, our method captures both pairwise and higher order interactions (using regional consistency) between data points. This manifold representation can help to improve the measure of within-class similarity, which further leads to improved performance of dimensionality reduction. Experimental results on multiple real-world data sets demonstrate the effectiveness of the proposed method.

Knowing how hands move and what object is being manipulated are two key sub-tasks for analyzing first-person (egocentric) action. However, lack of fully annotated hand data as well as imprecise foreground segmentation make either sub-task challenging. This work aims to explicitly ad dress these two issues via introducing a cascaded interactional targeting (i.e., infer both hand and active object regions) deep neural network. Firstly, a novel EM-like learning framework is proposed to train the pixel-level deep convolutional neural network (DCNN) by seamlessly integrating weakly supervised data (i.e., massive bounding box annotations) with a small set of strongly supervised data (i.e., fully annotated hand segmentation maps) to achieve state-of-the-art hand segmentation performance. Secondly, the resulting high-quality hand segmentation maps are further paired with the corresponding motion maps and object feature maps, in order to explore the contextual information among object, motion and hand to generate interactional foreground regions (operated objects). The resulting interactional target maps (hand + active object) from our cascaded DCNN are further utilized to form discriminative action representation. Experiments show that our framework has achieved the state-of-the-art egocentric action recognition performance on the benchmark dataset Activities of Daily Living (ADL).

Modeling human-object interactions and manipulating motions lies in the heart of fine-grained action recognition. Previous methods heavily rely on explicit detection of the object being interacted, which requires intensive human labour on object annotation. To bypass this constraint and achieve better classification performance, in this work, we propose a novel fine-grained action recognition pipeline by interaction part proposal and discriminative mid-level part mining. Firstly, we generate a large number of candidate object regions using off-the-shelf object proposal tool, e.g., BING. Secondly, these object regions are matched and tracked across frames to form a large spatio-temporal graph based on the appearance matching and the dense motion trajectories through them. We then propose an efficient approximate graph segmentation algorithm to partition and filter the graph into consistent local dense sub-graphs. These sub-graphs, which are spatio-temporal sub-volumes, represent our candidate interaction parts. Finally, we mine discriminative mid-level part detectors from the features computed over the candidate interaction parts. Bag-of-detection scores based on a novel Max-N pooling scheme are computed as the action representation for a video sample. We conduct extensive experiments on human-object interaction datasets including MPII Cooking and MSR Daily Activity 3D. The experimental results demonstrate that the proposed framework achieves consistent improvements over the state-of-the-art action recognition accuracies on the benchmarks, without using any object annotation.

The performance of current image retrieval system is largely determined by the quality and discriminative capability of features. Therefore, using what features and how to effectively combine the power of appropriate features are important in the system. We adopt the reciprocal neighbor based graph fusion approach for feature fusion. More importantly, we explicitly augment the original approach with the following two strategies: 1) we investigate the most suitable feature combinations on various datasets, including the deep learning feature, which has been popular for image retrieval recently; 2) we further improve the robustness of original graph fusion approach by the SVM prediction strategy.