Hoifung Poon

Renqian Luo, Liai Sun, Yingce Xia et al. · microsoft-research

Pre-trained language models have attracted increasing attention in the biomedical domain, inspired by their great success in the general natural language domain. Among the two main branches of pre-trained language models in the general language domain, i.e., BERT (and its variants) and GPT (and its variants), the first one has been extensively studied in the biomedical domain, such as BioBERT and PubMedBERT. While they have achieved great success on a variety of discriminative downstream biomedical tasks, the lack of generation ability constrains their application scope. In this paper, we propose BioGPT, a domain-specific generative Transformer language model pre-trained on large scale biomedical literature. We evaluate BioGPT on six biomedical NLP tasks and demonstrate that our model outperforms previous models on most tasks. Especially, we get 44.98%, 38.42% and 40.76% F1 score on BC5CDR, KD-DTI and DDI end-to-end relation extraction tasks respectively, and 78.2% accuracy on PubMedQA, creating a new record. Our case study on text generation further demonstrates the advantage of BioGPT on biomedical literature to generate fluent descriptions for biomedical terms. Code is available at https://github.com/microsoft/BioGPT.

Sheng Zhang, Hao Cheng, Jianfeng Gao et al. · microsoft-research

We present a bi-encoder framework for named entity recognition (NER), which applies contrastive learning to map candidate text spans and entity types into the same vector representation space. Prior work predominantly approaches NER as sequence labeling or span classification. We instead frame NER as a representation learning problem that maximizes the similarity between the vector representations of an entity mention and its type. This makes it easy to handle nested and flat NER alike, and can better leverage noisy self-supervision signals. A major challenge to this bi-encoder formulation for NER lies in separating non-entity spans from entity mentions. Instead of explicitly labeling all non-entity spans as the same class $\texttt{Outside}$ ($\texttt{O}$) as in most prior methods, we introduce a novel dynamic thresholding loss. Experiments show that our method performs well in both supervised and distantly supervised settings, for nested and flat NER alike, establishing new state of the art across standard datasets in the general domain (e.g., ACE2004, ACE2005) and high-value verticals such as biomedicine (e.g., GENIA, NCBI, BC5CDR, JNLPBA). We release the code at github.com/microsoft/binder.

Yiqing Xie, Sheng Zhang, Hao Cheng et al. · microsoft-research

Medical text generation aims to assist with administrative work and highlight salient information to support decision-making. To reflect the specific requirements of medical text, in this paper, we propose a set of metrics to evaluate the completeness, conciseness, and attribution of the generated text at a fine-grained level. The metrics can be computed by various types of evaluators including instruction-following (both proprietary and open-source) and supervised entailment models. We demonstrate the effectiveness of the resulting framework, DocLens, with three evaluators on three tasks: clinical note generation, radiology report summarization, and patient question summarization. A comprehensive human study shows that DocLens exhibits substantially higher agreement with the judgments of medical experts than existing metrics. The results also highlight the need to improve open-source evaluators and suggest potential directions.

Chunyuan Li, Cliff Wong, Sheng Zhang et al. · cambridge, microsoft-research

Conversational generative AI has demonstrated remarkable promise for empowering biomedical practitioners, but current investigations focus on unimodal text. Multimodal conversational AI has seen rapid progress by leveraging billions of image-text pairs from the public web, but such general-domain vision-language models still lack sophistication in understanding and conversing about biomedical images. In this paper, we propose a cost-efficient approach for training a vision-language conversational assistant that can answer open-ended research questions of biomedical images. The key idea is to leverage a large-scale, broad-coverage biomedical figure-caption dataset extracted from PubMed Central, use GPT-4 to self-instruct open-ended instruction-following data from the captions, and then fine-tune a large general-domain vision-language model using a novel curriculum learning method. Specifically, the model first learns to align biomedical vocabulary using the figure-caption pairs as is, then learns to master open-ended conversational semantics using GPT-4 generated instruction-following data, broadly mimicking how a layperson gradually acquires biomedical knowledge. This enables us to train a Large Language and Vision Assistant for BioMedicine (LLaVA-Med) in less than 15 hours (with eight A100s). LLaVA-Med exhibits excellent multimodal conversational capability and can follow open-ended instruction to assist with inquiries about a biomedical image. On three standard biomedical visual question answering datasets, LLaVA-Med outperforms previous supervised state-of-the-art on certain metrics. To facilitate biomedical multimodal research, we will release our instruction-following data and the LLaVA-Med model.

Sheng Zhang, Yanbo Xu, Naoto Usuyama et al. · cambridge, microsoft-research

Biomedical data is inherently multimodal, comprising physical measurements and natural language narratives. A generalist biomedical AI model needs to simultaneously process different modalities of data, including text and images. Therefore, training an effective generalist biomedical model requires high-quality multimodal data, such as parallel image-text pairs. Here, we present PMC-15M, a novel dataset that is two orders of magnitude larger than existing biomedical multimodal datasets such as MIMIC-CXR, and spans a diverse range of biomedical image types. PMC-15M contains 15 million biomedical image-text pairs collected from 4.4 million scientific articles. Based on PMC-15M, we have pretrained BiomedCLIP, a multimodal foundation model, with domain-specific adaptations tailored to biomedical vision-language processing. We conducted extensive experiments and ablation studies on standard biomedical imaging tasks from retrieval to classification to visual question-answering (VQA). BiomedCLIP achieved new state-of-the-art results in a wide range of standard datasets, substantially outperforming prior approaches. Intriguingly, by large-scale pretraining on diverse biomedical image types, BiomedCLIP even outperforms state-of-the-art radiology-specific models such as BioViL in radiology-specific tasks such as RSNA pneumonia detection. In summary, BiomedCLIP is a fully open-access foundation model that achieves state-of-the-art performance on various biomedical tasks, paving the way for transformative multimodal biomedical discovery and applications. We release our models at https://aka.ms/biomedclip to facilitate future research in multimodal biomedical AI.

Harsha Nori, Yin Tat Lee, Sheng Zhang et al. · microsoft-research

Generalist foundation models such as GPT-4 have displayed surprising capabilities in a wide variety of domains and tasks. Yet, there is a prevalent assumption that they cannot match specialist capabilities of fine-tuned models. For example, most explorations to date on medical competency benchmarks have leveraged domain-specific training, as exemplified by efforts on BioGPT and Med-PaLM. We build on a prior study of GPT-4's capabilities on medical challenge benchmarks in the absence of special training. Rather than using simple prompting to highlight the model's out-of-the-box capabilities, we perform a systematic exploration of prompt engineering. We find that prompting innovation can unlock deeper specialist capabilities and show that GPT-4 easily tops prior leading results for medical benchmarks. The prompting methods we explore are general purpose, and make no specific use of domain expertise, removing the need for expert-curated content. Our experimental design carefully controls for overfitting during the prompt engineering process. We introduce Medprompt, based on a composition of several prompting strategies. With Medprompt, GPT-4 achieves state-of-the-art results on all nine of the benchmark datasets in the MultiMedQA suite. The method outperforms leading specialist models such as Med-PaLM 2 by a significant margin with an order of magnitude fewer calls to the model. Steering GPT-4 with Medprompt achieves a 27% reduction in error rate on the MedQA dataset over the best methods to date achieved with specialist models and surpasses a score of 90% for the first time. Beyond medical problems, we show the power of Medprompt to generalize to other domains and provide evidence for the broad applicability of the approach via studies of the strategy on exams in electrical engineering, machine learning, philosophy, accounting, law, nursing, and clinical psychology.

Benedikt Boecking, Naoto Usuyama, Shruthi Bannur et al. · cambridge, microsoft-research

Multi-modal data abounds in biomedicine, such as radiology images and reports. Interpreting this data at scale is essential for improving clinical care and accelerating clinical research. Biomedical text with its complex semantics poses additional challenges in vision--language modelling compared to the general domain, and previous work has used insufficiently adapted models that lack domain-specific language understanding. In this paper, we show that principled textual semantic modelling can substantially improve contrastive learning in self-supervised vision--language processing. We release a language model that achieves state-of-the-art results in radiology natural language inference through its improved vocabulary and novel language pretraining objective leveraging semantics and discourse characteristics in radiology reports. Further, we propose a self-supervised joint vision--language approach with a focus on better text modelling. It establishes new state of the art results on a wide range of publicly available benchmarks, in part by leveraging our new domain-specific language model. We release a new dataset with locally-aligned phrase grounding annotations by radiologists to facilitate the study of complex semantic modelling in biomedical vision--language processing. A broad evaluation, including on this new dataset, shows that our contrastive learning approach, aided by textual-semantic modelling, outperforms prior methods in segmentation tasks, despite only using a global-alignment objective.

Wenxuan Zhou, Sheng Zhang, Yu Gu et al. · microsoft-research

Large language models (LLMs) have demonstrated remarkable generalizability, such as understanding arbitrary entities and relations. Instruction tuning has proven effective for distilling LLMs into more cost-efficient models such as Alpaca and Vicuna. Yet such student models still trail the original LLMs by large margins in downstream applications. In this paper, we explore targeted distillation with mission-focused instruction tuning to train student models that can excel in a broad application class such as open information extraction. Using named entity recognition (NER) for case study, we show how ChatGPT can be distilled into much smaller UniversalNER models for open NER. For evaluation, we assemble the largest NER benchmark to date, comprising 43 datasets across 9 diverse domains such as biomedicine, programming, social media, law, finance. Without using any direct supervision, UniversalNER attains remarkable NER accuracy across tens of thousands of entity types, outperforming general instruction-tuned models such as Alpaca and Vicuna by over 30 absolute F1 points in average. With a tiny fraction of parameters, UniversalNER not only acquires ChatGPT's capability in recognizing arbitrary entity types, but also outperforms its NER accuracy by 7-9 absolute F1 points in average. Remarkably, UniversalNER even outperforms by a large margin state-of-the-art multi-task instruction-tuned systems such as InstructUIE, which uses supervised NER examples. We also conduct thorough ablation studies to assess the impact of various components in our distillation approach. We release the distillation recipe, data, and UniversalNER models to facilitate future research on targeted distillation.

Wenxuan Zhou, Sheng Zhang, Hoifung Poon et al.

Large language models (LLMs) encode parametric knowledge about world facts and have shown remarkable performance in knowledge-driven NLP tasks. However, their reliance on parametric knowledge may cause them to overlook contextual cues, leading to incorrect predictions in context-sensitive NLP tasks (e.g., knowledge acquisition tasks). In this paper, we seek to assess and enhance LLMs' contextual faithfulness in two aspects: knowledge conflict and prediction with abstention. We demonstrate that LLMs' faithfulness can be significantly improved using carefully designed prompting strategies. In particular, we identify opinion-based prompts and counterfactual demonstrations as the most effective methods. Opinion-based prompts reframe the context as a narrator's statement and inquire about the narrator's opinions, while counterfactual demonstrations use instances containing false facts to improve faithfulness in knowledge conflict situations. Neither technique requires additional training. We conduct experiments on three datasets of two standard NLP tasks, machine reading comprehension and relation extraction, and the results demonstrate significant improvement in faithfulness to contexts. Code and data are released at https://github.com/wzhouad/context-faithful-llm.

Qianchu Liu, Stephanie Hyland, Shruthi Bannur et al. · cambridge, microsoft-research

The recent success of general-domain large language models (LLMs) has significantly changed the natural language processing paradigm towards a unified foundation model across domains and applications. In this paper, we focus on assessing the performance of GPT-4, the most capable LLM so far, on the text-based applications for radiology reports, comparing against state-of-the-art (SOTA) radiology-specific models. Exploring various prompting strategies, we evaluated GPT-4 on a diverse range of common radiology tasks and we found GPT-4 either outperforms or is on par with current SOTA radiology models. With zero-shot prompting, GPT-4 already obtains substantial gains ($\approx$ 10% absolute improvement) over radiology models in temporal sentence similarity classification (accuracy) and natural language inference ($F_1$). For tasks that require learning dataset-specific style or schema (e.g. findings summarisation), GPT-4 improves with example-based prompting and matches supervised SOTA. Our extensive error analysis with a board-certified radiologist shows GPT-4 has a sufficient level of radiology knowledge with only occasional errors in complex context that require nuanced domain knowledge. For findings summarisation, GPT-4 outputs are found to be overall comparable with existing manually-written impressions.

Cliff Wong, Sheng Zhang, Yu Gu et al. · cambridge, microsoft-research

Clinical trial matching is a key process in health delivery and discovery. In practice, it is plagued by overwhelming unstructured data and unscalable manual processing. In this paper, we conduct a systematic study on scaling clinical trial matching using large language models (LLMs), with oncology as the focus area. Our study is grounded in a clinical trial matching system currently in test deployment at a large U.S. health network. Initial findings are promising: out of box, cutting-edge LLMs, such as GPT-4, can already structure elaborate eligibility criteria of clinical trials and extract complex matching logic (e.g., nested AND/OR/NOT). While still far from perfect, LLMs substantially outperform prior strong baselines and may serve as a preliminary solution to help triage patient-trial candidates with humans in the loop. Our study also reveals a few significant growth areas for applying LLMs to end-to-end clinical trial matching, such as context limitation and accuracy, especially in structuring patient information from longitudinal medical records.

Yu Gu, Jianwei Yang, Naoto Usuyama et al. · microsoft-research

Rapid progress has been made in instruction-learning for image editing with natural-language instruction, as exemplified by InstructPix2Pix. In biomedicine, such methods can be applied to counterfactual image generation, which helps differentiate causal structure from spurious correlation and facilitate robust image interpretation for disease progression modeling. However, generic image-editing models are ill-suited for the biomedical domain, and counterfactual biomedical image generation is largely underexplored. In this paper, we present BiomedJourney, a novel method for counterfactual biomedical image generation by instruction-learning from multimodal patient journeys. Given a patient with two biomedical images taken at different time points, we use GPT-4 to process the corresponding imaging reports and generate a natural language description of disease progression. The resulting triples (prior image, progression description, new image) are then used to train a latent diffusion model for counterfactual biomedical image generation. Given the relative scarcity of image time series data, we introduce a two-stage curriculum that first pretrains the denoising network using the much more abundant single image-report pairs (with dummy prior image), and then continues training using the counterfactual triples. Experiments using the standard MIMIC-CXR dataset demonstrate the promise of our method. In a comprehensive battery of tests on counterfactual medical image generation, BiomedJourney substantially outperforms prior state-of-the-art methods in instruction image editing and medical image generation such as InstructPix2Pix and RoentGen. To facilitate future study in counterfactual medical generation, we plan to release our instruction-learning code and pretrained models.

Yu Gu, Sheng Zhang, Naoto Usuyama et al. · cambridge, microsoft-research

Large language models (LLMs), such as GPT-4, have demonstrated remarkable capabilities across a wide range of tasks, including health applications. In this paper, we study how LLMs can be used to scale biomedical knowledge curation. We find that while LLMs already possess decent competency in structuring biomedical text, by distillation into a task-specific student model through self-supervised learning, substantial gains can be attained over out-of-box LLMs, with additional advantages such as cost, efficiency, and white-box model access. We conduct a case study on adverse drug event (ADE) extraction, which is an important area for improving care. On standard ADE extraction evaluation, a GPT-3.5 distilled PubMedBERT model attained comparable accuracy as supervised state-of-the-art models without using any labeled data. Despite being over 1,000 times smaller, the distilled model outperformed its teacher GPT-3.5 by over 6 absolute points in F1 and GPT-4 by over 5 absolute points. Ablation studies on distillation model choice (e.g., PubMedBERT vs BioGPT) and ADE extraction architecture shed light on best practice for biomedical knowledge extraction. Similar gains were attained by distillation for other standard biomedical knowledge extraction tasks such as gene-disease associations and protected health information, further illustrating the promise of this approach.

Wenxuan Zhou, Sheng Zhang, Tristan Naumann et al. · cambridge, microsoft-research

Relation extraction (RE), which has relied on structurally annotated corpora for model training, has been particularly challenging in low-resource scenarios and domains. Recent literature has tackled low-resource RE by self-supervised learning, where the solution involves pretraining the entity pair embedding by RE-based objective and finetuning on labeled data by classification-based objective. However, a critical challenge to this approach is the gap in objectives, which prevents the RE model from fully utilizing the knowledge in pretrained representations. In this paper, we aim at bridging the gap and propose to pretrain and finetune the RE model using consistent objectives of contrastive learning. Since in this kind of representation learning paradigm, one relation may easily form multiple clusters in the representation space, we further propose a multi-center contrastive loss that allows one relation to form multiple clusters to better align with pretraining. Experiments on two document-level RE datasets, BioRED and Re-DocRED, demonstrate the effectiveness of our method. Particularly, when using 1% end-task training data, our method outperforms PLM-based RE classifier by 10.5% and 6.1% on the two datasets, respectively.

Nan Xu, Fei Wang, Sheng Zhang et al. · microsoft-research

Motivated by in-context learning (ICL) capabilities of Large Language Models (LLMs), multimodal LLMs with additional visual modality are also exhibited with similar ICL abilities when multiple image-text pairs are provided as demonstrations. However, relatively less work has been done to investigate the principles behind how and why multimodal ICL works. We conduct a systematic and principled evaluation of multimodal ICL for models of different scales on a broad spectrum of new yet critical tasks. Through perturbations over different modality information, we show that modalities matter differently across tasks in multimodal ICL. Guided by task-specific modality impact, we recommend modality-driven demonstration strategies to boost ICL performance. We also find that models may follow inductive biases from multimodal ICL even if they are rarely seen in or contradict semantic priors from pretraining data. Our principled analysis provides a comprehensive way of understanding the role of demonstrations in multimodal in-context learning, and sheds light on effectively improving multimodal ICL on a wide range of tasks.

Fangyu Liu, Qianchu Liu, Shruthi Bannur et al. · cambridge, deepmind

Label scarcity is a bottleneck for improving task performance in specialised domains. We propose a novel compositional transfer learning framework (DoT5 - domain compositional zero-shot T5) for zero-shot domain transfer. Without access to in-domain labels, DoT5 jointly learns domain knowledge (from MLM of unlabelled in-domain free text) and task knowledge (from task training on more readily available general-domain data) in a multi-task manner. To improve the transferability of task training, we design a strategy named NLGU: we simultaneously train NLG for in-domain label-to-data generation which enables data augmentation for self-finetuning and NLU for label prediction. We evaluate DoT5 on the biomedical domain and the resource-lean subdomain of radiology, focusing on NLI, text summarisation and embedding learning. DoT5 demonstrates the effectiveness of compositional transfer learning through multi-task learning. In particular, DoT5 outperforms the current SOTA in zero-shot transfer by over 7 absolute points in accuracy on RadNLI. We validate DoT5 with ablations and a case study demonstrating its ability to solve challenging NLI examples requiring in-domain expertise.

Sam Preston, Mu Wei, Rajesh Rao et al. · cambridge, microsoft-research

Objective: The majority of detailed patient information in real-world data (RWD) is only consistently available in free-text clinical documents. Manual curation is expensive and time-consuming. Developing natural language processing (NLP) methods for structuring RWD is thus essential for scaling real-world evidence generation. Materials and Methods: Traditional rule-based systems are vulnerable to the prevalent linguistic variations and ambiguities in clinical text, and prior applications of machine-learning methods typically require sentence-level or report-level labeled examples that are hard to produce at scale. We propose leveraging patient-level supervision from medical registries, which are often readily available and capture key patient information, for general RWD applications. To combat the lack of sentence-level or report-level annotations, we explore advanced deep-learning methods by combining domain-specific pretraining, recurrent neural networks, and hierarchical attention. Results: We conduct an extensive study on 135,107 patients from the cancer registry of a large integrated delivery network (IDN) comprising healthcare systems in five western US states. Our deep learning methods attain test AUROC of 94-99% for key tumor attributes and comparable performance on held-out data from separate health systems and states. Discussion and Conclusion: Ablation results demonstrate clear superiority of these advanced deep-learning methods over prior approaches. Error analysis shows that our NLP system sometimes even corrects errors in registrar labels. We also conduct a preliminary investigation in accelerating registry curation and general RWD structuring via assisted curation for over 1.2 million cancer patients in this healthcare network.

Cai Yang, Addie Woicik, Hoifung Poon et al. · stanford

Language models pre-trained on scientific literature corpora have substantially advanced scientific discovery by offering high-quality feature representations for downstream applications. However, these features are often not interpretable, and thus can reveal limited insights to domain experts. Instead of obtaining features from language models, we propose BLIAM, a literature-based data synthesis approach to directly generate training data points that are interpretable and model-agnostic to downstream applications. The key idea of BLIAM is to create prompts using existing training data and then use these prompts to synthesize new data points. BLIAM performs these two steps iteratively as new data points will define more informative prompts and new prompts will in turn synthesize more accurate data points. Notably, literature-based data augmentation might introduce data leakage since labels of test data points in downstream applications might have already been mentioned in the language model corpus. To prevent such leakage, we introduce GDSC-combo, a large-scale drug combination discovery dataset that was published after the biomedical language model was trained. We found that BLIAM substantially outperforms a non-augmented approach and manual prompting in this rigorous data split setting. BLIAM can be further used to synthesize data points for novel drugs and cell lines that were not even measured in biomedical experiments. In addition to the promising prediction performance, the data points synthesized by BLIAM are interpretable and model-agnostic, enabling in silico augmentation for in vitro experiments.

Qianchu Liu, Sheng Zhang, Guanghui Qin et al. · microsoft-research

Frontier models have demonstrated remarkable capabilities in understanding and reasoning with natural-language text, but they still exhibit major competency gaps in multimodal understanding and reasoning especially in high-value verticals such as biomedicine. Medical imaging report generation is a prominent example. Supervised fine-tuning can substantially improve performance, but they are prone to overfitting to superficial boilerplate patterns. In this paper, we introduce Universal Report Generation (UniRG) as a general framework for medical imaging report generation. By leveraging reinforcement learning as a unifying mechanism to directly optimize for evaluation metrics designed for end applications, UniRG can significantly improve upon supervised fine-tuning and attain durable generalization across diverse institutions and clinical practices. We trained UniRG-CXR on publicly available chest X-ray (CXR) data and conducted a thorough evaluation in CXR report generation with rigorous evaluation scenarios. On the authoritative ReXrank benchmark, UniRG-CXR sets new overall SOTA, outperforming prior state of the art by a wide margin.

Theodore Zhengde Zhao, Sid Kiblawi, Jianwei Yang et al.

Self-supervised learning (SSL) faces a fundamental conflict between semantic understanding and image reconstruction. High-level semantic SSL (e.g., DINO) relies on global tokens that are forced to be location-invariant for augmentation alignment, a process that inherently discards the spatial coordinates required for reconstruction. Conversely, generative SSL (e.g., MAE) preserves dense feature grids for reconstruction but fails to produce high-level abstractions. We introduce STELLAR, a framework that resolves this tension by factorizing visual features into a low-rank product of semantic concepts and their spatial distributions. This disentanglement allows us to perform DINO-style augmentation alignment on the semantic tokens while maintaining the precise spatial mapping in the localization matrix necessary for pixel-level reconstruction. We demonstrate that as few as 16 sparse tokens under this factorized form are sufficient to simultaneously support high-quality reconstruction (2.60 FID) and match the semantic performance of dense backbones (79.10% ImageNet accuracy). Our results highlight STELLAR as a versatile sparse representation that bridges the gap between discriminative and generative vision by strategically separating semantic identity from spatial geometry. Code available at https://aka.ms/stellar.

Theodore Zhao, Mu Wei, J. Samuel Preston et al.

Large Language Models (LLMs) have shown impressive abilities in many applications. When a concrete and precise answer is desired, it is important to have a quantitative estimation of the potential error rate. However, this can be challenging due to the text-in-text-out nature of generative models. We present a method based on Pareto optimization that generates a risk score to estimate the probability of error in an LLM response by integrating multiple sources of information. We prove theoretically that the error estimator optimized in our framework aligns with the LLM and the information sources in an Pareto optimal manner. Experimental results show that the risk scores estimated by our method are well correlated with the true LLM error rate, thus facilitating error correction. By dynamically combining with prompting strategies such as self-verification and information retrieval, we demonstrate the proposed method can be utilized to increase the performance of an LLM, surpassing state-of-the-art task specific models.

Javier González, Risa Ueno, Cliff Wong et al.

The rapid digitization of real-world data presents an unprecedented opportunity to optimize healthcare delivery and accelerate biomedical discovery. However, these data are often found in unstructured forms such as clinical notes in electronic medical records (EMRs), and is typically plagued by confounders, making it challenging to generate robust real-world evidence (RWE). Therefore, we present TRIALSCOPE, a framework designed to distil RWE from population level observational data at scale. TRIALSCOPE leverages biomedical language models to structure clinical text at scale, employs advanced probabilistic modeling for denoising and imputation, and incorporates state-of-the-art causal inference techniques to address common confounders in treatment effect estimation. Extensive experiments were conducted on a large-scale dataset of over one million cancer patients from a single large healthcare network in the United States. TRIALSCOPE was shown to automatically curate high-quality structured patient data, expanding the dataset and incorporating key patient attributes only available in unstructured form. The framework reduces confounding in treatment effect estimation, generating comparable results to randomized controlled lung cancer trials. Additionally, we demonstrate simulations of unconducted clinical trials - including a pancreatic cancer trial with varying eligibility criteria - using a suite of validation tests to ensure robustness. Thorough ablation studies were conducted to better understand key components of TRIALSCOPE and establish best practices for RWE generation from EMRs. TRIALSCOPE was able to extract data cancer treatment data from EMRs, overcoming limitations of manual curation. We were also able to show that TRIALSCOPE could reproduce results of lung and pancreatic cancer clinical trials from the extracted real world data.

Tinghui Zhu, Sheng Zhang, James Y. Huang et al.

Video diffusion models have made rapid progress in perceptual realism and temporal coherence, but they remain primarily optimized for plausible generation rather than verifiable reasoning. This limitation is especially pronounced in tasks where generated videos must satisfy explicit spatial, temporal, or logical constraints. Inspired by the role of reinforcement learning with verifiable rewards (RLVR) in reasoning-oriented language models, we introduce VideoRLVR, a practical recipe for optimizing video diffusion models with rule-based feedback. VideoRLVR formulates video reasoning as the generation of verifiable visual trajectories and consists of an SDE-GRPO optimization backbone, dense decomposed rewards, and an Early-Step Focus strategy for efficient training. The Early-Step Focus strategy restricts policy optimization to the early denoising phase, reducing training latency by about 40% while preserving performance. We evaluate VideoRLVR on Maze, FlowFree, and Sokoban, three procedurally generated domains with objective success criteria. Across these tasks, VideoRLVR consistently improves over supervised fine-tuning baselines, with dense decomposed rewards proving especially important in low-success-rate settings. Our RL-optimized model also outperforms the evaluated proprietary and open-source video generation models on these verifiable reasoning benchmarks and out-of-domain benchmarks. These results suggest that verifiable RL can move video models beyond perceptual imitation toward more reliable rule-consistent visual reasoning.

Juan Manuel Zambrano Chaves, Shih-Cheng Huang, Yanbo Xu et al. · microsoft-research

The scaling laws and extraordinary performance of large foundation models motivate the development and utilization of such models in biomedicine. However, despite early promising results on some biomedical benchmarks, there are still major challenges that need to be addressed before these models can be used in real-world clinics. Frontier general-domain models such as GPT-4V still have significant performance gaps in multimodal biomedical applications. More importantly, less-acknowledged pragmatic issues, including accessibility, model cost, and tedious manual evaluation make it hard for clinicians to use state-of-the-art large models directly on private patient data. Here, we explore training open-source small multimodal models (SMMs) to bridge competency gaps for unmet clinical needs in radiology. To maximize data efficiency, we adopt a modular approach by incorporating state-of-the-art pre-trained models for image and text modalities, and focusing on training a lightweight adapter to ground each modality to the text embedding space, as exemplified by LLaVA-Med. For training, we assemble a large dataset of over 697 thousand radiology image-text pairs. For evaluation, we propose CheXprompt, a GPT-4-based metric for factuality evaluation, and demonstrate its parity with expert evaluation. For best practice, we conduct a systematic ablation study on various choices in data engineering and multimodal training. The resulting LlaVA-Rad (7B) model attains state-of-the-art results on standard radiology tasks such as report generation and cross-modal retrieval, even outperforming much larger models such as GPT-4V and Med-PaLM M (84B). The inference of LlaVA-Rad is fast and can be performed on a single V100 GPU in private settings, offering a promising state-of-the-art tool for real-world clinical applications.

Qianchu Liu, Sheng Zhang, Guanghui Qin et al. · microsoft-research

Recent proprietary models (e.g., o3) have begun to demonstrate strong multimodal reasoning capabilities. Yet, most existing open-source research concentrates on training text-only reasoning models, with evaluations limited to mainly mathematical and general-domain tasks. Therefore, it remains unclear how to effectively extend reasoning capabilities beyond text input and general domains. This paper explores a fundamental research question: Is reasoning generalizable across modalities and domains? Our findings support an affirmative answer: General-domain text-based post-training can enable such strong generalizable reasoning. Leveraging this finding, we introduce X-Reasoner, a vision-language model post-trained solely on general-domain text for generalizable reasoning, using a two-stage approach: an initial supervised fine-tuning phase with distilled long chain-of-thoughts, followed by reinforcement learning with verifiable rewards. Experiments show that X-Reasoner successfully transfers reasoning capabilities to both multimodal and out-of-domain settings, outperforming existing state-of-the-art models trained with in-domain and multimodal data across various general and medical benchmarks (Figure 1). Additionally, we find that X-Reasoner's performance in specialized domains can be further enhanced through continued training on domain-specific text-only data. Building upon this, we introduce X-Reasoner-Med, a medical-specialized variant that achieves new state of the art on numerous text-only and multimodal medical benchmarks.

Zelalem Gero, Chandan Singh, Yiqing Xie et al. · microsoft-research

Summarizing clinical text is crucial in health decision-support and clinical research. Large language models (LLMs) have shown the potential to generate accurate clinical text summaries, but still struggle with issues regarding grounding and evaluation, especially in safety-critical domains such as health. Holistically evaluating text summaries is challenging because they may contain unsubstantiated information. Here, we explore a general mitigation framework using Attribute Structuring (AS), which structures the summary evaluation process. It decomposes the evaluation process into a grounded procedure that uses an LLM for relatively simple structuring and scoring tasks, rather than the full task of holistic summary evaluation. Experiments show that AS consistently improves the correspondence between human annotations and automated metrics in clinical text summarization. Additionally, AS yields interpretations in the form of a short text span corresponding to each output, which enables efficient human auditing, paving the way towards trustworthy evaluation of clinical information in resource-constrained scenarios. We release our code, prompts, and an open-source benchmark at https://github.com/microsoft/attribute-structuring.

Timothy Ossowski, Sheng Zhang, Qianchu Liu et al.

High-quality and carefully curated data is a cornerstone of training medical large language models, as it directly impacts both generalization and robustness to unseen clinical tasks. We investigate strategies for training and data curation to develop a robust multimodal reasoning model in the medical domain. Our work focuses on supervised fine-tuning (SFT) and explores data recipes that leverage structured reasoning traces. Using our proposed data recipe, we scale experiments to a dataset of over 8 million examples and 6.8 billion response tokens, achieving state-of-the-art performance among open-source models across diverse out-of-distribution medical benchmark tasks. Our results further indicate that curating a high-quality, diverse training dataset with varying structured reasoning trace lengths enables the fine-tuned model to self-calibrate its reasoning trajectory lengths based on the downstream task, without explicit supervision. We present key insights, describe the data curation strategy, and outline next steps toward developing robust medical vision-language reasoning system.

James Y. Huang, Sheng Zhang, Qianchu Liu et al.

Large Language Models (LLMs) have demonstrated remarkable capabilities in challenging, knowledge-intensive reasoning tasks. However, extending LLMs to perceive and reason over a new modality (e.g., vision), often requires costly development of large-scale vision language models (VLMs) with LLMs as backbones. Smaller VLMs are more efficient and adaptable but often lack the broad knowledge and reasoning capabilities of frontier LLMs. In this work, we propose BeMyEyes, a modular, multi-agent framework for extending LLMs to multimodal reasoning by orchestrating collaboration between efficient, adaptable VLMs as perceivers and powerful LLMs as reasoners through conversations. We then introduce a data synthesis and supervised fine-tuning pipeline to train the perceiver agent to effectively collaborate with the reasoner agent. By combining the complementary strengths of perception and reasoning agents, BeMyEyes avoids the need for training large-scale multimodal models, preserves the generalization and reasoning capabilities of LLMs, and allows flexible extension to new domains and modalities. Experiments show that our framework unlocks the multimodal reasoning capabilities for LLMs, enabling a lightweight and fully open-source solution, i.e. equipping text-only DeepSeek-R1 with Qwen2.5-VL-7B perceiver, to outperform large-scale proprietary VLMs such as GPT-4o on a wide range of knowledge-intensive multimodal tasks. These results demonstrate the effectiveness, modularity, and scalability of our multi-agent approach for building future multimodal reasoning systems.

Theodore Zhao, Yu Gu, Jianwei Yang et al.

Biomedical image analysis is fundamental for biomedical discovery in cell biology, pathology, radiology, and many other biomedical domains. Holistic image analysis comprises interdependent subtasks such as segmentation, detection, and recognition of relevant objects. Here, we propose BiomedParse, a biomedical foundation model for imaging parsing that can jointly conduct segmentation, detection, and recognition for 82 object types across 9 imaging modalities. Through joint learning, we can improve accuracy for individual tasks and enable novel applications such as segmenting all relevant objects in an image through a text prompt, rather than requiring users to laboriously specify the bounding box for each object. We leveraged readily available natural-language labels or descriptions accompanying those datasets and use GPT-4 to harmonize the noisy, unstructured text information with established biomedical object ontologies. We created a large dataset comprising over six million triples of image, segmentation mask, and textual description. On image segmentation, we showed that BiomedParse is broadly applicable, outperforming state-of-the-art methods on 102,855 test image-mask-label triples across 9 imaging modalities (everything). On object detection, which aims to locate a specific object of interest, BiomedParse again attained state-of-the-art performance, especially on objects with irregular shapes (everywhere). On object recognition, which aims to identify all objects in a given image along with their semantic types, we showed that BiomedParse can simultaneously segment and label all biomedical objects in an image (all at once). In summary, BiomedParse is an all-in-one tool for biomedical image analysis by jointly solving segmentation, detection, and recognition for all major biomedical image modalities, paving the path for efficient and accurate image-based biomedical discovery.

Fei Wang, Xingyu Fu, James Y. Huang et al.

We introduce MuirBench, a comprehensive benchmark that focuses on robust multi-image understanding capabilities of multimodal LLMs. MuirBench consists of 12 diverse multi-image tasks (e.g., scene understanding, ordering) that involve 10 categories of multi-image relations (e.g., multiview, temporal relations). Comprising 11,264 images and 2,600 multiple-choice questions, MuirBench is created in a pairwise manner, where each standard instance is paired with an unanswerable variant that has minimal semantic differences, in order for a reliable assessment. Evaluated upon 20 recent multi-modal LLMs, our results reveal that even the best-performing models like GPT-4o and Gemini Pro find it challenging to solve MuirBench, achieving 68.0% and 49.3% in accuracy. Open-source multimodal LLMs trained on single images can hardly generalize to multi-image questions, hovering below 33.3% in accuracy. These results highlight the importance of MuirBench in encouraging the community to develop multimodal LLMs that can look beyond a single image, suggesting potential pathways for future improvements.

Yifeng Liu, Hanwen Xu, Tangqi Fang et al.

As a fundamental task in computational chemistry, retrosynthesis prediction aims to identify a set of reactants to synthesize a target molecule. Existing template-free approaches only consider the graph structures of the target molecule, which often cannot generalize well to rare reaction types and large molecules. Here, we propose T-Rex, a text-assisted retrosynthesis prediction approach that exploits pre-trained text language models, such as ChatGPT, to assist the generation of reactants. T-Rex first exploits ChatGPT to generate a description for the target molecule and rank candidate reaction centers based both the description and the molecular graph. It then re-ranks these candidates by querying the descriptions for each reactants and examines which group of reactants can best synthesize the target molecule. We observed that T-Rex substantially outperformed graph-based state-of-the-art approaches on two datasets, indicating the effectiveness of considering text information. We further found that T-Rex outperformed the variant that only use ChatGPT-based description without the re-ranking step, demonstrate how our framework outperformed a straightforward integration of ChatGPT and graph information. Collectively, we show that text generated by pre-trained language models can substantially improve retrosynthesis prediction, opening up new avenues for exploiting ChatGPT to advance computational chemistry. And the codes can be found at https://github.com/lauyikfung/T-Rex.

Lis Pereira, Xiaodong Liu, Hao Cheng et al.

We present a simple yet effective Targeted Adversarial Training (TAT) algorithm to improve adversarial training for natural language understanding. The key idea is to introspect current mistakes and prioritize adversarial training steps to where the model errs the most. Experiments show that TAT can significantly improve accuracy over standard adversarial training on GLUE and attain new state-of-the-art zero-shot results on XNLI. Our code will be released at: https://github.com/namisan/mt-dnn.

Chenyan Xiong, Zhenghao Liu, Si Sun et al.

Neural rankers based on deep pretrained language models (LMs) have been shown to improve many information retrieval benchmarks. However, these methods are affected by their the correlation between pretraining domain and target domain and rely on massive fine-tuning relevance labels. Directly applying pretraining methods to specific domains may result in suboptimal search quality because specific domains may have domain adaption problems, such as the COVID domain. This paper presents a search system to alleviate the special domain adaption problem. The system utilizes the domain-adaptive pretraining and few-shot learning technologies to help neural rankers mitigate the domain discrepancy and label scarcity problems. Besides, we also integrate dense retrieval to alleviate traditional sparse retrieval's vocabulary mismatch obstacle. Our system performs the best among the non-manual runs in Round 2 of the TREC-COVID task, which aims to retrieve useful information from scientific literature related to COVID-19. Our code is publicly available at https://github.com/thunlp/OpenMatch.

Xiaodong Liu, Hao Cheng, Pengcheng He et al.

Generalization and robustness are both key desiderata for designing machine learning methods. Adversarial training can enhance robustness, but past work often finds it hurts generalization. In natural language processing (NLP), pre-training large neural language models such as BERT have demonstrated impressive gain in generalization for a variety of tasks, with further improvement from adversarial fine-tuning. However, these models are still vulnerable to adversarial attacks. In this paper, we show that adversarial pre-training can improve both generalization and robustness. We propose a general algorithm ALUM (Adversarial training for large neural LangUage Models), which regularizes the training objective by applying perturbations in the embedding space that maximizes the adversarial loss. We present the first comprehensive study of adversarial training in all stages, including pre-training from scratch, continual pre-training on a well-trained model, and task-specific fine-tuning. ALUM obtains substantial gains over BERT on a wide range of NLP tasks, in both regular and adversarial scenarios. Even for models that have been well trained on extremely large text corpora, such as RoBERTa, ALUM can still produce significant gains from continual pre-training, whereas conventional non-adversarial methods can not. ALUM can be further combined with task-specific fine-tuning to attain additional gains. The ALUM code is publicly available at https://github.com/namisan/mt-dnn.

Xiaodong Liu, Yu Wang, Jianshu Ji et al.

We present MT-DNN, an open-source natural language understanding (NLU) toolkit that makes it easy for researchers and developers to train customized deep learning models. Built upon PyTorch and Transformers, MT-DNN is designed to facilitate rapid customization for a broad spectrum of NLU tasks, using a variety of objectives (classification, regression, structured prediction) and text encoders (e.g., RNNs, BERT, RoBERTa, UniLM). A unique feature of MT-DNN is its built-in support for robust and transferable learning using the adversarial multi-task learning paradigm. To enable efficient production deployment, MT-DNN supports multi-task knowledge distillation, which can substantially compress a deep neural model without significant performance drop. We demonstrate the effectiveness of MT-DNN on a wide range of NLU applications across general and biomedical domains. The software and pre-trained models will be publicly available at https://github.com/namisan/mt-dnn.

Timothy Ossowski, Xinchi Liu, Danyal Maqbool et al.

Clinical reasoning agents based on large language models (LLMs) aim to automate tasks such as intensive care unit (ICU) monitoring and patient state tracking from electronic health records (EHRs). Existing systems typically rely on manually curated clinical tools or skills for concepts such as sepsis detection and organ failure assessment. However, maintaining these tool libraries requires substantial expert effort, while zero-shot querying or code generation often produces inefficient and unreliable reasoning chains, especially under institution-specific clinical policies. We introduce CodeClinic, a benchmark built on MIMIC-IV for evaluating whether LLM agents can synthesize and compose reusable clinical skills instead of relying on fixed toolboxes. The benchmark contains two complementary tasks: longitudinal ICU surveillance and compositional information seeking. The longitudinal setting simulates monitoring patient trajectories with structured decisions every four hours across 25 findings and eight clinical families, while the compositional setting spans 63k instances across 259 tasks in nine domains and is stratified by compositional dependency depth to evaluate increasingly complex multi-step reasoning. We further propose an offline autoformalization pipeline that converts natural-language clinical guidelines into reusable and verified Python skill libraries through iterative LLM refinement. Compared with zero-shot code generation, the resulting libraries improve consistency while reducing per-query token usage by up to 40%.

Ho Hin Lee, Yu Gu, Theodore Zhao et al.

Recent advancements in biomedical image analysis have been significantly driven by the Segment Anything Model (SAM). This transformative technology, originally developed for general-purpose computer vision, has found rapid application in medical image processing. Within the last year, marked by over 100 publications, SAM has demonstrated its prowess in zero-shot learning adaptations for medical imaging. The fundamental premise of SAM lies in its capability to segment or identify objects in images without prior knowledge of the object type or imaging modality. This approach aligns well with tasks achievable by the human visual system, though its application in non-biological vision contexts remains more theoretically challenging. A notable feature of SAM is its ability to adjust segmentation according to a specified resolution scale or area of interest, akin to semantic priming. This adaptability has spurred a wave of creativity and innovation in applying SAM to medical imaging. Our review focuses on the period from April 1, 2023, to September 30, 2023, a critical first six months post-initial publication. We examine the adaptations and integrations of SAM necessary to address longstanding clinical challenges, particularly in the context of 33 open datasets covered in our analysis. While SAM approaches or achieves state-of-the-art performance in numerous applications, it falls short in certain areas, such as segmentation of the carotid artery, adrenal glands, optic nerve, and mandible bone. Our survey delves into the innovative techniques where SAM's foundational approach excels and explores the core concepts in translating and applying these models effectively in diverse medical imaging scenarios.

James Y. Huang, Wenxuan Zhou, Fei Wang et al. · microsoft-research

Despite the strong capabilities of Large Language Models (LLMs) to acquire knowledge from their training corpora, the memorization of sensitive information in the corpora such as copyrighted, biased, and private content has led to ethical and legal concerns. In response to these challenges, unlearning has emerged as a potential remedy for LLMs affected by problematic training data. However, previous unlearning techniques are either not applicable to black-box LLMs due to required access to model internal weights, or violate data protection principles by retaining sensitive data for inference-time correction. We propose δ-Unlearning, an offset unlearning framework for black-box LLMs. Instead of tuning the black-box LLM itself, δ-Unlearning learns the logit offset needed for unlearning by contrasting the logits from a pair of smaller models. Experiments demonstrate that δ- Unlearning can effectively unlearn target data while maintaining similar or even stronger performance on general out-of-forget-scope tasks. δ-Unlearning also effectively incorporates different unlearning algorithms, making our approach a versatile solution to adapting various existing unlearning algorithms to black-box LLMs.

Wenhui Wang, Shuming Ma, Hanwen Xu et al.

This technical report presents LongViT, a vision Transformer that can process gigapixel images in an end-to-end manner. Specifically, we split the gigapixel image into a sequence of millions of patches and project them linearly into embeddings. LongNet is then employed to model the extremely long sequence, generating representations that capture both short-range and long-range dependencies. The linear computation complexity of LongNet, along with its distributed algorithm, enables us to overcome the constraints of both computation and memory. We apply LongViT in the field of computational pathology, aiming for cancer diagnosis and prognosis within gigapixel whole-slide images. Experimental results demonstrate that LongViT effectively encodes gigapixel images and outperforms previous state-of-the-art methods on cancer subtyping and survival prediction. Code and models will be available at https://aka.ms/LongViT.

Sheng Zhang, Qianchu Liu, Guanghui Qin et al. · microsoft-research

Reinforcement learning from verifiable rewards (RLVR) has recently gained attention for its ability to elicit self-evolved reasoning capabilitie from base language models without explicit reasoning supervisions, as demonstrated by DeepSeek-R1. While prior work on RLVR has primarily focused on mathematical and coding domains, its applicability to other tasks and domains remains unexplored. In this work, we investigate whether medical reasoning can emerge from RLVR. We introduce Med-RLVR as an initial study of RLVR in the medical domain leveraging medical multiple-choice question answering (MCQA) data as verifiable labels. Our results demonstrate that RLVR is not only effective for math and coding but also extends successfully to medical question answering. Notably, Med-RLVR achieves performance comparable to traditional supervised fine-tuning (SFT) on in-distribution tasks while significantly improving out-of-distribution generalization, with an 8-point accuracy gain. Further analysis of training dynamics reveals that, with no explicit reasoning supervision, reasoning emerges from the 3B-parameter base model. These findings underscore the potential of RLVR in domains beyond math and coding, opening new avenues for its application in knowledge-intensive fields such as medicine.

Qin Liu, Wenxuan Zhou, Nan Xu et al. · microsoft-research

One critical challenge for large language models (LLMs) for making complex reasoning is their reliance on matching reasoning patterns from training data, instead of proactively selecting the most appropriate cognitive strategy to solve a given task. Existing approaches impose fixed cognitive structures that enhance performance in specific tasks but lack adaptability across diverse scenarios. To address this limitation, we introduce METASCALE, a test-time scaling framework based on meta-thoughts -- adaptive thinking strategies tailored to each task. METASCALE initializes a pool of candidate meta-thoughts, then iteratively selects and evaluates them using a multi-armed bandit algorithm with upper confidence bound selection, guided by a reward model. To further enhance adaptability, a genetic algorithm evolves high-reward meta-thoughts, refining and extending the strategy pool over time. By dynamically proposing and optimizing meta-thoughts at inference time, METASCALE improves both accuracy and generalization across a wide range of tasks. Experimental results demonstrate that MetaScale consistently outperforms standard inference approaches, achieving an 11% performance gain in win rate on Arena-Hard for GPT-4o, surpassing o1-mini by 0.9% under style control. Notably, METASCALE scales more effectively with increasing sampling budgets and produces more structured, expert-level responses.

Zifeng Wang, Hanyin Wang, Benjamin Danek et al.

The integration of Large Language Models (LLMs) into medical applications has sparked widespread interest across the healthcare industry, from drug discovery and development to clinical decision support, assisting telemedicine, medical devices, and healthcare insurance applications. This perspective paper aims to discuss the inner workings of building LLM-powered medical AI applications and introduces a comprehensive framework for their development. We review existing literature and outline the unique challenges of applying LLMs in specialized medical contexts. Additionally, we introduce a three-step framework to organize medical LLM research activities: 1) Modeling: breaking down complex medical workflows into manageable steps for developing medical-specific models; 2) Optimization: optimizing the model performance with crafted prompts and integrating external knowledge and tools, and 3) System engineering: decomposing complex tasks into subtasks and leveraging human expertise for building medical AI applications. Furthermore, we offer a detailed use case playbook that describes various LLM-powered medical AI applications, such as optimizing clinical trial design, enhancing clinical decision support, and advancing medical imaging analysis. Finally, we discuss various challenges and considerations for building medical AI applications with LLMs, such as handling hallucination issues, data ownership and compliance, privacy, intellectual property considerations, compute cost, sustainability issues, and responsible AI requirements.

Shane Waxler, Paul Blazek, Davis White et al.

Realizing personalized medicine at scale calls for methods that distill insights from longitudinal patient journeys, which can be viewed as a sequence of medical events. Foundation models pretrained on large-scale medical event data represent a promising direction for scaling real-world evidence generation and generalizing to diverse downstream tasks. Using Epic Cosmos, a dataset with medical events from de-identified longitudinal health records for 16.3 billion encounters over 300 million unique patient records from 310 health systems, we introduce the Curiosity models, a family of decoder-only transformer models pretrained on 118 million patients representing 115 billion discrete medical events (151 billion tokens). We present the largest scaling-law study of medical event data, establishing a methodology for pretraining and revealing power-law scaling relationships for compute, tokens, and model size. Consequently, we pretrained a series of compute-optimal models with up to 1 billion parameters. Conditioned on a patient's real-world history, Curiosity autoregressively predicts the next medical event to simulate patient health timelines. We studied 78 real-world tasks, including diagnosis prediction, disease prognosis, and healthcare operations. Remarkably for a foundation model with generic pretraining and simulation-based inference, Curiosity generally outperformed or matched task-specific supervised models on these tasks, without requiring task-specific fine-tuning or few-shot examples. Curiosity's predictive power consistently improves as the model and pretraining scale. Our results show that Curiosity, a generative medical event foundation model, can effectively capture complex clinical dynamics, providing an extensible and generalizable framework to support clinical decision-making, streamline healthcare operations, and improve patient outcomes.

Cliff Wong, Sam Preston, Qianchu Liu et al. · microsoft-research

A significant fraction of real-world patient information resides in unstructured clinical text. Medical abstraction extracts and normalizes key structured attributes from free-text clinical notes, which is the prerequisite for a variety of important downstream applications, including registry curation, clinical trial operations, and real-world evidence generation. Prior medical abstraction methods typically resort to building attribute-specific models, each of which requires extensive manual effort such as rule creation or supervised label annotation for the individual attribute, thus limiting scalability. In this paper, we show that existing frontier models already possess the universal abstraction capability for scaling medical abstraction to a wide range of clinical attributes. We present UniMedAbstractor (UMA), a unifying framework for zero-shot medical abstraction with a modular, customizable prompt template and the selection of any frontier large language models. Given a new attribute for abstraction, users only need to conduct lightweight prompt adaptation in UMA to adjust the specification in natural languages. Compared to traditional methods, UMA eliminates the need for attribute-specific training labels or handcrafted rules, thus substantially reducing the development time and cost. We conducted a comprehensive evaluation of UMA in oncology using a wide range of marquee attributes representing the cancer patient journey. These include relatively simple attributes typically specified within a single clinical note (e.g. performance status), as well as complex attributes requiring sophisticated reasoning across multiple notes at various time points (e.g. tumor staging). Based on a single frontier model such as GPT-4o, UMA matched or even exceeded the performance of state-of-the-art attribute-specific methods, each of which was tailored to the individual attribute.

Yu Gu, Jingjing Fu, Xiaodong Liu et al.

Large frontier models like GPT-5 now achieve top scores on medical benchmarks. But our stress tests tell a different story. Leading systems often guess correctly even when key inputs like images are removed, flip answers under trivial prompt changes, and fabricate convincing yet flawed reasoning. These aren't glitches; they expose how today's benchmarks reward test-taking tricks over medical understanding. We evaluate six flagship models across six widely used benchmarks and find that high leaderboard scores hide brittleness and shortcut learning. Through clinician-guided rubric evaluation, we show that benchmarks vary widely in what they truly measure yet are treated interchangeably, masking failure modes. We caution that medical benchmark scores do not directly reflect real-world readiness. If we want AI to earn trust in healthcare, we must demand more than leaderboard wins and must hold systems accountable for robustness, sound reasoning, and alignment with real medical demands.

Bangzheng Li, Fei Wang, Wenxuan Zhou et al. · microsoft-research

Vision-Language Models (VLMs) leverage aligned visual encoders to transform images into visual tokens, allowing them to be processed similarly to text by the backbone large language model (LLM). This unified input paradigm enables VLMs to excel in vision-language tasks such as visual question answering (VQA). To improve fine-grained visual reasoning, recent advancements in vision-language modeling introduce image cropping techniques that feed all encoded sub-images into the model. However, this approach significantly increases the number of visual tokens, leading to inefficiency and potential distractions for the LLM. To address the generalization challenges of image representation in VLMs, we propose a lightweight, universal framework that seamlessly integrates with existing VLMs to enhance their ability to process finegrained details. Our method leverages textual semantics to identify key visual areas, improving VQA performance without requiring any retraining of the VLM. Additionally, it incorporates textual signals into the visual encoding process, enhancing both efficiency and effectiveness. The proposed method, SEMCLIP, strengthens the visual understanding of a 7B VLM, LLaVA-1.5 by 3.3% on average across 7 benchmarks, and particularly by 5.3% on the challenging detailed understanding benchmark V*.

Theodore Zhao, Sid Kiblawi, Naoto Usuyama et al.

Detecting and segmenting small objects, such as lung nodules and tumor lesions, remains a critical challenge in image analysis. These objects often occupy less than 0.1% of an image, making traditional transformer architectures inefficient and prone to performance degradation due to redundant attention computations on irrelevant regions. Existing sparse attention mechanisms rely on rigid hierarchical structures, which are poorly suited for detecting small, variable, and uncertain object locations. In this paper, we propose BoltzFormer, a novel transformer-based architecture designed to address these challenges through dynamic sparse attention. BoltzFormer identifies and focuses attention on relevant areas by modeling uncertainty using a Boltzmann distribution with an annealing schedule. Initially, a higher temperature allows broader area sampling in early layers, when object location uncertainty is greatest. As the temperature decreases in later layers, attention becomes more focused, enhancing efficiency and accuracy. BoltzFormer seamlessly integrates into existing transformer architectures via a modular Boltzmann attention sampling mechanism. Comprehensive evaluations on benchmark datasets demonstrate that BoltzFormer significantly improves segmentation performance for small objects while reducing attention computation by an order of magnitude compared to previous state-of-the-art methods.

James Y. Huang, Wenxuan Zhou, Nan Xu et al.

The ability of Large Language Models (LLMs) to generate structured outputs that follow arbitrary schemas is crucial to a wide range of downstream tasks that require diverse structured representations of results such as information extraction, table generation, and function calling. While modern LLMs excel in generating unstructured responses in natural language, whether this advancement translates to a strong performance on text-to-structure tasks remains unclear. To bridge this gap, we first introduce OmniStruct, a comprehensive benchmark for assessing LLMs' capabilities on diverse text-to-structure tasks such as information extraction, table generation, and function calling. We build OmniStruct by identifying existing datasets across a wide range of tasks that are suitable for a structured answer format, and adapting them under a unified text-to-structure problem setting. To facilitate the development of efficient text-to-structure models, we collect high-quality training data via synthetic task generation. Without using any supervised data for OmniStruct tasks, our experiments demonstrate the possibility of fine-tuning much smaller models on synthetic data into universal structured generation models that can rival the performance of GPT-4o.

Qin Liu, Jacob Dineen, Yuxi Huang et al. · microsoft-research

Benchmarks are central to measuring the capabilities of large language models and guiding model development, yet widespread data leakage from pretraining corpora undermines their validity. Models can match memorized content rather than demonstrate true generalization, which inflates scores, distorts cross-model comparisons, and misrepresents progress. We introduce ArenaBencher, a model-agnostic framework for automatic benchmark evolution that updates test cases while preserving comparability. Given an existing benchmark and a diverse pool of models to be evaluated, ArenaBencher infers the core ability of each test case, generates candidate question-answer pairs that preserve the original objective, verifies correctness and intent with an LLM as a judge, and aggregates feedback from multiple models to select candidates that expose shared weaknesses. The process runs iteratively with in-context demonstrations that steer generation toward more challenging and diagnostic cases. We apply ArenaBencher to math problem solving, commonsense reasoning, and safety domains and show that it produces verified, diverse, and fair updates that uncover new failure modes, increase difficulty while preserving test objective alignment, and improve model separability. The framework provides a scalable path to continuously evolve benchmarks in step with the rapid progress of foundation models.

Alyssa Unell, Noel C. F. Codella, Sam Preston et al.

The National Comprehensive Cancer Network (NCCN) provides evidence-based guidelines for cancer treatment. Translating complex patient presentations into guideline-compliant treatment recommendations is time-intensive, requires specialized expertise, and is prone to error. Advances in large language model (LLM) capabilities promise to reduce the time required to generate treatment recommendations and improve accuracy. We present an LLM agent-based approach to automatically generate guideline-concordant treatment trajectories for patients with non-small cell lung cancer (NSCLC). Our contributions are threefold. First, we construct a novel longitudinal dataset of 121 cases of NSCLC patients that includes clinical encounters, diagnostic results, and medical histories, each expertly annotated with the corresponding NCCN guideline trajectories by board-certified oncologists. Second, we demonstrate that existing LLMs possess domain-specific knowledge that enables high-quality proxy benchmark generation for both model development and evaluation, achieving strong correlation (Spearman coefficient r=0.88, RMSE = 0.08) with expert-annotated benchmarks. Third, we develop a hybrid approach combining expensive human annotations with model consistency information to create both the agent framework that predicts the relevant guidelines for a patient, as well as a meta-classifier that verifies prediction accuracy with calibrated confidence scores for treatment recommendations (AUROC=0.800), a critical capability for communicating the accuracy of outputs, custom-tailoring tradeoffs in performance, and supporting regulatory compliance. This work establishes a framework for clinically viable LLM-based guideline adherence systems that balance accuracy, interpretability, and regulatory requirements while reducing annotation costs, providing a scalable pathway toward automated clinical decision support.