The content of the data that a language model is trained on can have profound effects on its performance and the efficiency of the training process [Rae et al., 2021, Longpre et al., 2023, Penedo

Deep learning in small and imbalanced biomedical datasets remains fundamentally constrained by unstable optimization and poor generalization. We present the first biomedical implementation of FOSSIL (Flexible Optimization via Sample-Sensitive Importance Learning), a regret-minimizing weighting framework that adaptively balances training emphasis according to sample difficulty. Using softmax-based uncertainty as a continuous measure of difficulty, we construct a four-stage curriculum (Easy-Very Hard) and integrate FOSSIL into both convolutional and transformer-based architectures for Mpox skin lesion diagnosis. Across all settings, FOSSIL substantially improves discrimination (AUC = 0.9573), calibration (ECE = 0.053), and robustness under real-world perturbations, outperforming conventional baselines without metadata, manual curation, or synthetic augmentation. The results position FOSSIL as a generalizable, data-efficient, and interpretable framework for difficulty-aware learning in medical imaging under data scarcity.

Recent advances in Multimodal Large Language Models (MLLMs) have significantly enhanced the naturalness and flexibility of human computer interaction by enabling seamless understanding across text, vision, and audio modalities. Among these, voice enabled models such as SpeechGPT have demonstrated considerable improvements in usability, offering expressive, and emotionally responsive interactions that foster deeper connections in real world communication scenarios. However, the use of voice introduces new security risks, as attackers can exploit the unique characteristics of spoken language, such as timing, pronunciation variability, and speech to text translation, to craft inputs that bypass defenses in ways not seen in text-based systems. Despite substantial research on text based jailbreaks, the voice modality remains largely underexplored in terms of both attack strategies and defense mechanisms. In this work, we present an adversarial attack targeting the speech input of aligned MLLMs in a white box scenario. Specifically, we introduce a novel token level attack that leverages access to the model's speech tokenization to generate adversarial token sequences. These sequences are then synthesized into audio prompts, which effectively bypass alignment safeguards and to induce prohibited outputs. Evaluated on SpeechGPT, our approach achieves up to 89 percent attack success rate across multiple restricted tasks, significantly outperforming existing voice based jailbreak methods. Our findings shed light on the vulnerabilities of voice-enabled multimodal systems and to help guide the development of more robust next-generation MLLMs.

We introduce EXIT, an extractive context compression framework that enhances both the effectiveness and efficiency of retrieval-augmented generation (RAG) in question answering (QA). Current RAG systems often struggle when retrieval models fail to rank the most relevant documents, leading to the inclusion of more context at the expense of latency and accuracy. While abstractive compression methods can drastically reduce token counts, their token-by-token generation process significantly increases end-to-end latency. Conversely, existing extractive methods reduce latency but rely on independent, non-adaptive sentence selection, failing to fully utilize contextual information. EXIT addresses these limitations by classifying sentences from retrieved documents - while preserving their contextual dependencies - enabling parallelizable, context-aware extraction that adapts to query complexity and retrieval quality. Our evaluations on both single-hop and multi-hop QA tasks show that EXIT consistently surpasses existing compression methods and even uncompressed baselines in QA accuracy, while also delivering substantial reductions in inference time and token count. By improving both effectiveness and efficiency, EXIT provides a promising direction for developing scalable, high-quality QA solutions in RAG pipelines. Our code is available at https://github.com/ThisIsHwang/EXIT

Chronic obstructive pulmonary disease (COPD) is a chronic inflammatory lung disease that causes obstructed airflow from the lungs. In the United States, more than 15.7 million Americans have been diagnosed with COPD, with 96% of individuals living with at least one other chronic health condition. It is the 4th leading cause of death in the country. Over 2.2 million patients are admitted to hospitals annually due to COPD exacerbations. Monitoring and predicting patient exacerbations on-time could save their life. This paper presents two different predictive models to predict COPD exacerbation using AI and natural language processing (NLP) approaches. These models use respiration summary notes, symptoms, and vital signs. To train and test these models, data records containing physiologic signals and vital signs time series were used. These records were captured from patient monitors and comprehensive clinical data obtained from hospital medical information systems for tens of thousands of Intensive Care Unit (ICU) patients. We achieved an area under the Receiver operating characteristic (ROC) curve of 0.82 in detection and prediction of COPD exacerbation.

Respiratory diseases impose a significant burden on global health, with current diagnostic and management practices primarily reliant on specialist clinical testing. This work aims to develop machine learning-based algorithms to facilitate at-home respiratory disease monitoring and assessment for patients undergoing continuous positive airway pressure (CPAP) therapy. Data were collected from 30 healthy adults, encompassing respiratory pressure, flow, and dynamic thoraco-abdominal circumferential measurements under three breathing conditions: normal, panting, and deep breathing. Various machine learning models, including the random forest classifier, logistic regression, and support vector machine (SVM), were trained to predict breathing types. The random forest classifier demonstrated the highest accuracy, particularly when incorporating breathing rate as a feature. These findings support the potential of AI-driven respiratory monitoring systems to transition respiratory assessments from clinical settings to home environments, enhancing accessibility and patient autonomy. Future work involves validating these models with larger, more diverse populations and exploring additional machine learning techniques.

In the rapidly evolving field of healthcare management, the analysis of medical claims data has become an essential component for improving the quality and equity of healthcare services. The nature of care delivery in the United states is heavily influenced by its fragmentation--care is often spread across multiple disconnected providers (e.g., primary-care physicians, specialists). Settings with greater care fragmentation have been shown to inhibit effective communication and coordination between care team members, thus contributing to higher costs and lower quality of treatment [13,33,21,1,7]. Despite the well-understood impacts of fragmentation, there are still few quantitative tools that can capture the mechanisms of care delivery networks at scale [14]. Standard analyses of local infrastructure features, often executed using tabular data, are limited in their ability to distill complex dynamics between physicians.

Selecting high-quality data for pre-training is crucial in shaping the downstream task performance of language models. A major challenge lies in identifying this optimal subset, a problem generally considered intractable, thus necessitating scalable and effective heuristics. In this work, we propose a data selection method, CoLoR-Filter (Conditional Loss Reduction Filtering), which leverages an empirical Bayes-inspired approach to derive a simple and computationally efficient selection criterion based on the relative loss values of two auxiliary models. In addition to the modeling rationale, we evaluate CoLoR-Filter empirically on two language modeling tasks: (1) selecting data from C4 for domain adaptation to evaluation on Books and (2) selecting data from C4 for a suite of downstream multiple-choice question answering tasks. We demonstrate favorable scaling both as we subselect more aggressively and using small auxiliary models to select data for large target models. As one headline result, CoLoR-Filter data selected using a pair of 150m parameter auxiliary models can train a 1.2b parameter target model to match a 1.2b parameter model trained on 25b randomly selected tokens with 25x less data for Books and 11x less data for the downstream tasks.

The increasing use of transformer-based large language models brings forward the challenge of processing long sequences. In document visual question answering (DocVQA), leading methods focus on the single-page setting, while documents can span hundreds of pages. We present GRAM, a method that seamlessly extends pre-trained single-page models to the multi-page setting, without requiring computationally-heavy pretraining. To do so, we leverage a single-page encoder for local page-level understanding, and enhance it with document-level designated layers and learnable tokens, facilitating the flow of information across pages for global reasoning. To enforce our model to utilize the newly introduced document-level tokens, we propose a tailored bias adaptation method. For additional computational savings during decoding, we introduce an optional compression stage using our C-Former model, which reduces the encoded sequence length, thereby allowing a tradeoff between quality and latency. Extensive experiments showcase GRAM's state-of-the-art performance on the benchmarks for multi-page DocVQA, demonstrating the effectiveness of our approach.