Breast cancer is the most diagnosed cancer in women, with HER2 status critically guiding treatment decisions. Noninvasive prediction of HER2 status from dynamic contrast-enhanced MRI (DCE-MRI) could streamline diagnostics and reduce reliance on biopsy. However, preprocessing high-dynamic-range DCE-MRI into standardized 8-bit RGB format for pretrained neural networks is nontrivial, and normalization strategy significantly affects model performance. We benchmarked intensity normalization strategies using a Triple-Head Dual-Attention ResNet that processes RGB-fused temporal sequences from three DCE phases. Trained on a multicenter cohort (n=1,149) from the I-SPY trials and externally validated on BreastDCEDL_AMBL (n=43 lesions), our model outperformed transformer-based architectures, achieving 0.75 accuracy and 0.74 AUC on I-SPY test data. N4 bias field correction slightly degraded performance. Without fine-tuning, external validation yielded 0.66 AUC, demonstrating cross-institutional generalizability. These findings highlight the effectiveness of dual-attention mechanisms in capturing transferable spatiotemporal features for HER2 stratification, advancing reproducible deep learning biomarkers in breast cancer imaging.

Cellular image segmentation is essential for quantitative biology yet remains difficult due to heterogeneous modalities, morphological variability, and limited annotations. We present GenCellAgent, a training-free multi-agent framework that orchestrates specialist segmenters and generalist vision-language models via a planner-executor-evaluator loop (choose tool $\rightarrow$ run $\rightarrow$ quality-check) with long-term memory. The system (i) automatically routes images to the best tool, (ii) adapts on the fly using a few reference images when imaging conditions differ from what a tool expects, (iii) supports text-guided segmentation of organelles not covered by existing models, and (iv) commits expert edits to memory, enabling self-evolution and personalized workflows. Across four cell-segmentation benchmarks, this routing yields a 15.7\% mean accuracy gain over state-of-the-art baselines. On endoplasmic reticulum and mitochondria from new datasets, GenCellAgent improves average IoU by 37.6\% over specialist models. It also segments novel objects such as the Golgi apparatus via iterative text-guided refinement, with light human correction further boosting performance. Together, these capabilities provide a practical path to robust, adaptable cellular image segmentation without retraining, while reducing annotation burden and matching user preferences.

Jailbreaking techniques pose a significant threat to the safety of Large Language Models (LLMs). Existing defenses typically focus on single-turn attacks, lack coverage across languages, and rely on limited taxonomies that either fail to capture the full diversity of attack strategies or emphasize risk categories rather than the jailbreaking techniques. To advance the understanding of the effectiveness of jailbreaking techniques, we conducted a structured red-teaming challenge. The outcome of our experiments are manifold. First, we developed a comprehensive hierarchical taxonomy of 50 jailbreak strategies, consolidating and extending prior classifications into seven broad families, including impersonation, persuasion, privilege escalation, cognitive overload, obfuscation, goal conflict, and data poisoning. Second, we analyzed the data collected from the challenge to examine the prevalence and success rates of different attack types, providing insights into how specific jailbreak strategies exploit model vulnerabilities and induce misalignment. Third, we benchmark a popular LLM for jailbreak detection, evaluating the benefits of taxonomy-guided prompting for improving automatic detection. Finally, we compiled a new Italian dataset of 1364 multi-turn adversarial dialogues, annotated with our taxonomy, enabling the study of interactions where adversarial intent emerges gradually and succeeds in bypassing traditional safeguards.

Recent advances in large language models (LLMs) for mathematical reasoning have largely focused on tasks with easily verifiable final answers; however, generating and verifying natural language math proofs remains an open challenge. We identify the absence of a reliable, fine-grained evaluator for LLM-generated math proofs as a critical gap. To address this, we propose a systematic methodology for developing and validating evaluators that assign fine-grained scores on a 0-7 scale to model-generated math proofs. To enable this study, we introduce ProofBench, the first expert-annotated dataset of fine-grained proof ratings, spanning 145 problems from six major math competitions (USAMO, IMO, Putnam, etc) and 435 LLM-generated solutions from Gemini-2.5-pro, o3, and DeepSeek-R1. %with expert gradings. Using ProofBench as a testbed, we systematically explore the evaluator design space across key axes: the backbone model, input context, instructions and evaluation workflow. Our analysis delivers ProofGrader, an evaluator that combines a strong reasoning backbone LM, rich context from reference solutions and marking schemes, and a simple ensembling method; it achieves a low Mean Absolute Error (MAE) of 0.926 against expert scores, significantly outperforming naive baselines. Finally, we demonstrate its practical utility in a best-of-$n$ selection task: at $n=16$, ProofGrader achieves an average score of 4.14 (out of 7), closing 78% of the gap between a naive binary evaluator (2.48) and the human oracle (4.62), highlighting its potential to advance downstream proof generation.

DSC-MRI perfusion is a medical imaging technique for diagnosing and prognosing brain tumors and strokes. Its analysis relies on mathematical deconvolution, but noise or motion artifacts in a clinical environment can disrupt this process, leading to incorrect estimate of perfusion parameters. Although deep learning approaches have shown promising results, their calibration typically rely on third-party deconvolution algorithms to generate reference outputs and are bound to reproduce their limitations. To adress this problem, we propose a physics-informed autoencoder that leverages an analytical model to decode the perfusion parameters and guide the learning of the encoding network. This autoencoder is trained in a self-supervised fashion without any third-party software and its performance is evaluated on a database with glioma patients. Our method shows reliable results for glioma grading in accordance with other well-known deconvolution algorithms despite a lower computation time. It also achieved competitive performance even in the presence of high noise which is critical in a medical environment.

This study presents a comparative evaluation of ten state-of-the-art large language models (LLMs) applied to unstructured text categorization using the Interactive Advertising Bureau (IAB) 2.2 hierarchical taxonomy. The analysis employed a uniform dataset of 8,660 human-annotated samples and identical zero-shot prompts to ensure methodological consistency across all models. Evaluation metrics included four classic measures - accuracy, precision, recall, and F1-score - and three LLM-specific indicators: hallucination ratio, inflation ratio, and categorization cost. Results show that, despite their rapid advancement, contemporary LLMs achieve only moderate classic performance, with average scores of 34% accuracy, 42% precision, 45% recall, and 41% F1-score. Hallucination and inflation ratios reveal that models frequently overproduce categories relative to human annotators. Among the evaluated systems, Gemini 1.5/2.0 Flash and GPT 20B/120B offered the most favorable cost-to-performance balance, while GPT 120B demonstrated the lowest hallucination ratio. The findings suggest that scaling and architectural improvements alone do not ensure better categorization accuracy, as the task requires compressing rich unstructured text into a limited taxonomy - a process that challenges current model architectures. To address these limitations, a separate ensemble-based approach was developed and tested. The ensemble method, in which multiple LLMs act as independent experts, substantially improved accuracy, reduced inflation, and completely eliminated hallucinations. These results indicate that coordinated orchestration of models - rather than sheer scale - may represent the most effective path toward achieving or surpassing human-expert performance in large-scale text categorization.

Large Language Models (LLMs) are increasingly used to simulate public opinion and other social phenomena. Most current studies constrain these simulations to multiple-choice or short-answer formats for ease of scoring and comparison, but such closed designs overlook the inherently generative nature of LLMs. In this position paper, we argue that open-endedness, using free-form text that captures topics, viewpoints, and reasoning processes "in" LLMs, is essential for realistic social simulation. Drawing on decades of survey-methodology research and recent advances in NLP, we argue why this open-endedness is valuable in LLM social simulations, showing how it can improve measurement and design, support exploration of unanticipated views, and reduce researcher-imposed directive bias. It also captures expressiveness and individuality, aids in pretesting, and ultimately enhances methodological utility. We call for novel practices and evaluation frameworks that leverage rather than constrain the open-ended generative diversity of LLMs, creating synergies between NLP and social science.

Microwell microfluidics has been utilized for single-cell analysis to reveal heterogeneity in gene expression, signaling pathways, and phenotypic responses for identifying rare cell types, understanding disease progression, and developing more precise therapeutic strategies. However, designing microwell microfluidics is a considerably complex task, requiring knowledge, experience, and CAD software, as well as manual intervention, which often fails initial designs, demanding multiple costly and time-consuming iterations. In this study, we establish an autonomous large language model (LLM)-driven microwell design framework to generate code-based computer-aided design (CAD) scripts, that enables the rapid and reproducible creation of microwells with diverse geometries and imaging-based analysis. We propose a multimodal large language model (MLLM)-logistic regression framework based on integrating high-level semantic descriptions generated by MLLMs with image embeddings for image classification tasks, aiming to identify microwell occupancy and microwell shape. The fused multimodal representation is input to a logistic regression model, which is both interpretable and computationally efficient. We achieved significant improvements, exceeding 0.92 for occupancy classification and 0.99 for shape classification, across all evaluated MLLMs, compared with 0.50 and 0.55, respectively, when relying solely on direct classification. The MLLM-logistic regression framework is a scalable, efficient solution for high-throughput microwell image analysis. Our study demonstrates an autonomous design microwell platform by translating natural language prompts into optimized device geometries, CAD scripts and image analysis, facilitating the development of next-generation digital discovery by integration of literature mining, autonomous design and experimental data analysis.

In recent years, natural language generative models have shown outstanding performance in text generation tasks. However, when facing specific tasks or particular requirements, they may exhibit poor performance or require adjustments that demand large amounts of additional data. This work introduces PAGE (Prompt Augmentation for text Generation Enhancement), a framework designed to assist these models through the use of simple auxiliary modules. These modules, lightweight models such as classifiers or extractors, provide inferences from the input text. The output of these auxiliaries is then used to construct an enriched input that improves the quality and controllability of the generation. Unlike other generation-assistance approaches, PAGE does not require auxiliary generative models; instead, it proposes a simpler, modular architecture that is easy to adapt to different tasks. This paper presents the proposal, its components and architecture, and reports a proof of concept in the domain of requirements engineering, where an auxiliary module with a classifier is used to improve the quality of software requirements generation.

We explore a class of supervised training objectives that allow a language model to dynamically and autonomously scale the number of compute steps used for each input token. For any token, the model can request additional compute steps by emitting a output. If the model is granted a delay, a specialized token is inserted at the next input step, providing the model with additional compute resources to generate an output. The model can request multiple pauses. To train the model to use outputs judiciously and to calibrate its uncertainty, we frame the selection of each output token as a sequential-decision problem with a time cost. We refer to the class of methods as $\textit{Catch Your Breath}$ losses and we study three methods in this class: CYB-AP frames the model's task as anytime prediction, where an output may be required at any step and accuracy is discounted over time; CYB-VA is a variational approach that aims to maximize prediction accuracy subject to a specified distribution over stopping times; and CYB-DP imposes a penalty based on a computational budget. Through fine-tuning experiments, we identify the best performing loss variant. The CYB model needs only one third as much training data as the baseline (no pause) model needs to achieve the same performance, and half as much data as a model with pauses and a cross-entropy loss. We find that the CYB model requests additional steps when doing so improves accuracy, and the model adapts its processing time to token-level complexity and context. For example, it often pauses after plural nouns like $\textit{patients}$ and $\textit{challenges}$ but never pauses after the first token of contracted words like $\textit{wasn}$ and $\textit{didn}$, and it shows high variability for ambiguous tokens like $\textit{won}$, which could function as either a verb or part of a contraction.