IBM Research - Ireland

CURE2026 workshop provides a venue for research on how interactive and intelligent interfaces can be leveraged to ensure effective uncertainty communication. With the increased capabilities of black-box AI models, particularly large language models (LLMs), there is a growing urgency to understand exactly what these models can and cannot do. Effective uncertainty communication can help users calibrate their trust and reliance on the AI system. However, effectively conveying uncertainty, be it through conversation or by visualizing their confidence levels, remains a major challenge, as users often get confused or misinterpret uncertainty information due to poorly designed interfaces. CURE2026 workshop brings together researchers from HCI, AI, design, and psychology to investigate how we can design, develop, and evaluate intelligent and interactive interfaces for uncertainty communication. The workshop includes a mini-conference section aimed at inspiring discussion and investigating the role of intelligent and interactive interfaces in uncertainty communication. The workshop also features keynotes by Professors Jessica Hullman and Katrien Verbert. Finally, a part of the workshop is dedicated to an interactive activity to discuss and develop guidelines for communicating uncertainty across different domains and tasks.

Bioactive peptides are an important class of natural products with great functional diversity. Chemical modifications can improve their pharmacology, yet their structural diversity presents unique challenges for computational modeling. Furthermore, data for standard peptides (composed of the 20 canonical amino acids) is more abundant than for modified ones. Thus, we set out to identify whether predictive models fitted to standard data are reliable when applied to modified peptides. To do this, we first considered two critical aspects of the modeling problem, namely, choice of similarity function for guiding dataset partitioning and choice of molecular representation. Similarity-based dataset partitioning is an evaluation technique that divides the dataset into train and test subsets, such that the molecules in the test set are different from those used to fit the model. We demonstrate, across four peptide function prediction tasks, that chemical fingerprint-based similarity measures outperform traditional sequence alignment-based metrics for partitioning standard peptide datasets, challenging usual practices. Similarly, we have found that chemical fingerprints are the best option for building predictive models able to interpolate between classes (standard to standard and modified to modified). Despite a 50% drop in performance, chemical fingerprints are also the best choice in the more challenging extrapolation scenario (standard to modified). All code and data necessary for reproducing the experiments are available in Github (https://github.com/IBM/PeptideGeneralizationBenchmark).

Foundational models (FMs), pretrained on extensive datasets using self-supervised techniques, are capable of learning generalized patterns from large amounts of data. This reduces the need for extensive labeled datasets for each new task, saving both time and resources by leveraging the broad knowledge base established during pretraining. Most research on FMs has primarily focused on unstructured data, such as text and images, or semi-structured data, like time-series. However, there has been limited attention to structured data, such as tabular data, which, despite its prevalence, remains under-studied due to a lack of clean datasets and insufficient research on the transferability of FMs for various tabular data tasks. In response to this gap, we introduce a framework called TabularFM 2 , which incorporates state-of-the-art methods for developing FMs specifically for tabular data. This includes variations of neural architectures such as GANs, VAEs, and Transformers. We have curated a million of tabular datasets and released cleaned versions to facilitate the development of tabular FMs. We pretrained FMs on this curated data, benchmarked various learning methods on these datasets, and released the pretrained models along with leaderboards for future comparative studies. Our fully open-sourced system provides a comprehensive analysis of the transferability of tabular FMs. By releasing these datasets, pretrained models, and leaderboards, we aim to enhance the validity and usability of tabular FMs in the near future. * Equal contribution 2 https://tabularfm.github.io/

Recent research has begun to assess people’s perceptions of voice user interfaces (VUIs) as dialogue partners, termed partner models. Current self-report measures are only available in English, limiting research to English-speaking users. To improve the diversity of user samples and contexts that inform partner modelling research, we translated, localized, and evaluated the Partner Modelling Questionnaire (PMQ) for non-English speaking Western (German, n=185) and East Asian (Japanese, n=198) cohorts where VUI use is popular. Through confirmatory factor analysis (CFA), we find that the scale produces equivalent levels of “goodness-to-fit” for both our German and Japanese translations, confirming its cross-cultural validity. Still, the structure of the communicative flexibility factor did not replicate directly across Western and East Asian cohorts. We discuss how our translations can open up critical research on cultural similarities and differences in partner model use and design, whilst highlighting the challenges for ensuring accurate translation across cultural contexts.

IntroductionIntraoperative indocyanine green fluorescence angiography (ICGFA) perfusion assessment has been demonstrated to reduce complications in reconstructive surgery. This study sought to advance ICGFA flap perfusion assessment via quantification methodologies. MethodPatients undergoing pedicled and free flap reconstruction were subjected to intraoperative ICGFA flap perfusion assessment using either an open or endoscopic system. Patient demographics, clinical impact of ICGFA and outcomes were documented. From the ICGFA recordings, fluorescence signal quality, as well as inflow/outflow milestones for the flap and surrounding (control) tissue were computationally quantified post hoc and compared on a region of interest (ROI) level. Further software development intended full flap quantification, metric computation and heatmap generation. ResultsFifteen patients underwent ICGFA assessment at reconstruction (8 head and neck, 6 breast and 1 perineum) including 10 free and 5 pedicled flaps. Visual ICGFA interpretation altered on-table management in 33.3% of cases, with flap edges trimmed in 4 and a re-anastomosis in 1 patient. One patient suffered post-operative flap dehiscence. Laparoscopic camera use proved feasible but recorded a lower quality signal than the open system.Using established and novel metrics, objective ICGFA signal ROI quantification permitted perfusion comparisons between the flap and surrounding tissue. Full flap assessment feasibility was demonstrated by computing all pixels and subsequent outputs summarisation as heatmaps. ConclusionThis trial demonstrated the feasibility and potential for ICGFA with operator based and quantitative flap perfusion assessment across several reconstructive applications. Further development and implementation of these computational methods requires technique and device standardisation.

This paper addresses the challenge of binary relation classification in biomedical Natural Language Processing (NLP), focusing on diverse domains including gene-disease associations, compound protein interactions, and social determinants of health (SDOH). We evaluate different approaches, including fine-tuning Bidirectional Encoder Representations from Transformers (BERT) models and generative Large Language Models (LLMs), and examine their performance in zero and few-shot settings. We also introduce a novel dataset of biomedical text annotated with social and clinical entities to facilitate research into relation classification. Our results underscore the continued complexity of this task for both humans and models. BERT-based models trained on domain-specific data excelled in certain domains and achieved comparable performance and generalization power to generative LLMs in others. Despite these encouraging results, these models are still far from achieving human-level performance. We also highlight the significance of high-quality training data and domain-specific fine-tuning on the performance of all the considered models.

In this article, we generalize the approximate quantum compiling algorithm into a new method for cnot-depth reduction, which is apt to process wide target quantum circuits. Combining this method with state-of-the-art techniques for error mitigation and circuit compiling, we present a ten-qubit experimental demonstration of iterative amplitude estimation on a quantum computer. The target application is a derivation of the expected value of contract portfolios in the energy industry. In parallel, we also introduce a new variant of the quantum amplitude estimation algorithm, which we call dynamic amplitude estimation, as it is based on the dynamic circuit capability of quantum devices. The algorithm achieves a reduction in the circuit width in the order of the binary precision compared to the typical implementation of quantum amplitude estimation, while simultaneously decreasing the number of quantum-classical iterations (again in the order of the binary precision) compared to the iterative amplitude estimation. The calculation of the expected value, value at risk, and conditional value at risk of contract portfolios on quantum hardware provides a proof of principle of the new algorithm. INDEX TERMS Approximate quantum compiling (AQC), dynamic circuits, error mitigation, quantum amplitude estimation (QAE), quantum computing, risk analysis.