The notion of explainable user interfaces (XUI) has been proposed as a way to address the growing complexity and unpredictability of many UIs. A key question is whether this aspirational goal can be achieved with reasonable amounts of development effort. This paper takes React as its focus as it is one the most widely used web development frameworks. It shows through practical demonstration that the existing event mechanisms within React can be repurposed with the addition of a small amount of annotation to implement one of the proposed XUI techniques.

This paper summarises the third edition of the SYNERGY workshop on Designing and Building Hybrid Human-AI Systems, held at AVI 2026 in Venice, Italy. The workshop addresses the critical challenge of designing AI systems that genuinely augment human capabilities through meaningful combination, moving beyond approaches where humans function as mere cogs in the machine. Directly aligned with AVI 2026's theme of "Interactive Creativity: Agencies, Interfaces, and Ethics", the workshop brings together researchers and practitioners to advance both theoretical frameworks and practical implementations that preserve human agency while leveraging AI's computational power.

Abstract Whilst it is commonly reported that healthcare is set to benefit from advances in Artificial Intelligence (AI), there is a consensus that, for clinical AI, a gulf exists between conception and implementation. Here we advocate the increased use of situated design and evaluation to close this gap, showing that in the literature there are comparatively few prospective situated studies. Focusing on the combined human-machine decision-making process - modelling, exchanging and resolving - we highlight the need for advances in exchanging and resolving. We present a novel relational space - contextual dimensions of combination - a means by which researchers, developers and clinicians can begin to frame the issues that must be addressed in order to close the chasm. We introduce a space of eight initial dimensions, namely participating agents, control relations, task overlap, temporal patterning, informational proximity, informational overlap, input influence and output representation coverage. We propose that our awareness of where we are in this space of combination will drive the development of interactions and the designs of AI models themselves. Designs that take account of how user-centered they will need to be for their performance to be translated into societal and individual benefit.

This workshop explores the evolving landscape of Human-AI collaboration, focusing on Advanced Visual Interfaces and Artificial Intelligence to enhance human cognition. We explore synergistic models of collaboration that merge human insights with AI capabilities, addressing ethical dimensions and practical AI applications. Our goal is to foster rich interdisciplinary dialogue and challenge existing paradigms of human-machine interaction. We aim to redefine interaction paradigms and establish new benchmarks for intelligent systems, ensuring AI not only supports but significantly augments human decision-making processes.

Research has shown that data-driven robots deployed in social settings are likely to unconsciously perpetuate systemic social biases. Despite this, robots can also be deployed to promote fair behaviour in humans. These phenomena have led to the development of two broad sub-disciplines in HRI concerning ‘fairness’: a data-centric approach to ensuring robots operate fairly and a human-centric approach which aims to use robots as interventions to promote fairness in society. To date, these two fields have developed independently, thus it is unknown how data-driven robots can be used to suppress biases in human-human interactions. In this paper, we present a conceptual framework and hypothetical example of how robots might deploy data-driven fairness interventions, to actively suppress social biases in human-human interactions.

Building upon the Semantic and Spatial Matching for Visual Place Recognition (SSM-VPR) model for Waterborne Imagery, we develop a novel pipeline that leverages segmentation at multiple stages in order to enhance performance, filter out non-discriminative features, and to limit spatial matching to relevant class based edge lines. Our approach is motivated by the unique nature of waterborne imagery, where salient land features often make up a minority of the overall image, with the rest being non-discriminative sea and sky. In order to measure improvements in performance, speed, and storage space We apply each individual segmentation-aware method to the pipeline in, before fusing these methods into a single highly robust and efficient pipeline that balances the benefits of each. We evaluate this approach on a waterborne image dataset and compare our novel approach to the original SSM-VPR model, showing improvements in precision versus recall, total recall, and providing efficient inference time for real-world application.

Image based place recognition has achieved state of the art performance on terrestrial image datasets, however there is very little publicly available research on how these systems perform on waterborne imagery in order to carry out place recognition for autonomous sea vessels. This domain may provide new visual challenges such as water obstruction, distance from shore, lower atmospheric visibility, and camera stability. In this paper, we compare performance and saliency of state of the art place recognition on both terrestrial imagery and waterborne imagery from the Symphony Lake dataset, to see how capable modern pipelines are at adapting to the latter. We utilize convolutional neural network features to highlight salient regions of the candidate image that contributed to its retrieval to gain further insight into what key features are being extracted for each.

The problem addressed here is the classification of videos at the highest level into pre-defined genre. The approach adopted is based on the dynamic content of short sequences (/spl sim/30 secs). This paper presents two methods of extracting motion from a video sequence: foreground object motion and background camera motion. These dynamics are extracted, processed and applied to classify 3 broad classes: sports, cartoons and news. Experimental results for this 3 class problem give error rates of 17%, 8% and 6% for camera motion, object motion and both combined respectively, on /spl sim/30 second sequences.

In this paper we propose an approach to high-level classification of video into genre: sport, cartoon, news, commercial and music. An important issue for automatic high-level classification systems is the amount of time needed to classify a video. Here we investigate classification performance as a function of the test sequence length. In addition we present performance against different orders and combinations of static and dynamic mel-frequency cepstral coefficients (MFCC). We find that static and delta MFCCs perform well for this classification task. A test sequence length of approximately 25 seconds for the 5 class problem gives approximately 80 % correct identification. 1.

This paper gives an insight into biometrics used for speaker recognition. Three different biometrics are presented, based on: acoustic, geometric lip, and holistic facial features. Experiments are carried out using a corpus of the DAVID audio-visual database. Recognition accuracy is found to be similar in the 2 domains. The geometric visual feature is based on a method of signature coding of the contour of the lips and the holistic feature is based on a mean dynamic signature, a method of capturing the motions of the face during a spoken utterance. Physical biometrics (static measurements) demand only small model sizes, perhaps just a single template, and therefore require less training data. Conversely behavioral biometrics contain more variation and demand more training data.