Today’s online platforms, e.g. in e-commerce, often offer users numerous competing options in single product categories, e.g. televisions or watches, making it difficult for the users to identify the best option to suit their preferences. To ease this process, many platforms provide users with simple scores resulting from the aggregation of other users’ reviews. However, these scoring systems may oversimplify the underlying information and lack explanatory context. Our main contribution in this demonstration paper is a novel online system for aggregating customer reviews and explaining the aggregation to users.The system operates through a multi-stage pipeline: it first applies novel automatic ontology extraction methods using BERT or Large Language Models to identify key aspects from customer reviews, then constructs support and attack relations between these aspects using Argumentative Dialogical Agents (ADAs), an existing methodology for generating argumentative analyses of aspects. Finally, it generates ontology-driven, explainable aggregations of the reviews. We evaluate the performance of our system (which we call ADA-X) on the Amazon and Disneyland review datasets, focusing the ontology quality using the LLM-as-a-judge method and aggregation performance against the original Amazon and Disneyland ratings. The demonstration is available at https://ada-x.co.uk/.
Taxonomies play a vital role in structuring and categorizing information across domains. However, many existing taxonomies suffer from limited coverage and outdated or ambiguous nodes, reducing their effectiveness in knowledge retrieval. To address this, we present Taxoria, a novel taxonomy enrichment pipeline that leverages Large Language Models (LLMs) to enhance a given taxonomy. Unlike approaches that extract internal LLM taxonomies, Taxoria uses an existing taxonomy as a seed and prompts an LLM to propose candidate nodes for enrichment. These candidates are then validated to mitigate hallucinations and ensure semantic relevance before integration. The final output includes an enriched taxonomy with provenance tracking and visualization of the final merged taxonomy for analysis.
It is known that a simple numeric planning problem (SNP) with one numeric variable is decidable but undecidable with three (Helmert 2002). A more recent result (Gnad et. al 2023) showed undecidability for two numeric and one propositional variable. In this paper, we show the decidability of SNP with exactly two numeric variables. For this, we first partition the state space into a finite number of regions and demonstrate the decidability of SNP when restricted to any of these regions. Afterwards, we develop a correct search algorithm that abstracts from these regions by tracking an infinite number of states following an arithmetic progression pattern. Finally, we prove termination of the search and draw conclusions about the reasons for undecidability for general SNP.
This paper presents FactNET, a framework for human-centered granular fact-checking. Each claim to check is decomposed into sub claims that are fact-checked independently by using an LLM’s internal knowledge. The user interacts with the framework through an innovative interface, displaying the complex claim in the form of a graph, allowing the input of human knowledge on the topic, and evaluating the trust given in the generated evidence. A video demonstrating the system is available at TO_BE_PUBLISHED (in submission documents for review phase).
People have long asked complex existential and ethical questions, e.g., 'What is a meaningful life?' or 'How can I address inequality?' Theological texts offer deep insights on these critical questions, but their complexity often limits their access to a larger audience. We present a system that uses retrieval-augmented generation (RAG) with large language models (LLMs) to make this wisdom more accessible. By retrieving content from theological sources and synthesizing it into understandable answers, the system addresses a wide range of ethical questions. Technically, we propose ensemble RAG methods with iterative query simplification to align better with the ancient vocabulary. Our application highlights the potential of AI to support ethical reflection. A video is available here.
In this demo paper, we present SPARK (Semantic Planning with Augmented Retrieval and Knowledge), an AI-driven web application that generates personalized itineraries with natural language input. By addressing the limitations of traditional routing systems and machine learning models that lack adaptability to user intent and real-time contexts, SPARK integrates a Neo4j [6] graph that built on OpenStreetMap [8] data with semantic analysis, custom route optimization, and retrieval-augmented generation. By using large language models (LLM), the system outputs travel planning with real-time and enriched data via external APIs. Also, SPARK dynamically adapts to personalized user queries and optimizes routes based on contextual relevance. As demonstrated in an urban scenario, this demo paper highlights the potential of combining graph-based retrieval and LLMs to deliver flexible and context-aware route planning. Meanwhile, our demo shows that the proposed system is scalable and deployable for travel guidance.
AI holds great promise for industrial applications, but its complexity and opacity can hinder trust and limit adoption, particularly in high-stakes sectors, where transparency and accountability are essential. Question Answering with Explanation (QAE) helps address this challenge by enabling natural interaction with AI and providing justifications that allow users to assess, verify, and better understand AI-generated answers, thus improving perceived credibility. However, despite its clear benefits, QAE remains underexplored in industrial contexts. To address this gap, we employ a user-centric framework for QAE, and study it in system information modelling (SIM), a realistic and valuable industrial use case spanning disciplines, lifecycle stages, and downstream engineering tasks. Our framework integrates retrieval-augmented generation (RAG), with instruction, chain-of-thought, and few-shot prompting to guide LLMs in producing reasoning-rich explanations, and incorporates both automatic and user-perspective human evaluation. Extensive experiments on PDF and RDF datasets covering five key SIM topics demonstrate near-human-level performance. Case studies further illustrate our strengths in delivering strong argumentation, contextual evidence, and complex reasoning chains.
Traditionally, flight control laws are verified, among other analyses, by demonstrating robust stability in the flight envelope (namely, in the region of operation of aircraft). Stability requirements are given in terms of gain and phase margins. These margins give an indication of the robustness of the control laws in presence of unmodeled effects and uncertainty. This paper proposes an AI-based approach to support the stability analysis tasks by identifying a structured partition of the flight envelope, where each region of the partition exhibits locally homogeneous stability characteristics. Building on recursive partitioning methods from Machine Learning, the proposed approach leverages the interpretability of tree-based models to facilitate human expert validation and usage of the obtained results.
Analysis of rugby match and training footage is particularly useful for coaches and players to understand and improve their tackling technique, and potentially lower the rate of injuries. Machine learning models (in particular for pose estimation) promise to streamline rugby analysis. However models trained for “general purpose” computer vision tasks, such as pose estimation and object detection, frequently fail as a result of the challenging conditions and significant domain shift that rugby footage presents: high-impact, close-contact play causes problems such as occlusions, motion blur, and unconventional body orientations. It is therefore crucial to understand the specific conditions which cause these systems to fail so they can be prioritised during pre-processing and expensive manual data collection. In this paper we leverage Met-Pose, a metamorphic testing system to understand the specific conditions that cause pose estimation systems to fail. Metamorphic testing is particularly advantageous as this approach side-steps the need for costly, manually labelled data. Our ongoing project on applying pose estimation for rugby analysis employs MediaPipe, a popular, widely used pose estimation system, on rugby broadcast footage. We show how applying metamorphic testing to a sport analytics application can reveal situations that challenge the model without the need for any manual data labelling. For example, our results show that in this context, MediaPipe is particularly sensitive to motion blur and colour loss, but less so to lighting and resolution changes. Furthermore, we show how this process can be adapted to focus on particular aspects of an application by proposing a new metamorphic rule exploring the effect of including or excluding context on MediaPipe’s results. Our results show where MediaPipe struggles in complex, real-world sporting scenarios and also offer concrete insights for improving data augmentation, data collection and system design in sports analytics.
Digital health interventions often require structured, protocol-driven dialogues delivered with high fidelity. Evaluating whether an agent employing a Large Language Model (LLM) can meet these requirements remains challenging, especially in early development stages. In this work, we present VALISE (Virtual Agent Laboratory for Instruction-Following Simulation and Evaluation), a modular framework for simulating and evaluating LLM agent behavior in delivering structured health interventions. VALISE enables configurable agent-patient simulations using synthetic personas and evaluates protocol adherence through a customizable, automated grid assessed by ensembles of LLM-based judges. We demonstrate its use with Brief Action Planning (BAP), a short intervention promoting behavior change in sedentary individuals. Our results strongly align LLM-based and expert annotations, supporting VALISE's effectiveness for early-stage evaluations. VALISE offers a reproducible, extensible platform for testing instruction-following capabilities of LLM agents in digital health.
In the era of social media and instantaneous content consumption, the demand for quick post-event sports highlights has emerged, requiring systems to deliver compelling summaries in a minimal turnaround time. This paper introduces an AI-driven framework designed to streamline the creation of match recaps by leveraging real-time data acquisition and efficient post-processing. The system analyzes multi-modal inputs —including live game statistics, audio-visual feeds, and contextual cues— to automatically identify key moments (e.g., goals, pivotal plays) as they occur. By integrating lightweight neural models and rule-based prioritization, it generates timestamped clips immediately after the match concludes, significantly reducing manual editing effort. The solution not only supports human editors by providing pre-curated material but also enables fully automated highlight production for platforms requiring instant content delivery. Evaluations on soccer and basketball matches demonstrate the system’s ability to cut post-event processing time by 87.5% while maintaining 90% accuracy in event selection compared to manual curation. The work underscores the potential of hybrid AI systems to bridge real-time analytics with post-production workflows, offering scalability across sports and media formats. The generated highlights are already being published on a production platform (https://tiivii.gal), demonstrating real-world applicability.
Automated planning, a core area of artificial intelligence, aims to generate action sequences that achieve specified goals based on a formal model. In classical planning, where only Boolean state variables are allowed, conditional effects are the standard approach for modelling actions with state-dependent outcomes. However, unlike in the classical setting, relatively little research has focused on developing planning methods for numeric problems with conditional effects. To address this gap in the literature, this work studies numeric planning with conditional effects. We formalise its semantics and revise existing classical planning compilations for conditional effects to account for the specific features of numeric planning. This results in three encodings: two are designed for the full class of numeric planning problems, while the third is specific to tasks with conditional effects that increase or decrease variables by a constant, transforming such problems into instances of Simple Numeric Planning, a well-known and practically significant subclass of numeric tasks. The experimental evaluation compares these compilations across both newly designed and compelling benchmarks as well as existing domains featuring conditional effects. Our empirical findings reveal complementary behaviour among the approaches, highlighting the practical impact of selecting the appropriate compilation for different problem structures.
Sorting of parcels is a critical process in intralogistics for the proper processing and dispatching of packages. Commonly, such a process is manually executed by operators along the plant, without any added value, and might result in musculoskeletal injuries due to the non-ergonomic working conditions. Automation solutions are also present in the market and scientific literature. However, available solutions are usually implemented with pre-defined, simplified sorting rules/finite state machines capable of managing only a limited number of parcel types/sorting scenarios. To generalize and fully automate the sorting process in intralogistics, we propose to employ Reinforcement Learning (RL) for the derivation of sorting policies in combination with machine vision for the online tracking of the parcels, used as the state of the RL. More in detail, the on-policy Proximal Policy Optimization (PPO) algorithm is used for RL, and Yolo is chosen as the machine vision algorithm for parcel recognition and tracking. Based on the AMS sorting module of the SAIET Engineering company, a modular kinematic model (with parcels collision modeling) of the sorting system (an n by m AMS - i.e., 2-action actuators - matrix) is derived, and used as the environment for the PPO. Offline sorting policy training is performed by randomizing the parcel number, size, and entry positions. The trained policy is then deployed to the sorting module, which is equipped with cameras for machine vision implementation and performance evaluation. In-distribution and out-of-distribution (i.e., with parcel types not considered in the off-line training) tests achieved the target performance of 96.5% and 94% sorting accuracy, respectively.
The accumulation of litter in natural environments poses significant ecological and social challenges, motivating the development of automated solutions for litter detection. However, collecting and centrally aggregating large-scale annotated datasets for training object detectors often raises privacy and ownership concerns. In this work, we propose a Federated Learning (FL) framework to train a lightweight litter detection model based on the YOLO architecture, which enables collaborative model development without requiring centralized access to raw data. Each participating client locally trains the model on site-specific datasets collected in the wild, and only model updates are shared with a central server for aggregation. We compare and contrast different FL process configurations involving mixed and heterogeneous training datasets built starting from two commonly used benchmark datasets collected across different locations and having very different visual data distributions, i.e. TACO and PlastOPol. Experimental results show that the federated model, trained across these non-IID data distributions, achieves superior generalization in cross-dataset evaluation compared to the corresponding centrally trained models.
Interactive segmentation has become critical for efficiently delineating dune boundaries from remote sensing landform images, enabling geographers to iteratively refine model predictions through minimal user guidance. However, geographers report two major challenges when working with existing tools: (1) handling segmentation around ambiguous dune boundaries forces geographers into dense, repetitive clicking, making the interaction tedious and reducing annotation efficiency; (2) conventional desktop-based annotation platforms mainly support sequential, isolated interactions, hindering the smooth, co-located collaboration necessary for dealing with difficult cases. We thus propose FenGePad, a tangible collaborative interactive segmentation framework. It supports flexible prompt types-clicks, polylines, and scribbles-designed to accommodate geographers' diverse annotation preferences and improve annotation efficiency. To enhance model robustness and generalization, we introduce prompt generation strategies that simulate realistic annotation behaviors of geographers during training. Finally, we instantiate a tablet-based application supporting FenGePad's tangible annotation and collaboration. Comprehensive experiments demonstrate that FenGePad achieves competitive segmentation performance while effectively improving annotation quality and collaborative efficiency. Our results demonstrate the promise of tangible interactive frameworks for applying deep learning in geographic research.
We present ICX (Individual Consistency eXplorer), an interactive dashboard designed to support stakeholders in exploring individual fairness notions within algorithmic decision-making systems. ICX focuses on a set of metrics based on the consistency score, a key measure of individual fairness, by allowing the visualisation of how the classification of an individual compares with that of similar individuals. Stakeholders can define and fine-tune the notion of similarity according to domain-specific criteria, and examine individual-level views that highlight comparable individuals and their classification outcomes. ICX empowers users to interrogate, analyse and interpret fairness at the individual level, making algorithmic decision-making more transparent and accountable.
In this paper, we examine a probabilistic intelligent system that provides posterior information, with the goal of recovering either the prior beliefs or the new data received by the system. We address this inverse problem within a general probabilistic framework and present several key contributions. Specifically, we demonstrate that when only the posterior beliefs and the data used are known, the corresponding prior beliefs form a convex set of probability distributions. We characterize this set by determining its lower and upper probability bounds. Additionally, we show that, in the general case, the minimum number of queries required to recover the prior information grows exponentially with the number of variables. We analyze this problem both when prior beliefs are represented in a non-factorized form and when they are factorized. Finally, we generalize our findings to scenarios where the available information is expressed using an alternative, non-additive theory of uncertainty.
Developments in Artificial Intelligence (AI) today have sparked growing interest in topics such as agentic workflows and Long-Term Memory (LTM) architectures, which extend the capabilities of Large Language Models (LLMs) beyond their current limitations. Agentic workflows represent a significant paradigm shift by enabling LLMs to exhibit goal-oriented behavior, decision-making, and adaptability within dynamic environments. On the other hand, LTM systems enable LLMs to retain and retrieve information across multiple interactions, allowing for personalization, context-aware reasoning, and additional functionalities. This paper elucidates two illustrative use cases, namely the Agentic Event Planner and MemoryGraph, which exemplify the integration of agentic workflows, long-term memory, LLM switching, and related elements into cohesive, hybrid AI pipelines for experimentation purposes. Implemented on the AI-Builder platform, these prototypes benefit from modularity, reusability, and a user-friendly drag-and-drop design environment, with modest orchestration complexity compared to the flexibility and interoperability afforded by the platform.
Infrared-visible object detection aims to leverage the complementary information between infrared and visible modalities to improve detection performance in challenging environments. However, existing infrared-visible object detection methods face several limitations: (1) difficulty in effectively extracting and decomposing modality-common and modality-specific features; (2) interference from modality-irrelevant or redundant information; and (3) insufficient fusion of cross-modal complementary cues. To address these issues, we propose a novel Modality Decomposition and Compensation Fusion Network (MDCF-Net). Specifically, MDCF-Net first decomposes the common and unique features across different modalities. It then performs selective enhancement and interaction between these features via cross-modality compensation. Finally, a dynamic fusion strategy based on spatial and channel attention is applied to adaptively integrate the enhanced features. Extensive experiments on two public datasets, LLVIP and FLIR, demonstrate that our proposed method achieves superior detection performance and exhibits robust generalisation across various challenging conditions. The Code is available at https://github.com/fanjiangtao666/MDCF-Net/tree/main
Bidirectional heuristic search potentially decreases search effort in combinatorial search problems amenable to backward search. To date, bidirectional search has been limited to minimization or shortest path problems. This paper extends the notion of bidirectional heuristic search to (constrained) longest-path problems. We present a bidirectional heuristic search algorithm for longest simple path (LSP) in undirected graphs, and prove its correctness. We then suggest several refinements, as well as a generalization to other types of longest path problems, such as Coil-in-a-box (CIB). Empirical evaluation shows that, as with many forms of bidirectional search, sometimes unidirectional search wins, but for a sizable chunk of problem instances, bidirectional search performs better by expanding fewer nodes and achieves a shorter runtime despite the increased overhead per expansion.
