1,721,026 research outputs found
FORTNIoT: Intelligible Predictions to Improve User Understanding of Smart Home Behavior
No full textFig. 1. Based on self-sustaining predictions (e.g. the sun will set), FORTNIoT can deduce when trigger-condition-action rules (e.g. IF sun set AND anyone home THEN lower the rolling shutter) will trigger in the near future and what effects they will cause (e.g. the rolling shutter will lower). Ubiquitous environments, such as smart homes, are becoming more intelligent and autonomous. As a result, their behavior becomes harder to grasp and unintended behavior becomes more likely. Researchers have contributed tools to better understand and validate an environments' past behavior (e.g. logs, end-user debugging), and to prevent unintended behavior. There is, however, a lack of tools that help users understand the future behavior of such an environment. Information about the actions it will perform, and why it will perform them, remains concealed. In this paper, we contribute FORTNIoT, a well-defined approach that combines self-sustaining predictions (e.g. weather forecasts) and simulations of trigger-condition-action rules to deduce when these rules will trigger in the future and what state changes they will cause to connected smart home entities. We implemented a proof-of-concept of this approach, as well as a visual demonstrator that shows such predictions, including causes and effects, in an overview of a smart home's behavior. A between-subject evaluation with 42 participants indicates that FORTNIoT predictions lead to a more accurate understanding of the future behavior, more confidence in that understanding, and more appropriate trust in what the system will (not) do. We envision a wide variety of situations where predictions about the future are beneficial to inhabitants of smart homes, such as debugging unintended behavior and managing conflicts by exception, and hope to spark a new generation of intelligible tools for ubiquitous environments
Improving AI Text Classification: A Cascaded Approach
No full textLLMs have rapidly evolved into versatile ''foundation models'', repurposed - despite persistent gaps in reliability - for a variety of tasks, such as legal document summarization, medical question answering, and text classification. In this paper, we propose an approach to engineer better text classification solutions for educational grading. We address this challenge with a solution that couples (i) a transformer cascade for rubric-level prediction with (ii) a transparent, traffic-light feedback interface powered by a Mixture-of-Agents LLM system. We compared our approach to a standard LLM and a single transformer architecture using the ASAG dataset. Results show that our approach increases recall for incorrect answers by more than 50% and precision on fully correct answers by 20% compared to a single transformer. Finally, we describe a prototype implementing our approach in an end-to-end, minimally intrusive solution for semi-automatic grading, which allows the teaching staff to review and revise the feedback generated by a Mixture-of-Agents LLM system based on the grade classification.This work was supported by the Special Research Fund (BOF)
of Hasselt University (BOF24OWB28). This research was made possible with support
from the MAXVR-INFRA project, a scalable and flexible infrastructure that facilitates
the transition to digital-physical work environments. The MAXVR-INFRA project is
funded by the European Union - NextGenerationEU and the Flemish Government.
The authors would like to thank Ruben Swidzinski for providing us with Figure 3
NexOz - A Wizard of Oz Approach to Facilitate the Integration of AI in Interactive Systems
No full textThis paper introduces NexOz, an approach to facilitate incremental integration of AI components into interactive systems using a Wizard of Oz (WOz). The paper explores various challenges of AI integration, such as contextual understanding, conversational interaction, state tracking, and gesture recognition, particularly within the context of the Flanders Make OperatorAssist project in the manufacturing industry. NexOz allows developers to simulate AI functionalities through human intervention, enabling rapid prototyping, experimentation, and data collection. By leveraging human operators to emulate AI behavior, NexOz facilitates the design and testing of dialogue flows, interaction patterns, tracking mechanisms, and conversational designs in real-time, without having a fully-fledged implementation of AI-based solutions. Through iterative refinement and continuous feedback loops, NexOz offers a pragmatic approach to navigate the complexities of AI integration in interactive systems.Flanders Make, under the project OperatorAssist_SBO, project number 2021-0133
Flemish Government under the “Onderzoeksprogramma Artificiële Intelligentie (AI) Vlaanderen” progra
NexOz - A Wizard of Oz Approach to Facilitate the Integration of AI in Interactive Systems
No full textThis paper introduces NexOz, an approach to facilitate incremental integration of AI components into interactive systems using a Wizard of Oz (WOz). The paper explores various challenges of AI integration, such as contextual understanding, conversational interaction, state tracking, and gesture recognition, particularly within the context of the Flanders Make OperatorAssist project in the manufacturing industry. NexOz allows developers to simulate AI functionalities through human intervention, enabling rapid prototyping, experimentation, and data collection. By leveraging human operators to emulate AI behavior, NexOz facilitates the design and testing of dialogue flows, interaction patterns, tracking mechanisms, and conversational designs in real-time, without having a fully-fledged implementation of AI-based solutions. Through iterative refinement and continuous feedback loops, NexOz offers a pragmatic approach to navigate the complexities of AI integration in interactive systems.Flanders Make, under the project OperatorAssist_SBO, project number 2021-0133
Flemish Government under the “Onderzoeksprogramma Artificiële Intelligentie (AI) Vlaanderen” progra
Work-a-Pose: Ergonomic Feedback and Posture Improvement Interfaces for Long-Term Sustainable Work
No full textNon-ergonomic postures and the resulting musculoskeletal disorders are key factors in worker disability and well-being. This underlines the importance of designing ergonomic work environments and educating workers in performing tasks ergonomically. We present Work-a-Pose to increase awareness of non-ergonomic
postures and promote long-term sustainable work postures. To this end, we combine camera-based posture tracking with the automatic application of ergonomic guidelines. Glanceable visualizations highlight the worker’s posture and potential ergonomic risks. A complementary, personal tool provides a more detailed overview
of the worker’s ergonomic score and motivates the worker to strive for a healthy work posture through simple gamification techniques
Conception, Approval and First Evaluation of a New Master’s Program Engineering Technology: Software Systems (Informatics) in Belgium
No full textThe demand for skilled software engineers continues to outweigh the number of new graduates by far. Although trends such as AI-based code generation and low-code software development might seem to lessen the need for software engineers, the digital transformation of our society is expected to speed up because of these trends, requiring engineers with fitting proficiencies. This paper highlights the crucial steps in the development and governmental accreditation process of a new curriculum in software systems, and describes the lessons learned after a first generation of graduates. Based on interviews with and studies from diverse actors (e.g., trade unions, local government, EU, and professional organizations such as ACM and IEEE) and in response to top-of-mind concerns from regional industry leaders, we designed and deployed an engineering program that meets the identified needs and aims to educate a new generation of software engineers for the forthcoming digital society. The program educates systems thinkers who engineer this digital society by designing and implementing resilient, intelligent, user-centered solutions that integrate with existing processes and enable new, innovative processes. Our master’s program is a unique joint effort of two Flemish universities, Hasselt University and KU Leuven, and resides in the faculty of Engineering Technology
Work-a-Pose: Ergonomic Feedback and Posture Improvement Interfaces for Long-Term Sustainable Work
No full textNon-ergonomic postures and the resulting musculoskeletal disorders are key factors in worker disability and well-being. This underlines the importance of designing ergonomic work environments and educating workers in performing tasks ergonomically. We present Work-a-Pose to increase awareness of non-ergonomic
postures and promote long-term sustainable work postures. To this end, we combine camera-based posture tracking with the automatic application of ergonomic guidelines. Glanceable visualizations highlight the worker’s posture and potential ergonomic risks. A complementary, personal tool provides a more detailed overview
of the worker’s ergonomic score and motivates the worker to strive for a healthy work posture through simple gamification techniques
Towards Traceable Design Rationale in Augmented Reality
No full textWhen new things, such as buildings or physical products, are designed, the design process typically explores multiple design alternatives and undergoes several iterations. The associated artefacts typically grow from low-fdelity prototypes , such as paper sketches, to high-fdelity prototypes, such as 3D scale models. While previous work has focused on capturing the design rationale behind the decisions that happen during such a design process, this information typically remains secluded and is not easily accessible for the stakeholders. In this paper, we explore how to augment both physical and digital designs with their associated design rationale and decisions. Our exploratory inquiry with three experts in architecture provides qualitative feedback on our augmented reality tool and concepts. We expect that these preliminary results are valuable for future traceabil-ity tools for physical and digital designs, even beyond the domain of architecture
- …
