EXMOS: Explanatory Model Steering through Multifaceted Explanations and Data Configurations

Published: May 11, 2024

Explanations in interactive machine-learning systems facilitate debugging and improving prediction models. However, the effectiveness of various global model-centric data-centric explanations aiding domain experts to detect resolve potential data issues for model improvement remains unexplored. This research investigates influence that support healthcare optimising models through automated manual configurations. We conducted quantitative (n=70) qualitative (n=30) studies with explore impact different on trust, understandability improvement. Our results reveal insufficiency guiding users during configuration. Although enhanced understanding post-configuration system changes, a hybrid fusion both explanation types demonstrated highest effectiveness. Based our study results, we also present design implications effective explanation-driven systems.

Language: Английский

---

Human-in-the-loop machine learning: Reconceptualizing the role of the user in interactive approaches

Internet of Things, Journal Year: 2024, Volume and Issue: 25, P. 101048 - 101048

Published: Jan. 9, 2024

The rise of intelligent systems and smart spaces has opened up new opportunities for human-machine collaborations. Interactive Machine Learning (IML) contribute to fostering such Nonetheless, IML solutions tend overlook critical factors as the timing, frequency workload that drive this interaction are vital adapting these users' goals engagement. To address gap, work explores expectations towards in context an interactive hydration monitoring system workplace, which represents a challenging environment implement can collaborate with individuals. proposed involves users learning process by providing feedback on success detecting their drinking gestures enabling them additional examples data. A qualitative study was conducted evaluate use case, where participants completed specific tasks varying levels involvement. This provides promising insights into potential placing Human-in-the-Loop (HitL) adapt reconceptualize role solutions, highlighting importance considering human designing more effective flexible collaborative between humans machines.

Language: Английский

---

From High Dimensions to Human Insight: Exploring Dimensionality Reduction for Chemical Space Visualization

Molecular Informatics, Journal Year: 2024, Volume and Issue: 44(1)

Published: Dec. 5, 2024

Abstract Dimensionality reduction is an important exploratory data analysis method that allows high‐dimensional to be represented in a human‐interpretable lower‐dimensional space. It extensively applied the of chemical libraries, where structure ‐ as feature vectors‐are transformed into 2D or 3D space maps. In this paper, commonly used dimensionality techniques Principal Component Analysis (PCA), t‐Distributed Stochastic Neighbor Embedding (t‐SNE), Uniform Manifold Approximation and Projection (UMAP), Generative Topographic Mapping (GTM) are evaluated terms neighborhood preservation visualization capability sets small molecules from ChEMBL database.

Language: Английский

---

Navigating the landscape of concept-supported XAI: Challenges, innovations, and future directions

Multimedia Tools and Applications, Journal Year: 2024, Volume and Issue: 83(25), P. 67147 - 67197

Published: Jan. 22, 2024

Abstract This comprehensive review of concept-supported interpretation methods in Explainable Artificial Intelligence (XAI) navigates the multifaceted landscape. As machine learning models become more complex, there is a greater need for that deconstruct their decision-making processes. Traditional techniques frequently emphasise lower-level attributes, resulting schism between complex algorithms and human cognition. To bridge this gap, our research focuses on XAI, new line XAI emphasises higher-level attributes or 'concepts' are aligned with end-user understanding needs. We provide thorough examination over twenty-five seminal works, highlighting respective strengths weaknesses. A list available concept datasets, as opposed to training presented, along discussion sufficiency metrics importance robust evaluation methods. In addition, we identify six key factors influence efficacy interpretation: network architecture, settings, protocols, presence confounding standardised methodology. also investigate robustness these methods, emphasising potential significantly advance field by addressing issues like misgeneralization, information overload, trustworthiness, effective human-AI communication, ethical concerns. The paper concludes an exploration open challenges such development automatic discovery strategies expert-AI integration, optimising primary model managing designing efficient

Language: Английский

---

Mapping the Design Space of Teachable Social Media Feed Experiences

Published: May 11, 2024

Social media feeds are deeply personal spaces that reflect individual values and preferences. However, top-down, platform-wide content algorithms can reduce users' sense of agency fail to account for nuanced experiences values. Drawing on the paradigm interactive machine teaching (IMT), an interaction framework non-expert algorithmic adaptation, we map out a design space teachable social feed empower agential, personalized curation. To do so, conducted think-aloud study (N = 24) featuring four platforms—Instagram, Mastodon, TikTok, Twitter—to understand key signals users leveraged determine value post in their feed. We synthesized into taxonomies that, when combined with user interviews, inform five principles extend IMT setting. finally embodied our three designs present as sensitizing concepts moving forward.

Language: Английский

---

Towards Directive Explanations: Crafting Explainable AI Systems for Actionable Human-AI Interactions

Published: May 11, 2024

With Artificial Intelligence (AI) becoming ubiquitous in every application domain, the need for explanations is paramount to enhance transparency and trust among non-technical users. Despite potential shown by Explainable AI (XAI) enhancing understanding of complex systems, most XAI methods are designed technical experts rather than consumers. Consequently, such overwhelmingly seldom guide users achieving their desired predicted outcomes. This paper presents ongoing research crafting systems tailored outcomes through improved human-AI interactions. highlights objectives methods, key takeaways implications learned from user studies. It outlines open questions challenges enhanced collaboration, which author aims address future work.

Language: Английский

---

Explainable artificial intelligence for energy systems maintenance: A review on concepts, current techniques, challenges, and prospects

Renewable and Sustainable Energy Reviews, Journal Year: 2025, Volume and Issue: 216, P. 115668 - 115668

Published: April 8, 2025

Language: Английский

---

Toward trustworthy AI with integrative explainable AI frameworks

it - Information Technology, Journal Year: 2025, Volume and Issue: unknown

Published: April 30, 2025

Abstract As artificial intelligence (AI) increasingly permeates high-stakes domains such as healthcare, transportation, and law enforcement, ensuring its trustworthiness has become a critical challenge. This article proposes an integrative Explainable AI (XAI) framework to address the challenges of interpretability, explainability, interactivity, robustness. By combining XAI methods, incorporating human-AI interaction using suitable evaluation techniques, implementation this serves holistic approach. The discusses framework’s contribution trustworthy gives outlook on open related interdisciplinary collaboration, generalization evaluation.

Language: Английский

---

Interactive Explainable Anomaly Detection for Industrial Settings

Lecture notes in computer science, Journal Year: 2025, Volume and Issue: unknown, P. 133 - 147

Published: Jan. 1, 2025

Language: Английский

---

Unpacking Human-AI interactions: From Interaction Primitives to a Design Space

ACM Transactions on Interactive Intelligent Systems, Journal Year: 2024, Volume and Issue: 14(3), P. 1 - 51

Published: June 8, 2024

This article aims to develop a semi-formal representation for Human-AI (HAI) interactions, by building set of interaction primitives which can specify the information exchanges between users and AI systems during their interaction. We show how these be combined into patterns capture common interactions humans AI/ML models. The motivation behind this is twofold: firstly, provide compact generalization existing practices design implementation HAI interactions; secondly, support creation new extending space interactions. Taking consideration frameworks, guidelines, taxonomies related human-centered systems, we define vocabulary describing based on model’s characteristics interactional capabilities. Based vocabulary, message passing model models presented, demonstrate account approaches. Finally, build describe models, discuss approach used toward that creates possibilities designs as well keeping track issues concerns.

Language: Английский

---