We are delighted to announce the 2nd Edition of the Workshop on Emerging ICT Trends & Applications (WEITA 2025), dedicated to the theme:

“Explainable, Interpretable, and Trustworthy AI: Methods,
Challenges, and Applications.”

This highly anticipated event is proudly organized by ESPRIT—encompassing the School of Business, the School of Engineering, and ESPRIM—in collaboration with the IEEE Tunisia , EMSI Morocco and IEEE Student Branch.
📅 Dates Wednesday and Thursday, February 12–13, 2025.

General Co-Chairs:
  • Faouzi Kamoun
  • Lamjed Bettaieb
Scientific Committee:
  • Mourad Zarai (ESPRIT School of Engineering)
  • Mohamed Tabaa (EMSI, Morocco)
  • Heni Abidi (ESPRIT School of Business)
  • Mohamed Hedi Riahi (ESPRIT School of Engineering)
Organizing Committee:
  • Abderrazek Hachani (ESPRIT School of Engineering)
  • Aymen Ben Brik (ESPRIT School of Business)
  • Manel Madhioub (ESPRIT School of Engineering)
  • Walid El Ayeb (ESPRIT School of Business)
  • Hafedh Abdelghani (ESPRIM)
  • Meriem Chichti (ESPRIT School of Business)
  • Anis Zaouaoui (ESPRIT School of Engineering)
Communications & Outreach Committee
  • Chiraz Gharbi (ESPRIT School of Business)
  • Nebgha Ayachi (ESPRIT School of Engineering)

Artificial Intelligence (AI) systems have become an integral part of modern society, with their deployment spanning numerous critical applications such as healthcare, finance, energy, transportation, cybersecurity, justice systems, and manufacturing.

Many of these AI-driven systems are designed to assist in or directly make decisions that can have far-reaching implications. Moreover, the reliance on specialized hardware accelerators for AI workloads introduces the risk of hardware faults, which can pose significant challenges, particularly in safety-critical applications where accuracy and reliability cannot be compromised.

These challenges underscore the urgent need for AI models that are not only accurate but also explainable and interpretable, ensuring trust and accountability in high-stakes environments.

The advantages of explainability and interpretability in AI are multifaceted: They provide deeper insights into the decision-making processes of AI models, enabling stakeholders to understand the rationale behind the outcomes. Transparent models enhance accountability by making AI systems interpretable and explainable to human users. This builds confidence in AI systems, fostering fairness, trust and user acceptance.

However, the complexity and opacity of many AI models—particularly those leveraging deep learning models—present significant challenges. These “black box” systems often operate with little or no transparency, making it difficult to apprehend how and why specific decisions are made. This lack of transparency can lead to biased outcomes, raise ethical concerns, and ultimately erode trust. Addressing these issues requires rigorous approaches that make AI systems explainable and interpretable.

This workshop on eXplainable AI (XAI) aims to explore cutting-edge methods, tackle existing challenges and risks, and discuss practical applications of XAI to address the pressing need for transparency and trust in AI systems

By bringing together researchers and practitioners, this workshop aspires to provide an interdisciplinary forum for sharing recent research advancements in XAI approaches, best practices, design principles, open challenges, inherent risks, and practical applications. The program will feature keynote presentations, open discussions, idea exchanges, and practical, instructor-led hands-on training sessions.

The primary objectives of this workshop are as follows:

  • Capacity Building: Developing expertise in XAI among researchers and practitioners working in the fields of AI and human-computer interaction.
  • Awareness Raising: Highlighting the pivotal role of explainability, interpretability, and reliability as essential pillars of human-centered, responsible, and trustworthy AI.
  • Application Exposure: Showcasing the diverse range of use cases and application areas where XAI can create transformative impact.
  • Critical Reflection: Engaging participants to reflect upon the ethical, legal, security, and social concerns posed by the lack of transparency in AI systems.
  • Collaboration and Exchange: Seeding collaboration and promoting the exchange of ideas, perspectives, and experiences to advance research in XAI.
  • Experiential Learning: Immersing interested participants in practical applications of XAI by equipping them with hands-on experience in implementing XAI techniques to enhance model transparency, interpretability, and trustworthiness.
The event is structured around four main themes:

  • Technical aspects of XAI (methods, frameworks, and approaches).
  • XAI use cases and application areas
  • Ethical, security, and legal issues surrounding XAI.
  • Hands-on practical training on XAI techniques (Day 2).
The two-day workshop on XAI will be organized as follows:

DAY 1- February 12th: Onsite event with hybrid online participation (Free registration)
  • The first session (half day) will draw attention to the state-of-the art in XAI concepts and approaches.
  • The second session (half day) will provide a deep understanding of the use cases and applications of XAI, while addressing the potential ethical, security, and legal concerns.
DAY 2- February 13th: Onsite Participation only (Registration fee applies)
  • This full-day onsite event consists of an instructor-led hands-on practical training in XAI that aims to train participants on how to apply XAI techniques to understand Neural Network decisions, and boost transparency, fairness, accountability and trust in AI decision making.

Day 1 – Wednesday, February 12thAgenda of Invited Talks
Time Plenary Sessions
9:00 – 9:15 Welcome and opening
9:15 – 9:55 Prof. Dr. Samek Wojciech, Technical University of Berlin & Fraunhofer Heinrich Hertz Institute (HHI), Berlin, Germany.
Component-level Explanation and Validation of AI Models.
ABSTRACT
9:55 – 10:35 Prof. Wassila Ouerdane, CentraleSupélec, Paris-Saclay University, France.
Interpretable Image Classification Through an Argumentative Dialog Between Encoders.
ABSTRACT
10:35 – 10:50 Break
10:50 – 11:30 Dr. Celia Cintas, IBM Research Africa, Nairobi.
A Tale of Adversarial Attacks & Out-of-distribution Detection Stories in the Activation Space.
ABSTRACT
11:30 – 12:15 Dr. Mourad Zerai, ESPRIT School of Engineering, Tunisia.
Why Explainability Matters.
ABSTRACT
12:15 – 13:00 Break
13:00 – 13:40 Prof. Alberto Bosio, University of Lyon – Lyon Institute of Nanotechnology.
Reliable and Efficient Hardware for Trustworthy Deep Neural Networks.
ABSTRACT
13:40 – 14:20 Prof. Giuseppe Primiero, University of Milan, Italy.
BRIO: A Bias and Risk Assessment Formal Methodology and Tool.
ABSTRACT
14:20 – 15:00 Dr. Yazan Mualla, University of Technology of Belfort-Montbéliard (UTBM), France.
Human-Agent Explainability Architecture: Application of Remote Robots.
ABSTRACT
15:00 – 15:15 Break
15:15 – 15:55 Pr. Farkhund Iqbal, College of Technological Innovation, Zayed University, UAE.
The Impact of Explainable AI Models in Enhancing Interpretability and Transparency for Cybersecurity and Digital Forensics.
ABSTRACT
15:55 – 16:35 Dr. Amit Dhurandhar, IBM T.J. Watson, Yorktown Heights, NY, USA.
Explainable AI from Classification to Generation.
ABSTRACT
16:35 – 17:15 Michael Boone, NVIDIA, USA.
Trustworthy AI at NVIDIA.
ABSTRACT
17:15 – 17:25 Closing remarks
Location : Amphitheater Tahar Ben Lakhdar (ESB)

Prof. Dr. Wojciech Samek, Head of Artificial Intelligence Department, Fraunhofer HHI, Professor for ML & Communications, TU Berlin, Germany
Prof. Wojciech Samek is a professor in the department of Electrical Engineering and Computer Science at the Technical University of Berlin and is jointly heading the department of Artificial Intelligence and the Explainable AI Group at Fraunhofer Heinrich Hertz Institute (HHI), Berlin, Germany.
He studied computer science at Humboldt University of Berlin, Heriot-Watt University and University of Edinburgh and received the Dr. rer. nat. degree with distinction (summa cum laude) from the Technical University of Berlin in 2014. During his studies he was awarded scholarships from the German Academic Scholarship Foundation and the DFG Research Training Group GRK 1589/1, and was a visiting researcher at NASA Ames Research Center, Mountain View, USA.
Dr. Samek is a Fellow at the BIFOLD – Berlin Institute for the Foundation of Learning and Data and the ELLIS Unit Berlin, and Principal Investigator at the DFG Research Unit DeSBi and the DFG Graduate School BIOQIC. Furthermore, he an elected member and the Germany’s Platform for Artificial Intelligence, as well as member of the Scientific Advisory Board Member of the Center of Excellence in Artificial Intelligence at AGH University of Krakow and the Executive Boards of the Helmholtz Einstein School in Data Science HEIBRiDS and the DAAD Konrad Zuse School ELIZA. He is recipient of multiple best paper awards, including the 2020 Pattern Recognition Best Paper Award and the 2022 Digital Signal Processing Best Paper Prize, and part of the expert group developing the ISO/IEC MPEG-17 NNC standard.
He edited books on Explainable AI (2019) and xxAI – Beyond explainable AI (2022), was senior editor of IEEE TNNLS, associate editor of Pattern Recognition, Digital Signal Processing, PLoS ONE, area chair at NeurIPS, ICML and NAACL, program chair at IEEE MLSP 2023 and organizer of various special sessions, workshops and tutorials on topics such as explainable AI, neural network compression, and federated learning.
Dr. Samek has co-authored more than 200 peer-reviewed journal and conference papers; some of them listed as ESI Hot (top 0.1%) or Highly Cited Papers (top 1%).

Wassila Ouerdane is a Full Professor of Computer Science at CentraleSupélec, University Paris-Saclay, and a member of the MICS laboratory (Mathematics and Informatic), within the ARCADE research group. She earned her PhD in Computer Science from the University of Paris Dauphine in 2009.
Her research activities cover two areas: Artificial Intelligence and Decision Theory. She focuses on issues related to knowledge representation and reasoning in the context of explainable AI for various application domains (such as healthcare, cyber defense, etc.), with the goal of using formal tools from AI and decision theory to both specify the reasoning of artificial agents (such as decision explanations) and facilitate interaction between agents. She is also interested in the application of decision support methodologies in real-world contexts.
Dr. Ouerdane has supervised over a dozen PhD students, in collaboration with academics and industry partners (Airbus Defense and Space, IBM Zurich, CEA, etc.). Among her responsibilities, Dr. Ouerdane is co-leader of the national working group “Explainability and Trust” within the CNRS RADIA Research Group (Reasoning, Learning, and Decision in Artificial Intelligence).

Dr. Amit Dhurandhars, Principle Research Staff Member, IBM T.J. Watson, Yorktown Heights NY.
Dr. Dhurandhar is a Principal Research Staff Member at the IBM T. J. Watson Research Center. He received his Master and Ph.D. degrees in computer engineering from the University of Florida in 2005 and 2009, respectively.
His research has focused on understanding AI methods through their statistical/output behaviors.
He has worked on projects spanning multiple industries such as Semi-conductor manufacturing, Oil and Gas, Procurement, Retail, Utilities, Airline, Health Care. His current research includes proposing various methods for enhancing trust in AI systems.
His recent work was featured in Forbes and PC magazine with corresponding technical contribution in leading AI research venues such as NeurIPS.
His work has helped uncover interesting insights in fields such as Olfaction with papers in reputed journals such as Science and Nature Communications with extensive media coverage (Quartz, New Yorker, Atlantic, Biological Scene, Science News). His research also has received the AAAI deployed application award, AAAI HCOMP best paper honorable mention as well as being selected as Best of ICDM twice. For his work on explainability he was invited to attend Schloss Dagstuhl seminars as well as provide multiple invited talks including an industry keynote at ACM CODS-COMAD 2021.
He also Co-led the creation of the AI Explainability 360 open-source toolkit. Besides research impact, his work has also gone into IBM product, and he has received Outstanding Technical Achievement as well as IBM Corporate award. He has been an Area Chair and SPC member for top AI conferences as well as has served on National Science Foundation (NSF) panels for the small business innovative research (SBIR) program. He also serves on the invention disclosure committee (IDT) in IBM Research and is an honored listee on Marquis Who’s Who 2024.

Prof. Farkhund Iqbal, College of Technological Innovation, Zayed University, UAE.
Dr. Farkhund Iqbal is a full professor and team lead for the Cybersecurity and Digital Forensics (CAD) research group at the College of Technological Innovation, Zayed University, United Arab Emirates.
He is the Director of the Advanced Cyber Forensics Research Laboratory (ACFRL) at Zayed University. He is also an Affiliate Professor with the School of Information Studies at McGill University, Canada, and an Adjunct Professor with the Faculty of Business and Information Technology at Ontario Tech University, Canada.
He has more than 15+ years of teaching and research experience. He uses artificial intelligence, machine learning, service robotics, and data analytics techniques to solve problems in digital security, digital forensics, and healthcare. He has successfully executed 25+ internal and external research projects (with more than USD 3M funding), resulting in 150+ peer-reviewed impactful publications and a recently published lead-authored book with Springer.
He has more than 4678 citations with an h-index of 34 and i10-index of 91 on Google Scholar. He is the recipient of several prestigious awards.  He has served as a chair and co-chair for several IEEE/ACM conferences and has been a guest editor and reviewer for multiple high-rank journals.

Prof. Giuseppe Primiero, Department of Philosophy , University of Milan, Italy.
Giuseppe Primiero is Professor of Logic at the Department of Philosophy, University of Milan, Italy. He is also the co-founder and Chief Research Officer of MIRAI, a spin-off of the Department of Philosophy at the University of Milan.
Previously, Giuseppe was Senior Lecturer in Computing Sciences at the Foundations of Computing Group, Department of Computer Science, Middlesex University London (UK). Before that he was Post-Doctoral Researcher at the Centre for Logic and Philosophy of Science, Ghent University (Belgium) and Leiden University (the Netherlands).
He also held pre- and post-doctoral visiting positions at CUNY Graduate Centre, Oxford University, Stockholm University, Università di Sassari. Giuseppe received a PhD in Philosophy from the University of Palermo (Italy) and a MA in Philosophy from the University of Leiden (the Netherlands).
His research focuses on formal models of dynamic rationality for intelligent mechanical and natural information systems, the semantics and proof theory of computational systems, and the Philosophy of Computing and Information.
He is the author of the books: Information and Knowledge: A Constructive Type-theoretical Approach (Springer); On the Foundations of Computing (Oxford University Press).

Prof. Alberto Bosio, Ecole Centrale de Lyon – Lyon Institute of Nanotechnology
Alberto Bosio received his MSc (2003) and PhD (2006) in Computer Engineering in the area of digital systems dependability at the Politecnico di Torino (Italy).
He is now a Full Professor at Ecole Centrale de Lyon, Institue of Nanotechnology (France). His research activities are related to the design and test of advanced digital circuits and systems. He served as committee and organizing member in several international conferences including DATE (Track Chair) and ETS (Program Chair) as well as guest editors for many international journals.
He is a member of the IEEE and the Vice-Chair of the Europeen Test Technical Technology Council.

Dr. Celia Cintas, Research Scientist, IBM Research Africa – Nairobi
Celia Cintas is a Staff Research Scientist at IBM Research Africa, based in Nairobi, Kenya. She is a member of the AI Science team at the Lab.
Her current research focuses on the improvement of ML techniques to address challenges on global health in developing countries and exploring subset scanning for anomaly detection under generative models. Previously, grantee from the National Scientific and Technical Research Council (CONICET), working on Deep Learning for populations studies at LCI-UNS and IPCSH-CONICET (Argentina) as part of the Consortium for Analysis of the Diversity and Evolution of Latin America (CANDELA). During her PhD, she was a visitor student at the University College of London (UK). She was also a Postdoc researcher visitor at Jaén University (Spain).
She holds a Ph.D. in Computer Science from Universidad del Sur (Argentina). Co-chair of several Scipy Latinamerica conferences, Financial Aid Co-Chair for the SciPy (USA) Committee (2016-2019), and Diversity Co-Chair for SciPy (2020-2022).
Workshop Co-chair at ICLR 2023, Diversity Co-chair for ISBI-IEEE 2023 and 2024, MICCAI 2026, among others. A detailed list of talks, publications and events is available at https://celiacintas.io/

Dr. Yazan Mualla, University of Technology of Belfort-Montébliard (UTBM), France
Dr. Yazan Mualla is an Associate Professor at the Belfort-Montbéliard University of Technology, France.
He is the head of the specialty program: Software development. Additionally, he is a member of the “Connaissance and Intelligence Artificielle Distribués” (CIAD) research laboratory, and he participated in several national, European, and international projects. He has authored more than 50 research articles in international journals, conferences, and workshops.
Dr. Mualla received his Ph.D. degree in computer science from the Bourgogne-Franche-Comté University, France. His research interests include Explainable artificial intelligence (XAI), Human-machine/computer/robot/agent interaction/collaboration, Multi-agent systems, Agent-based modelling and simulation, and Intelligent transport systems.

Michael Boone, Manager, Trustworthy AI Product, NVIDIA . USA.
Michael Boone is NVIDIA’s Manager for Trustworthy AI Product. He is responsible for helping ensure NVIDIA’s technology is developed according to its guiding principles and driving implementation for building products and integrating processes to accomplish this.
His roadmap includes multimodal processes, tools, products, and services to enable the company, customers, and the larger ecosystem. Beginning as a licensed professional civil engineer, Michael pivoted from transportation infrastructure project management delivery, operations, and consulting to owning NVIDIA’s global core computer vision product marketing strategy portfolio and product feature definition and interaction for DRIVE AV.
Inventor and car enthusiast, Michael is a highly trusted collaborator, partner, and voice in deploying emerging technology in public and private settings. Michael has been with NVIDIA since 2020. Michael holds his bachelor’s degree in civil engineering from the University of Virginia and his master’s degree in Management from Wake Forest University.

Dr. Mourad Zerai, Associate Professor, ESPRIT School of Engineering Mourad Zerai is an Associate professor and Director of the Quality and Institutional Effectiveness office at ESPRIT School of Engineering.
He is also a research associate at the “Laboratoire de Modélisation Mathématique et Numérique pour les Sciences de l’Ingénieur” (LAMSIN) of the Ecole Nationale d’Ingénieurs de Tunis (ENIT).
He is a University Ambassador of the NVIDIA Deep Learning Institute. Mourad research interest is on the application of image processing, mathematical modeling, machine learning and AI approaches to address real-world problems. His current research area is the development of machine learning methods on constrained data that live in Riemannian manifolds. Dr Mourad holds an Engineering degree and a PhD in applied mathematics from ENIT.

Day 2: Hands-on XAI Workshop

Day 2 – Thursday, February 13th

Practical Workshop: Hands-on with eXplainable AI tools and techniques

Duration: 7 hours
Format: Onsite training at ESPRIT School of Engineering Campus
Location: ESPRIT School of Engineering, Bloc E, Salle EMBA-E10
Certificate: Participants will be awarded a certificate of participation issued by ESPRIT, recognizing their attendance and active engagement in the workshop.

Coverage*:

Model-agnostic explainability approaches: (Duration: 3h)
  • Presentation of LIME, SHAP and Anchor approaches
  • Hands-on with XAI open-source libraries: Alibi, Dalex and AIX360
  • Applications to Real-world case studies
Model-specific explainability approaches: (Duration: 3h)
  • Presentation of LRP, Integrated Gradient, Grad-CAM and Deep LIFT approaches
  • Hands-on with XAI open-source libraries: Captum and TF-Explain
  • Applications to Real-world case studies
Tools for LLM Interpretability: (Duration: 1h)
  • A brief introduction to LIT, Phoenix, and Comagra as a preview of interpretability techniques for LLMs.

Pre-requisites:

  1. 1. Foundational knowledge of Machine Learning (ML) models and the basics of Deep Learning.
  2. 2. Familiarity with Python programming and PyTorch/TensorFlow frameworks.

Requirements:

  • All attendees are required to bring their own laptops to fully participate in the hands-on activities and interactive sessions.

Language of instructions:

  • English

Registration fee:

  • 250 DT (for professionals)
  • 50 DT (for students)
Registration link (limited places): https://forms.office.com/pages/responsepage.aspx?id=lhpPYOjL-EOrv_jq9dhXMJR1cjIf8L1AmEL0NBE8wt1UNFZZMEIxMjdGOUE1U1RVUFJSV1NGWUJPNi4u&origin=lprLink&route=shorturl
Registration deadline: Sunday February 9th at midnight.
Payment: Only online payment via bank transfer to ESPRIT’s designated account is accepted. Payment details, including the account information, will be shared with the shortlisted applicants.

Workshop Planning*

Morning Session Afternoon Session
8:30 – 9:00 Registration & Setup 13:00–16:00 Model-specific explainability approaches
9:00 – 12:00 Model-agnostic explainability approaches 16:00–17:00 LLM Interpretability
12:00 – 13:00 1 hour Break 17:00–17:30 Wrap-up and Closing remarks
* Subject to minor changes without prior notice
Registration link (limited places): https://forms.office.com/pages/responsepage.aspx?id=lhpPYOjL-EOrv_jq9dhXMJR1cjIf8L1AmEL0NBE8wt1UQ08zSzhIQ1JaVldJQUlCNUoxVjRLUjFHTi4u&origin=lprLink&route=shorturl
Registration deadline: Sunday February 9th at midnight.
Registration link (limited places): https://forms.office.com/pages/responsepage.aspx?id=lhpPYOjL-EOrv_jq9dhXMJR1cjIf8L1AmEL0NBE8wt1UNFZZMEIxMjdGOUE1U1RVUFJSV1NGWUJPNi4u&origin=lprLink&route=shorturl
Registration deadline: Sunday February 9th at midnight.
Faouzi Kammoun Co-Chair WEITA 2025
Phone number : 99423944
Email : weita@esprit.tn

Background

Artificial Intelligence (AI) systems have become an integral part of modern society, with their deployment spanning numerous critical applications such as healthcare, finance, energy, transportation, cybersecurity, justice systems, and manufacturing. Many of these AI-driven systems are designed to assist in or directly make decisions that can have far-reaching implications. Hence the need for AI models that are not only accurate but also explainable and interpretable.

The advantages of explainability and interpretability in AI are multifaceted: They provide deeper insights into the decision-making processes of AI models, enabling stakeholders to understand the rationale behind the outcomes. Transparent models enhance accountability by making AI systems interpretable and explainable to human users. This builds confidence in AI systems, fostering fairness, trust and user acceptance.

However, the complexity and opacity of many AI models—particularly those leveraging deep learning models—present significant challenges. These “black box” systems often operate with little or no transparency, making it difficult to apprehend how and why specific decisions are made. This lack of transparency can lead to biased outcomes, raise ethical concerns, and ultimately erode trust. Addressing these issues requires rigorous approaches that make AI systems explainable and interpretable.

Aims

This workshop on eXplainable AI (XAI) aims to explore cutting-edge methods, tackle existing challenges and risks, and discuss practical applications of XAI to address the pressing need for transparency and trust in AI systems.

By bringing together researchers and practitioners, this workshop aspires to provide an interdisciplinary forum for sharing recent research advancements in XAI approaches, best practices, design principles, open challenges, inherent risks, and practical applications. The program will feature keynote presentations, open discussions, idea exchanges, and practical, instructor-led hands-on training sessions.

Key objectives

The primary objectives of this workshop are as follows:

  • Capacity Building: Developing expertise in XAI among researchers and practitioners working in the fields of AI and human-computer interaction.
  • Awareness Raising: Highlighting the pivotal role of explainability and interpretability as essential pillars of human-centered, responsible, and trustworthy AI.
  • Application Exposure: Showcasing the diverse range of use cases and application areas where XAI can create transformative impact.
  • Critical Reflection: Engaging participants to reflect upon the ethical, legal, security, and social concerns posed by the lack of transparency in AI systems.
  • Collaboration and Exchange: Seeding collaboration and promoting the exchange of ideas, perspectives, and experiences to advance research in XAI.
  • Experiential Learning: Immersing participants in practical applications of XAI by equipping them with hands-on experience in implementing XAI techniques to enhance model transparency, interpretability, and trustworthiness.
General Themes:
The event is structured around four main themes:
  • Technical aspects of XAI (methods, frameworks, and approaches).
  • XAI use cases and application areas.
  • Ethical, security, and legal issues surrounding XAI.
  • Hands-on practical training on XAI techniques.

Workshop Format

The two-day workshop on XAI will be organized as follows:

DAY 1- February 12th: Onsite event with hybrid online participation (Free registration)

  • The first session (half day) will draw attention to the state-of-the art in XAI concepts and approaches.
  • The second session (half day) will provide a deep understanding of the use cases and applications of XAI, while addressing the potential ethical, security, and legal concerns.

DAY 2- February 13th: Onsite Participation only (Registration fee applies)

  • This full-day onsite event consists of an instructor-led hands-on practical training in XAI that aims to train participants on how to apply XAI techniques to understand Neural Network decisions, and boost transparency, fairness, accountability and trust in AI decision making.

Note: Attendance on the first day, whether onsite or online, is free of charge. However, participation in the second day’s practical XAI training requires a registration fee and is conditional upon attendance on the first day.

General Co-Chairs

Faouzi KAMOUN - ESPRIT

Lamjed BETTAIEB - ESPRIT

Organizing Committee

Abderrazek HACHANI

Anis ZOUAOUI

Aymen BEN BRIK

Meriem CHICHTI

Mekki KSOURII

Heni ABIDI

Scientific Committee:

Mourad Zarai

Mohamed Tabaa

Heni Abidi

Mohamed Hedi Riahi

Communications & Outreach Committee

Chiraz GHARBI

Nebgha AYACHI

Speakers


TITLE: BRIO: A Bias and Risk Assessment Formal Methodology and Tool

Prof. Giuseppe Primiero, University of Milan, Italy

ABSTRACT:

Phenomena of bias by AI systems based on machine learning methods are well known and largely discussed in the literature. A variety of tools are being developed to assess these undesirable behaviours. In this talk I present a bias and risk assessment formal methodology and tool developed within the BRIO Research Project (https://sites.unimi.it/brio/).The tool is based on various formal logics and works on the I/O data of a ML system remaining agnostic on the model itself. The user can choose one of two distinct modules to evaluate either the difference in behaviour that the model displays on outputs produced by subclasses of inputs, or to evaluate against a desirable output. The type of distance and the threshold for admissible distance from the target distribution can also be selected. The result is a set of all the features and combinations thereof that produce violations with respect to the target distribution. These features can be fed into a risk function which computes an overall value weighting them on parameters such as size of the population and number of features involved, mapping naturally into notions of group and individual fairness

Reliable and Efficient hardware for Trustworthy Deep Neural Networks

Prof. Alberto Bosio, University of Lyon - Lyon Institute of Nanotechnology

ABSTRACT:

Deep Neural Networks (DNNs) are amongst the most intensively and widely used predictive models in machine learning. Nonetheless, increased computation speed and memory resources, along with significant energy consumption, are required to achieve the full potentials of DNNs. To be able to run DNNs algorithms out of the cloud and onto distributed Internet-of-Things (IoT) devices, customized HardWare platforms for Artificial Intelligence (HW-AI) are required. However, similar to traditional computing hardware, HW-AI is subject to hardware faults, occurring due to process, aging and environmental reliability threats. Although HW-AI comes with some inherent fault resilience, faults can lead to prediction failures seriously affecting the application execution. Typical reliability approaches, such as on-line testing and hardware redundancy, or even retraining, are less appropriate for HW-AI due to prohibited overhead; DNNs are large architectures with important memory requirements, coming along with an immense training set. This talk will address these limitations by exploiting the particularities of HW-AI architectures to develop low-cost and efficient reliability strategies.  

Title: Component-level Explanation and Validation of AI Models

Prof. Wojciech Samek, Technical University of Berlin & Fraunhofer Heinrich Hertz Institute (HHI), Berlin, Germany.

ABSTRACT:

Human-designed systems are constructed step by step, with each component serving a clear and well-defined purpose. For instance, the functions of an airplane’s wings and wheels are explicitly understood and independently verifiable. In contrast, modern AI systems are developed holistically through optimization, leaving their internal processes opaque and making verification and trust more difficult. This talk explores how explanation methods can uncover the inner workings of AI, revealing what knowledge models encode, how they use it to make predictions, and where this knowledge originates in the training data. It presents SemanticLens, a novel approach that maps hidden neural network knowledge into the semantically rich space of foundation models like CLIP. This mapping enables effective model debugging, comparison, validation, and alignment with reasoning expectations. The talk concludes by demonstrating how SemanticLens can help in identifying flaws in medical AI models, enhancing robustness and safety, and ultimately bridging the “trust gap” between AI systems and traditional engineering.

Title: A tale of adversarial attacks & out-of-distribution detection stories in the activation space

Dr. Celia Cintas, Research Scientist, IBM Research Africa – Nairobi

ABSTRACT:

Most deep learning models assume ideal conditions and rely on the assumption that test/production data comes from the in-distribution samples from the training data. However, this assumption is not satisfied in most real-world applications. Test data could differ from the training data either due to adversarial perturbations, different hardware devices, diverse patient populations with different or unknown conditions samples, generated content, noise, or other distribution changes. These shifts in the input data can lead to classifying unknown types, classes that do not appear during training, as known with high confidence. On the other hand, adversarial perturbations in the input data can cause a sample to be incorrectly classified. In this talk, we will discuss approaches based on subset scanning methods from the anomalous pattern detection domain and how they can be applied over off-the-shelf DL models.

Title: Explainable AI from Classification to Generation

Dr. Amit Dhurandhars, Principle Research Staff Member, IBM T.J. Watson, Yorktown Heights NY. USA.

ABSTRACT:

Over the past decade XAI has focused mainly on supervised learning problems. However, with the explosion of Large Language Models in the last couple of years explainability for generative tasks has become an important problem. In this talk, I will present two directions that we have been pushing extending approaches previously built for classification to now generation. One is a post-hoc explainability approach based on popular paradigm of contrastive/counterfactual explanations. The other is a directly interpretable approach generalizing CoFrNets a novel continued fraction inspired architecture that we had proposed few years back. Both of these complimentary directions aim at making the black-box nature of these ever more opaque models increasingly transparent.

Title: Human-Agent Explainability Architecture: Application of Remote Robots

Dr. Yazan Mualla, University of Technology of Belfort-Montébliard (UTBM), France.

ABSTRACT:

Recent studies in the goal-driven eXplainable AI (XAI) domain have confirmed that explaining the agent’s behavior to humans fosters the latter’s understandability of the agent and increases its acceptability. However, providing overwhelming information may also confuse human users and cause misunderstandings. For these reasons, the parsimony of explanations has been outlined as one of the key features facilitating successful human-agent interaction with a parsimonious explanation defined as the simplest explanation that describes the situation adequately. This keynote discusses a context-aware and adaptive formulation of parsimonious explanations and proposes a Human-Agent Explainability Architecture (HAExA). The latter relies first on generating normal and contrastive explanations and second on updating and filtering them before communicating them to the human. To evaluate HAExA, we design and conduct empirical human-computer interaction studies employing agent-based simulation of remote robots.

Title: Why Explainability Matters

Dr. Mourad Zerai, ESPRIT School of Engineering, Tunis, Tunisia.

ABSTRACT:

In a world increasingly shaped by AI, explainability is no longer optional-it’s essential. Regulations like the EU’s GDPR and AI Act, or the U.S. Equal Credit Opportunity Act, require AI systems to be transparent, especially in sensitive industries like healthcare, finance, and law enforcement. These laws aim to ensure fairness, accountability, and trust by demanding that AI decisions are understandable to humansThis talk breaks down what explainability means in practical terms, why regulators insist on it, and how it impacts both companies and individuals. From protecting rights to improving trust in technology, we’ll explore why making AI less of a “black box” benefits everyone.

Title: Trustworthy AI at NVIDIA

Michael Boone, Manager, Trustworthy AI Product, NVIDIA, USA.

ABSTRACT:

This presentation discusses what Trustworthy AI (TAI) means, what we are doing, and how we are deploying trustworthiness as an embedded property of technology for us, our customers and partners, and the greater ecosystem.

Title: The impact of Explainable AI models in enhancing interpretability and transparency for Cybersecurity and Digital Forensics

Pr. Farkhund Iqbal, College of Technological Innovation, Zayed University, UAE.

ABSTRACT:

Many computational and machine learning models operate as ‘black boxes,’ often delivering outputs—like classification accuracies—that are difficult for even experts to interpret. Cybersecurity professionals, vulnerability analysts, and risk analysis teams frequently struggle to trace how these models arrive at their conclusions, complicating efforts to predict and prepare for future attacks. This challenge also extends to the courtroom, where judges and police officers must understand and trust digital evidence to make informed decisions, yet often find themselves at a loss. Explainable AI (XAI) has become an invaluable tool in addressing these challenges. By enhancing the transparency and interpretability of AI models, XAI demystifies the results these technologies produce. This presentation will explore the critical roles that both supervised and unsupervised XAI models play in digital forensics. Supervised learning models, which are trained with labeled data, are particularly effective in areas like malware detection and email phishing identification. They provide detailed explanations of their predictions, making the evidence accessible and understandable for non-technical audiences. We will also discuss the tools and techniques used to implement XAI in cybersecurity, including popular frameworks like LIME and SHAP for supervised models, and feature important methods for unsupervised models. Ultimately, the use of explainable models in digital forensics does more than improve interpretability; they are essential in ensuring that justice is effectively served in our increasingly digital world.

TITLE: Interpretable Image Classification Through an Argumentative Dialog Between Encoders

Prof. Wassila Ouerdane, CentraleSupélec, Paris-Saclay University, France UAE.

ABSTRACT:

We address the problem of designing interpretable algorithms for image classification. Modern computer vision algorithms implement classification in two phases: feature extraction – the encoding – that relies on deep neural networks (DNN), followed by a task-oriented decision – the decoding – often also using a DNN. We propose to formulate this last phase as an argumentative DialoguE Between two agents relying on visual ATtributEs and Similarity to prototypes (DEBATES). DEBATES represents the combination of information provided by two encoders in a transparent and interpretable way. It relies on a dual process that combines similarity to prototypes and visual attributes, each extracted from an encoder. DEBATES makes explicit the agreements and conflicts between the two encoders managed by the two agents, reveals the causes of unintended behaviors, and helps identify potential corrective actions to improve performance. The approach is demonstrated on two problems of fine-grained image classification.