Skip to content

Who we are

The HES-SO

The HES-SO is a Swiss university of applied sciences that offers various academic programs, research projects, and services in the fields of technology, design, economy, and social work. It has a strong focus on practical and applied knowledge, and its mission is to contribute to the economic and social development of Switzerland through its research and innovation activities.

Regarding the advancement of machine learning, the HES-SO is a leading institution in Switzerland that provides high-quality education, research, and development in this field. It has a team of experienced and knowledgeable professors, researchers, and practitioners who work on cutting-edge projects and initiatives in collaboration with industry partners and other academic institutions.

The HES-SO also has a strong network of alumni and professional contacts in the industry, which ensures that its graduates are well-prepared and have excellent job prospects in the machine learning field.

Overall, the HES-SO is a dynamic and innovative institution that contributes to the development of machine learning in Switzerland and beyond.

Team members

Prof. Carlos Andrés Peña-Reyes

PhD, Director of the Institute of Information and Communication Technologies (IICT), HEIG-VD

Team member

Professor Peña has extensive expertise in Explainable Artificial Intelligence (XAI). Across several CTI and SNSF projects (DiagnoSuite, BossExplorer, nanoFUGE, ISyPeM) he has developed and successfully applied his interpretable modelling approach, dubbed Fuzzy CoCo and based on evolutionary algorithms and fuzzy logic, to several diagnostic and biomarker-discovery tasks. Based on this algorithm he has published open-source packages (Trefle1, fugeR2) and produced an unpublished package that led to the creation of SimplicityBio, a company today integrated into Precision QuartzBio and that also led to the obtention of a patent (WO/2017/199067).

From 2016 to 2018, with the project "D-Rex: Rule extraction from deep neural networks", (Hasler foundation grant)3, 4, his group focused on the investigation of knowledge extraction methods from deep neural networks (CNNs). The methods developed in this project were able to extract simple rules based on high-level features4, extract class-relevant input features and highlight them in specific inputs 4, and recreate prototypical images for specific classes or internal features allowing users to gain insights into the inner representations of the trained network4.

Since 2020, the team participates to the SNSF NRP77 project "EXPLaiN: Ethical and Legal issues of Mobile Health Data – Improving understanding and eXPlainability of digitaL transformAtion and data technologies using artificial IntelligeNce", which focus on the implementation and test of explainable models in the context of mobile health data.

Prof. Marcel Graf

PhD, Professor, HEIG-VD

Team member

Professor Graf has more than 20 years of experience in software engineering, including 16 years at IBM Research. He has developed digital platforms that give structure to technical expertise and make it available to a broader non-technical audience5. In an Innosuisse project with Avalia Systems he has developed a platform that allows experts in software engineering to organise and streamline their work in software assessments and communicate their insights to business clients. The platform can be flexibly deployed on Avalia's or the client's IT. In another Innosuisse project with Flybotix he is developing a cloud-based platform for drone inspection flights. The platform uses short-lived massive cloud resources for photogrammetric processing with low latency and enables remote inspection experts to follow and supervise a drone flight remotely.

Prof. Guido Bologna

PhD, Professor, HEPIA-GE

Team member

Professor Bologna introduced a neural network model from which propositional rules are generated 6. These rules are natural with respect to the logic of human reasoning. A clear advantage of the introduced model, which is called Discretized Interpretable Multi Layer Perceptron (DIMLP) is that for a given input, we can refer the activated rules and determine range inputs relevant to the decision. Later, the DIMLP transparency framework was applied to ensembles 7. A rule extraction technique that is essential to the explicability of deep Multi-Layer Perceptrons (MLPs) was presented in 8. Subsequently, a comprehensive comparison of rule extraction techniques was proposed for 25 classification problems 9. The compared models were ensembles of DIMLPs, ensembles of decision trees and Support Vector Machines (SVMs). Later, in 10 the rule extraction problem was tackled from a simple convolutional architecture with respect to textual data and melanoma images 11. Finally, in 12, he introduced in collaboration with Jean-Marc Boutay and Damian Boquete Costa, a newly developed algorithm called Fidex, as a significant advancement in rule extraction. Fidex is notably faster and capable of deriving rules from data trained by various models, including Convolutional Neural Networks (CNNs). Many of about 100 papers he has authored or co-authored are available on ResearchGate 13.

Xavier Brochet

Senior R&D Scientist in Bioinformatics, PhD, HEIG-VD

Team member

Senior bioinformatician with over 12 years of experience, Xavier Brochet is passionate about working at the interface of disciplines and he enjoys solving real-world problems, specially those related with life sciences and biomedical. He has a significant experience in public research and has been working in multidisciplinary teams (biology and computer science). Working within a stimulating research framework and with researchers from other disciplines is particularly adapted to his education and is a significant motivation for him. He is an expert in creating information systems and developing bioinformatics analysis tools, softwares and web interfaces.

Arthur Babey

Ra&D Collaborator, HEIG-VD

Team member

Arthur Babey is a Life Sciences Engineer working as an Ra&D Collaborator at HEIG-VD. His expertise in Life Sciences, coupled with a strong inclination towards data science and machine learning, equips him with a distinctive set of skills. Currently, he is focused on the domain of machine learning and explainable artificial intelligence. By utilizing computational techniques and data science, he is deeply interested in solving intricate problems related to health and biology. Whether it involves the complexities of bioinformatics or the potential of deep learning, Arthur strives to push the boundaries of innovation and make meaningful contributions to the field.

Jean-Marc Boutay

HES Assistant, HEPIA-GE

Team member

Jean-Marc Boutay is an HES Assistant at HEPIA-GE and a proficient Machine Learning Engineer. He obtained his master's degree in computer science in 2022, focusing his thesis on real-time object detection using deep learning. He is currently developing and optimizing cutting-edge algorithms in Python and C++. He developed the Fidex12 algorithm in collaboration with Damian Boquete Costa. He is a serious-minded individual who takes his work very seriously and is always striving for excellence in his projects. Jean-Marc is a great team player and is always willing to lend a helping hand to his colleagues whenever needed.

Rémy Marquis

Ra&D Collaborator, HEIG-VD

Team member

Rémy Marquis is an R&D Collaborator at HEIG-VD, with a background in sciences and environmental engineering. His growing interest in the capabilities and wide-ranging applications of Machine Learning led him to embark on a gradual transition into the field. Currently, Rémy focuses on Machine Learning Infrastructure Engineering, with a specific emphasis on Machine Learning Operations (MLOps). He continuously expands his expertise in building and managing scalable systems for ML workflows, ensuring seamless integration and deployment of ML models. Rémy's proficiency in MLOps is steadily growing, as he remains committed to delivering high-quality work in this evolving field.

Damian Boquete Costa

HES Assistant, HEPIA-GE

Team member

Damian Boquete Costa is an HES Assistant at HEPIA-GE and a versatile Software Engineer. He earned a Bachelor's degree in Computer Science in 2023 and joined the team later that year to work with Jean-Marc Boutay and developed the Fidex12 algorithm with him. His tasks focus on optimizing Machine Learning algorithms, developing graphic solutions, and enhancing tool documentation. Damian's diverse skills bring fresh perspectives, and he is committed to finding innovative solutions. His motivation and dedication to continuously sharpening his skills are valuable assets to the project's progress.

  1. https://pypi.org/project/trefle/ ↩

  2. https://cran.r-project.org/web/packages/fugeR/ ↩

  3. Gomez Schnyder, Stéphane, Despraz, Jérémie, et Peña-Reyes, Carlos Andrés, « Improving neural network interpretability via rule extraction », Greece, p. 811‑813, octobre 2018. ↩

  4. J. Despraz, S. Gomez, H. F. Satizábal, et C. A. Peña-Reyes, « Exploring Internal Representations of Deep Neural Networks », in Computational Intelligence, vol. 829, C. Sabourin, J. J. Merelo, K. Madani, et K.Warwick, Éd. Cham: Springer International Publishing, 2019, p. 119‑138. doi: 10.1007/978-3-030-16469-0_7. ↩↩↩↩

  5. T. Grandison et al., « Elevating the Discussion on Security Management: The Data Centric Paradigm », in 2007 2nd IEEE/IFIP International Workshop on Business-Driven IT Management, Munich, Germany, mai 2007, p. 84‑93. doi: 10.1109/BDIM.2007.375015. ↩

  6. Bologna, Guido et Pellegrini, Christian, « Three medical examples in neural network rule extraction », 183‑187, 1997. ↩

  7. G. Bologna, « Is it worth generating rules from neural network ensembles? », Journal of Applied Logic, vol. 2, no 3, p. 325‑348, sept. 2004, doi: 10.1016/j.jal.2004.03.004. ↩

  8. G. Bologna et Y. Hayashi, « Characterization of Symbolic Rules Embedded in Deep DIMLP Networks: A Challenge to Transparency of Deep Learning », Journal of Artificial Intelligence and Soft Computing Research, vol. 7, no 4, p. 265‑286, oct. 2017, doi: 10.1515/jaiscr-2017-0019. ↩

  9. G. Bologna et Y. Hayashi, « A Comparison Study on Rule Extraction from Neural Network Ensembles, Boosted Shallow Trees, and SVMs », Applied Computational Intelligence and Soft Computing, vol. 2018, p. 1‑20, 2018, doi: 10.1155/2018/4084850. ↩

  10. G. Bologna, « A Simple Convolutional Neural Network with Rule Extraction », Applied Sciences, vol. 9, no 12, p. 2411, juin 2019, doi: 10.3390/app9122411. ↩

  11. G. Bologna et S. Fossati, « A Two-Step Rule-Extraction Technique for a CNN », Electronics, vol. 9, no 6, p. 990, juin 2020, doi: 10.3390/electronics9060990. ↩

  12. G. Bologna, J.-M. Boutay, Q. Leblanc, et D. Boquete. « Fidex: An Algorithm for the Explainability of Ensembles and SVMs », in International Work-Conference on the Interplay Between Natural and Artificial Computation, p. 378388. Cham: Springer Nature Switzerland, may 2024, doi: 10.1007/978-3-031-61137-7_35. ↩↩↩

  13. https://www.researchgate.net/profile/Guido-Bologna ↩