Skip to main content

Birkbeck Knowledge Lab Seminar: Set membership estimation based on interval arithmetic for feature relevance analysis in neural network explainability

Venue: Birkbeck Main Building, Malet Street

In this talk we describe how set membership estimation based on interval arithmetic can be applied to tackle challenges in neural computation and especially help with the problem of explainability-interpretability of neural classifiers’ decisions. The talk will give an overview of relevant concepts and highlight issues in neural computation that can be treated as parameter estimation problems, emphasizing on how uncertainty on the parameters can be quantified in terms of intervals. It will introduce interval methods and present relevant case studies with an emphasis on explainability-interpretability of the neural classifiers’ decisions in terms of feature relevance.

Contact name:

  • Dr Stavros P. Adam -

    Dr Adam serves as an Assistant Professor at the Department of Informatics and Telecommunications, University of Ioannina, Greece. His work focuses on the design of reliable computing algorithms to address neural computing and machine learning algorithms limitations towards effective machine learning in the wild. Before joining academia, he held R&D positions in the cement industry working on projects designing and implementing computer-based systems and application software for the supervision of energy consumption in industrial plants, simulation-based optimization of the use of water resources, equipment preventive maintenance, etc. He also spent 8 years in the defence industry, holding R&D management position of outstanding projects such as, building submarine and vessel simulators, designing secure military communication systems, and developing software radar systems. In 2016 he joined the UK Interval Methods Group (University of Manchester) and a year later the groups MEA (Méthodes Ensemblistes pour l’Automatique) and VS-CPS (Verification et Synthèse de Systèmes Cyber-Physiques) of C.N.R.S., the French National Research Center. More information about his work is available through Google scholar.