explAIn
Robustness in Explainable AI
Technical Workshop
26th September 2022
AI systems are increasingly used to make consequential decisions in a wide range of domains, including banking, healthcare and beyond. As a result, several methods and tools have been developed to explain the decisions of these systems, with the aim to deliver explainable AI (XAI) solutions suitable for a human-centred world.
While explanations can improve the interpretability of AI decisions, they have also been shown to lack robustness, e.g., they may produce completely different explanations for similar events. This phenomenon can have troubling implications, as lack of robustness may indicate that explanations are not capturing the underlying decision-making process of a system and thus cannot be trusted.
This workshop aims to bring together researchers from academia and industry working in XAI to explore limits of current explainability methods and discuss the role that robustness may play in delivering trustworthy explanations. The workshop will consist of invited and selected presentations and discussions.
Time (GMT+1) | Talk |
---|---|
09:15 - 09:30 | Registration |
09:30 - 09:45 | Welcome |
9:45 - 10:30 | Invited talk: Rafael Calvo |
10:30 - 10:40 | Supporting the Value-Sensitive Participatory Design of AI-Based Systems, Malak Sadek |
10:40 - 10:50 | Do users care about the robustness of AI explanations? A study proposal, Bence Palfi |
10:50 - 11:00 | Break |
11:00 - 11:45 | Invited talk: Saumitra Mishra |
11:45 - 11:55 | Formalising the Robustness of Counterfactual Explanations for Neural Networks, Junqi Jiang |
11:55 - 12:05 | On interactive explanations as non-monotonic reasoning, Guilherme Paulino-Passos |
12:05 - 12:30 | Group discussion |
12:30 - 13:30 | Lunch |
13:30 - 14:15 | Invited talk: Nicola Paoletti |
14:15 - 14:25 | Sonification in Explainable Artificial Intelligence: An Example Study on COVID-19 Detection from Audio, Alikan Akman |
14:25 - 14:35 | Towards a Theory of Faithfulness: Faithful Explanations of Differentiable Classifiers over Continuous Data, Xiang Yin |
14:35 - 14:45 | Break |
14:45 - 15:30 | Invited talk: Antonia Creswell |
15:30 - 15:40 | Using logical reasoning to create faithful explanations for NLI models, Joe Stacey |
15:40 - 15:50 | A framework for evaluating the cognitive capabilities of AI systems, Ryan Burnell |
15:50 - 16:20 | Group discussion |
16:20 - 16:30 | Concluding remarks |
Room 308, Huxley Building,
180 Queen's Gate,
South Kensington Campus,
Imperial College London,
London, SW7 2RH, UK.
Participation in the workshop is free, subject to availability.
If you would like to give a short presentation around the workshop's topics, please submit a title and a short abstract by Sep 14th, 3:00 pm, using the link below. To attend without presenting, please submit a short description of your research interests and previous (relevant) research, again by Sep 14th, 3:00 pm, using the same link.
We plan to notify presenters and attendees by Sep 16th. Lunch and refreshments will be provided for registered attendees (whether presenting or not).
Event organised by the Centre for eXplainable Artificial Intelligence.