Several members of the DAIS group presented papers at the annual SPIE Defense + Commercial Sensing event in the session Artificial Intelligence and Machine Learning for Multi-Domain Operations Applications.
Taking place between the 14th to 18th April in Baltimore, Maryland, the event has more than 4,500 attendees and is one of the leading conferences on sensors, infrared technology, laser systems, spectral imaging, radar and LIDAR.
DAIS Research Student Dave Braines from IBM Research UK presented his paper "Achieving useful AI explanations in a high-tempo complex environment”. The paper seeks to address how artificial intelligence (AI) and machine learning (ML) techniques are often inscrutable and hard for users to trust since they lack effective explanations for their outputs. Dave’s work investigates which explanation to choose for a particular user and task, considering information such as the time available to them, their level of skill and the device they are using. By defining a meta-model for AI/ML explanation provision, Dave shows how an interactive conversational interface can deliver explanations to users across a range of situations, datasets and modalities (text, images, audio).
Another member of the DAIS group presenting at the conference was Crime and Security Research Institute (CSRI) PhD Student Iain Barclay. In his paper "A conceptual architecture for contractual data sharing in a decentralised environment”, Iain examined how to provide ‘traceability’ for data in AI/ML systems. This will ensure that whenever a consumers’ data is sourced and used, they have assurance that the data accuracy is as described, has been obtained legitimately and the terms under which it is made available are understood. Iain’s approach uses blockchain-based distributed ledger technology, which can facilitate transactions in situations where parties do not have an established trust relationship or centralised command and control structures.
Richard Tomsett (IBM Research) presented "Uncertainty-aware situational understanding”, a paper co-authored by Dr Federico Cerutti at the CSRI as well as other DAIS colleagues. The work focuses on how AI/ML systems need to be aware of when they are uncertain about a prediction in order to be useful to human users. When observations are out of the ordinary, system confidence decreases because the relevant training data for the machine learning system is significantly smaller than the size of the training data set. The paper proposes new techniques based on subjective and uncertain Bayesian networks be employed to overcome this problem. The merits of these methods were evaluated using a case study developed in collaboration with professional intelligence analysts.
Currently, the only full paper available for download is “A conceptual architecture for contractual data sharing in a decentralised environment”. However, all of the above papers will become available shortly via orca.cardiff.ac.uk.