Scroll down for updates.
Around a hundred members of the US/UK Distributed Analytics and Information Sciences International Technology Alliance (DAIS ITA) are participating in the programme’s Annual Fall Meeting (AFM) in Weybridge, Surrey this week. The AFM provides an opportunity for the 15-partner consortium to meet together to share their latest research results and advance their collaborations. This year, the team will also receive feedback from the programme’s international Peer Review panel.
The DAIS programme is carrying out research to provide the fundamental scientific underpinnings to enable multi-partner coalition operations to exploit artificial intelligence, machine learning, and collective intelligence analytics technologies in front-line situations.
On Tuesday, members of the fourteen-strong Cardiff DAIS team are presenting highlights of their research over the past year including:
-- Making coalition information systems adaptable to the needs of users at the front line, and able to fully exploit a data analytics environment that is mobile, dynamic, unstable, and very much unlike a traditional data centre.
-- Providing assured predictive analytics operating synergistically between users and machines, using recent advances in artificial intelligence and machine learning.
-- Understanding complex adaptive human groups, their behaviours and how they evolve over time.
The DAIS consortium is training a cohort of over 40 PhD students across the 15 US and UK partners. The AFM includes an annual student cohort event, organised by IBM and aimed at building the collaborative community: this year this involved a boat trip on the river Thames.
Dave Braines from IBM Research (part-time CSRI PhD student) presents recent results on new techniques for social network analysis using ‘motifs’, which are small fragments of a social network graph that signify features of interest. This work focuses in particular on efficiently detecting features in large social networks that change over time (compared to processing the whole network). The paper, co-authored with Dr Liam Turner in CSRI and our colleagues at the University of Massachusetts Amherst, is available at https://arxiv.org/abs/1812.05473
CSRI DAIS PhD student Laura D’Arcy presents her poster on how we use reinforcement learning to learn data analytics workflows for our “distributed brain” architecture. The new idea here is to use deep Q-learning, a form of reinforcement learning based on deep neural networks, to generate directed graphs as needed for workflow pipelines. An earlier version of the work was presented in Long Beach, California at the International Conference on Machine Learning this summer https://arxiv.org/abs/1906.02280
Cardiff PhD student Liam Hiley demonstrates the “discriminative relevance” technique recently presented in Macau, China at the IJCAI 2019 Explainable Artificial Intelligence workshop -https://arxiv.org/abs/1908.01536. Here, the technique is used to explain decisions made by a deep neural network processing webcam footage. The person in the video performs a punching action which the AI software explains by highlighting the fist and head. Liam’s technique is novel in that it separates the relevant temporal and spatial features of the video: for example, the fist is important because it moves quickly when the person punches.