Biologically Inspired Information Fusion
22 August 2006 - 23 August 2006
The ability to process, interpret and act upon sensory information is perhaps one of the most remarkable aspects of human and animal cognition. Our sensory systems process large volumes of information at different scales in short periods of time, far out-performing current artificial systems, which struggle to usefully process just a single modality of information. For example, whereas speech recognition systems have achieved real-time continuous operation, artificial systems, designed for vision or olfaction are far less advanced, yet the combination of different information sources, or senses, may help overcome some of the processing limitations. This disparity between natural and artificial cognitive systems has been recognised in the recent UK Foresight Cognitive Systems Review, which suggests that our understanding of both natural and artificial systems of sensory processing can be significantly improved through collaboration between life and physical scientists.
The workshop was held to address this issue, attempting to bring together both life and physical scientists to discuss research from the perspective of the different disciplines, focused on the common theme of information fusion. The aim was to promote collaboration between the disciplines to develop an understanding of how to build adaptive information fusion systems. This initial workshop was targeted at bringing the disciplines together to develop a set of research priorities for future collaboration. This was achieved through a two-day programme of tutorials, discussion sessions, student presentations and brainstorming held on the 22nd and 23rd August 2006.
The workshop programme, proceedings, notes from sessions, and full workshop report can be found here. The workshop was sponsored by the University of Surrey's Institute of Advanced Studies and the EPSRC under grant number EP/E012795/1.
The workshop successfully attracted 47 people. Of these 15 were from the life sciences and 30 were from the physical sciences. A good mix of backgrounds was also achieved, with 18 students and 22 academics, together with 7 other representatives from commercial organisations, the scientific press, clinicians and teachers. Participation was also drawn from a number of countries, with representation from the USA (3), Italy (3), Germany (2), France (1), Turkey (1) and Australia (1), with the remaining 36 from the UK.
Contributions to the workshop were invited from researchers and students for discussion sessions and student papers. Submissions were peer reviewed by two members of the inter-disciplinary community. A policy of assigning reviewers from two different disciplines was adopted to help focus the material on cross-discipline understanding. Peer review feedback was used to select topics that would provoke discussion. Many of the reviewers’ comments were used to help authors improve on the discussion potential of the work. In general the standard of the student papers was very high, with each receiving favourable feedback on their work.
The programme was designed to assist in cross-discipline training, provoke discussion and to foster collaboration. This was achieved through the following activities:
- Invited tutorials from discipline leaders. Research leaders in the different disciplines were asked to summarise current knowledge of their field. Rising to this challenge were Barry Stein (Wake Forest University School of Medicine), Gemma Calvert (University of Bath), Charles Spence (Oxford University), John Foxe (City College of New York), Belur Dasarathy (Editor-in-Chief of the Elsevier Information Fusion Journal) and Gerard McKee (University of Reading).
- Discussion sessions. Resulting from the submitted discussion abstracts, researchers were asked to promote discussion of controversial ideas and to pose currently unanswered questions related to the workshop aim. The ideas were typically introduced in a short presentation by the researcher, and followed by an open discussion session. Discussion sessions were lead by Alex Thomson (London University), Hans Colonius (Oldenburg University), Bob Damper (University of Southampton) and Leslie Smith (University of Stirling).
- Research student presentations. Resulting from the submitted student papers, students were asked to summarise their current work related to the workshop aim. Each student gave a short presentation, and after all of the presentations an open discussion session was held, with a focus on discussing the presented ideas and giving feedback. Such presentations were restricted to students to promote the discussion of new ideas by young researchers. Presentations were given by Basabdatta Sen (University of Manchester), Timothy Dixon (University of Bristol) and Craig Bennett (University of Surrey).
- Workshop sessions. To bring together the ideas discussed in the tutorials, discussions and presentations, three workshop brainstorming sessions were held across the two days. At each session, participants were asked to answer key questions to help build a set of research priorities.
A highlight of the workshop was the tutorial given by Barry Stein, which was rated as the best session by 81% of the respondents to the workshop evaluation questionnaire. This reflects on Barry’s ability to present ideas, backed up by the impressive set of experiments reported. Also mentioned for the best session were Gemma Clavert, Chalres Spence and John Foxe for the psychology tutorial, and Alex Thomson, Bob Damper and Leslie Smith for their discussions.
The three brainstorming sessions held throughout the workshop were used to foster collaboration and to determine a set of community research priorities and their associated objectives. The sessions were run in groups that were randomly selected, but with a bias to ensure that each group had a good mix from the different disciplines, and with background as academics, students, and where possible, commercial representatives.
The collated set of the top three research priorities and associated objectives identified by the participants were:
1. Sensory fusion, disorder and clinical application
How does sensory fusion impact on disorder and how can this be applied? In particular, can we develop machine aids in the form of implants or prosthetics to overcome or reduce the effects of disorders? For example, hearing impaired people perform sub-optimally at using visual cues to amplify auditory cues. Could a training regime been developed to enhance the operation of multi-sensory integration to overcome the sub-optimal performance on visual cues?
Objectives identified are:
- Understand further how the brain achieves integration and what effect this has on particular disorders. For example, what role does filtering play in, say Schizophrenia, where multi-modal neurons may help in filtering important sensory information that are impaired in disorder?
- Evaluate the effectiveness of input channels and the resulting integration. Intuitively, systems that fuse information (for example medical diagnosis support) often result in much of the source information being discarded.
- Construct models of multi-sensory brain function and perform lesioning studies on the models to understand processes better.
- Explore whether brain function could be altered to compensate for deficits, perhaps through the use of reactive robots to stimulate patients.
2. Exploiting effective biological processes for sensory integration
What biological processes can be exploited by computer systems? What do biological sensory systems do well, and conversely what do computer sensory systems do well? Which biological sensory systems are optimal? Is fusion of benefit or not? Which senses dominate under which circumstances? How does memory impact on sensory integration?
Objectives identified are:
- Understand spatiotemporal dynamics of integration by applying non-linear dynamics concepts to models.
- Understand which biological fusion processes can be modelled in a computer system, focusing on those that appear optimal and taking into account sensory dominance and its relevance to computation.
- Explore how biological systems are robust to context in order to translate this to making computer systems less dependent on situation and not optimised for particular environments or contexts.
- Explore how memory impacts on sensory integration and perception.
Developing a common language for inter-disciplinary communication and collaboration
In order to develop collaboration further, a common language is required to overcome terminology differences and to assist in cross-discipline training and dissemination.
Objectives identified are:
- Provide further training to overcome discipline dominance and to break down barriers. Examples include repeating a similar workshop annually.
- Understand and promote the benefits of inter-disciplinary work.
The aim of the workshop was to provide a forum in which the different disciplines researching multi-sensory/information fusion could learn from each other. Specific objectives were targeted at identifying research priorities in the field and developing long-term collaboration. Taking into account the devised programme, outcomes of sessions and evaluation feedback, this aim has been met through this initial workshop. Follow-up work is targeted at establishing long-term collaboration.
Specifically, activities in progress are:
- Dissemination of the workshop findings and notes. Notes, references and other workshop materials are already available via the website, and a summary of the findings in the workshop report will be distributed via mailing lists and contacts.
- Journal special issues to provide further training and dissemination, and to promote collaboration. Currently a special issue each of the Perception and Information Fusion journals is being arranged for publication in 2007, covering the dissemination routes for both life and physical scientists.
- Establishment of follow-up projects. The identified research priorities and objectives will be used to promote follow-up projects to reinforce collaboration. In the first instance participants will be contacted to highlight opportunities and encourage interaction. A mailing list will be established to support this. Participants will be encouraged to suggest ideas via the mailing list to further promote collaboration. Specific projects will also be targeted by the workshop chairs with key collaborators. When details are known, these will also be advertised.
We would like to thank all those who took part in the workshop, especially those reviewing abstracts and papers. A special vote of thanks goes to Sophie Gautier and Heather Norman for all their effort in making the event go as smoothly as possible. Thanks also go to the Institute of Advanced Studies at the University of Surrey, especially Nigel Gilbert, and the Engineering and Physical Sciences Research Council (grant number EP/E012795/1) for providing funds to support this initiative.
Dr Matthew Casey