Attention Focuses on Focus of Attention at CAMP95 T J Fountain University College London In the magnificent surroundings of Lake Como, Northern Italy, nearly 100 delegates from seventeen countries assembled for the third International Workshop on Computer Architectures for Machine Perception - CAMP95. The meeting, organized by Professor Virginio Cantoni from Pavia University, and sponsored by Pavia University, IEEE, IAPR and ACMSIGART, took place in the opulent Villa Olmo between September 18th and September 20th 1995. Any delegates whose attention strayed momentarily from the current speaker could be inspired to ponder the difficulties of automated perception merely by raising their eyes to the roof of the lecture theater, which bore a strong and glorious resemblance to that of the Cistine Chapel. The delegates heard more than 40 presentations, organized into nine sessions whose titles ranged from real-time systems and techniques to basic parallel primitives and neural computing, and these were supported by a poster session and a discussion panel (of which more later). The overall theme of the conference was the inspiration to be gained (for perceptual architectures) from the human visual system, with particular reference to the role which could be played by focusing attention in order to avoid redundant processing. In particular, two of the invited papers set the main theme for the meeting. Professor Martin Levine (Ontario) explained how the idea of foveation, in which processing is concentrated towards the central area of an image, fosters the idea of focus of attention, and offers important gains in computational efficiency. Professor John Tsotsos (Ontario) concentrated on the use of onset and offset events in a scene as an attention focusing cue, and how this idea was applied to an automatic vision system, particularly by using a variety of spatial and temporal scales. Of the remaining invited papers two (Heirarchical and Modular Attention from Harry Wechsler of George Mason University, and Artificial Life and Evolutionary Computing in Machine Perception from Jean-Claude Heudin of Sodima S.A. in France)supported the general conference theme, whilst the third (Fault Tolerance and Reconfigurability Issues from Professor Stefanelli of Milan) explored the very important issue of ensuring computationally correct results from highly-parallel computers in which faulty elements are almost certain to exist. The discussion forum, which formed the final conference session, whilst ostensibly concerned with the paucity of research on autonomous systems of any sort, generated some controversy concerning simulation. It was agreed amongst members of the panel and the audience that research in this area was at a stage which demanded some form of simulation before valid systems could be constructed. However, there was disagreement (and some confusion) over the benefits to be obtained from hardware emulation as against software simulation. Some opinions suggested that construction of prototype hardware could demonstrate important emergent properties (and problems) which could not have been foreseen, others that software simulations could demonstrate all but the more trivial implementation details. This discussion reached no definite conclusion. As at all worthwhile workshops, much of interest was to be gleaned from informal conversations. Whilst the great majority of technical papers offered an optimistic view of the field, your reporter's concerns about some aspects were echoed in discussions with a surprising number of delegates. Whilst there is no doubt that the human vision system provides an inspinng target for much of this work, most research efforts are understandably focused on near-term attainable goals, and the disparity between the two seems just as great now as it was, say, twenty years ago. Certainly technological advances have been made, and this is one key area where perhaps only time could solve the problems of the past. However, many delegates expressed a worry that few conceptual stepping- stones seem to be available between the present position and our inspinng target. On a more optimistic note, it seems encouragingly possible that the concept of focused attention, so strong a theme at this conference, may provide just such a stepping-stone. At the next CAMP meeting, to be held under the chairmanship of Chip Weems in Boston, Massachusetts in 1997, it is to be hoped that other such concepts will be discussed. Information concerning the 1997 workshop, together with a call for papers, will be promulgated on http://www.cs.umass.edu/-camp97. The 461-page proceedings of CAMP95 are published by the IEEE Computer Society Press (order number PR07134). SIGART Bulletin, Vol. 6 No. 4

/lp/association-for-computing-machinery/attention-focuses-on-focus-of-attention-at-camp95-4wS0t6G0tX