In a recent Physical Review Letters paper, Vicsek et. al. propose a simple but compelling discrete-time model of n autonomous agents fi.e., points or particlesg all moving in the plane with the same speed but with dierent headings. Each agent's heading is updated using a local rule based

task, but differ completely at the levels of synaptic strengths and single-unit responses. However, hierarchical clustering techniques applied to NETtalk reveal that these different networks have similar internal representations of letter-to-sound correspondences within groups of processing units

closes together smoothly when the curve returns to its starting point. A handwritten rather than mechanical style is appropriate for mathematics because people generally create math with pen, pencil, or chalk. The Euler letters are upright, not italic, so that there is a general consistency

Recent studies of eye movements in reading and other information processing tasks, such as music reading, typing, visual search, and scene perception, are reviewed. The major emphasis of the review is on reading as a specific example of cognitive processing. Basic topics discussed with respect to reading are (a) the characteristics of eye movements, (b) the perceptual span, (c) integration of information across saccades, (d) eye movement control, and (e) individual differences (including dyslexia). Similar topics are discussed with respect to the other tasks examined. The basic theme of the review is that eye movement data reflect moment-to-moment cognitive processes in the various tasks examined. Theoretical and practical considerations concerning the use of eye movement data are also discussed. Many studies using eye movements to investigate cognitive processes have appeared over the past 20 years. In an earlier review, I (Rayner, 1978b) argued that since the mid-1970s we have been in a third era of eye movement research and that the success of research in the current era would depend on the ingenuity of researchers in designing interesting and informative

Multiclass learning problems involve nding a de nition for an unknown function f(x) whose range is a discrete set containing k>2values (i.e., k \classes"). The de nition is acquired by studying collections of training examples of the form hx i;f(x i)i. Existing approaches to multiclass learning problems include direct application of multiclass algorithms such as the decision-tree algorithms C4.5 and CART, application of binary concept learning algorithms to learn individual binary functions for each of the k classes, and application of binary concept learning algorithms with distributed output representations. This paper compares these three approaches to a new technique in which error-correcting codes are employed as a distributed output representation. We show that these output representations improve the generalization performance of both C4.5 and backpropagation on a wide range of multiclass learning tasks. We also demonstrate that this approach is robust with respect to changes in the size of the training sample, the assignment of distributed representations to particular classes, and the application of over tting avoidance techniques such as decision-tree pruning. Finally,we show that|like the other methods|the error-correcting code technique can provide reliable class probability estimates. Taken together, these results demonstrate that error-correcting output codes provide a general-purpose method for improving the performance of inductive learning programs on multiclass problems.

INTRODUCTION A logo is "a single piece of type bearing two or more usually separate elements" [1]. In this paper, we use logos to display aligned sets of sequences. Sequence logos concentrate the following information into a single graphic [2]: 1. The general consensus of the sequences. National Cancer Institute, Frederick Cancer Research and Development Center, Laboratory of Mathematical Biology, P. O. Box B, Frederick, MD 21701. Internet addresses: toms@ncifcrf.gov and stephens@ncifcrf.gov. y corresponding author 1 2. The order of predominance of the residues at every position. 3. The relative frequencies of every residue at every position. 4. The amount of information present at every position in the sequence, measured in bits. 5. An initiation point, cut point, or other significant location (if appropriate) . Any aligned set of DNA, RNA or protein sequences can be represented using this technique. CREATION OF BINDING S

> ! "GZip" page 64, Table 2.5, line "progp": "43,379" ! "49,379" page 68, Table 2.6: "Mbyte/sec" ! "Mbyte/min" twice in the body of the table, and in the caption "Mbyte/second" ! "Mbyte/minute" page 70, para 4, line 5: "Santos" ! "Santis" page 71, line 11: "Fiala and Greene (1989)" ! "Fiala and Green (1989)" Chapter Three page 89, para starting "Using this method", line 2: "hapax legomena " ! "hapax legomenon " page 96, line 5: "a such a" ! "such a" page 98, line 6: "shows that in fact none is an answer to this query" ! "shows that only document 2 is an answer to this query" page 106, para 3, line 9: "the bitstring in Figure 3.7b" ! "the bitstring in Figure 3.7c" page 107, Figure 3.7: The coding shown in part (c) cannot be decoded ambiguously. For example, the sequence "1010 0000 0001 0000

Detection of a visual signal requires information to reach a system capable of eliciting arbitrary responses required by the experimenter. Detection latencies are reduced when subjects receive a cue that indicates where in the visual field the signal will occur. This shift in efficiency appears to be due to an alignment (orient-ing) of the central attentional system with the pathways to be activated by the visual input. It would also be possible to describe these results as being due to a reduced criterion at the expected target position. However, this description ignores impor-tant constraints about the way in which expectancy improves performance. First, when subjects are cued on each trial, they show stronger expectancy effects than when a probable position is held constant for a block, indicating the active nature of the expectancy. Second, while information on spatial position improves per-formance, information on the form of the stimulus does not. Third, expectancy may lead to improvements in latency without a reduction in accuracy. Fourth, there appears to be little ability to lower the criterion at two positions that are

The literature on interference in the Stroop Color-Word Task, covering over 50 years and some 400 studies, is organized and reviewed. In so doing, a set ofl 8 reliable empirical findings is isolated that must be captured by any successful theory of the Stroop effect. Existing theoretical positions are summarized and evaluated in view of this critical evidence and the 2 major candidate theories--relative speed of processing and automaticity of reading--are found to be wanting. It is concluded that recent theories placing the explanatory weight on parallel processing of the irrelevant and the relevant dimensions are likely to be more successful than are earlier theories attempting to locate a single bottleneck in attention. In 1935, J. R. Stroop published his landmark article on attention and interference, an article more influential now than it was then. Why has the Stroop task continued to fascinate us? Perhaps the task is seen as tapping into the primitive operations of cognition, offering clues to the fundamental process of attention. Perhaps the robustness of the phenomenon provides a special challenge to decipher. Together these are powerful attractions

We develop a connectionist approach to processing in quasi-regular domains, as exemplified by English word reading. A consideration of the shortcomings of a previous implementation (Seidenberg & McClelland, 1989, Psych. Rev.) in reading nonwords leads to the development of orthographic and phonological representations that capture better the relevant structure among the written and spoken forms of words. In a number of simulation experiments, networks using the new representations learn to read both regular and exception words, including low-frequency exception words, and yet are still able to read pronounceable nonwords as well as skilled readers. A mathematical analysis of the effects of word frequency and spelling-sound consistency in a related but simpler system serves to clarify the close relationship of these factors in influencing naming latencies. These insights are verified in subsequent simulations, including an attractor network that reproduces the naming latency data directly in its time to settle on a response. Further analyses of the network's ability to reproduce data on impaired reading in surface dyslexia support a view of the reading system that incorporates a graded division-of-labor between semantic and phonological processes. Such a view is consistent with the more general Seidenberg and McClelland framework and has some similarities with---but also important differences from---the standard dual-route account.