Michael R. W. Dawson

The How and Why of What Went Where in Apparent Motion: Modeling Solutions to the Motion Correspondence Problem

A model that is capable of maintaining the identities of individuated elements as they move is described. It solves a particular problem of underdetermination, the motion correspondence problem, by simultaneously applying 3 constraints: the nearest neighbor principle, the relative velocity principle, and the element integrity principle. The model generates the same correspondence solutions as does the human visual system for a variety of displays, and many of its properties are consistent with what is known about the physiological mechanisms underlying human motion perception. The model can also be viewed as a proposal of how the identities of attentional tags are maintained by visual cognition, and thus it can be differentiated from a system that serves merely to detect movement.

Online human training of a myoelectric prosthesis controller via actor-critic reinforcement learning

As a contribution toward the goal of adaptable, intelligent artificial limbs, this work introduces a continuous actor-critic reinforcement learning method for optimizing the control of multi-function myoelectric devices. Using a simulated upper-arm robotic prosthesis, we demonstrate how it is possible to derive successful limb controllers from myoelectric data using only a sparse human-delivered training signal, without requiring detailed knowledge about the task domain. This reinforcement-based machine learning framework is well suited for use by both patients and clinical staff, and may be easily adapted to different application domains and the needs of individual amputees. To our knowledge, this is the first my-oelectric control approach that facilitates the online learning of new amputee-specific motions based only on a one-dimensional (scalar) feedback signal provided by the user of the prosthesis.

Modifying the Generalized Delta Rule to Train Networks of Non-monotonic Processors for Pattern Classification

A modification of the generalized delta rule is described that is capable of training multilayer networks of value units, i.e. units defined by a particular non-monotonic activation function, the Gaussian, For simple problems of pattern classification, this rule produces networks with several advantages over standard feedforward networks: they require fewer processing units and can be trained much more quickly. Though superficially similar, there are fundamental differences between the networks trained by this new learning rule and radial basis function networks. These differences suggest that value unit networks may be better suited for learning some pattern classification tasks and for answering general questions related to the organization of neurophysiological systems.

Effects of adapting luminance and stimulus contrast on the temporal and spatial limits of short-range motion.

Perception of short-range motion was studied as a function of adapting luminance and of stimulus contrast in computer simulations and in psychophysical experiments. The stimuli were random dots plotted on an oscilloscope in two sequential frames, separated by an inter-stimulus interval (ISI). Both in the simulations and in the empirical studies, we estimated the maximum spatial displacement (dmax) between corresponding dots in the successive frames, and the maximum ISI (tmax) at which coherent motion was perceived 80% of the time. Conceptually, the research was based on a rectified elaborated Reichardt detector with spatial and temporal filters at its inputs. Extant psychophysical and neurophysiological data were employed to provide estimates of the effects of luminance and contrast on the spatial and temporal characteristics of the input filters. Results showed a strong effect of adapting luminance on both the spatial and temporal limits of motion perception: decrements in luminance produced marked increments in both dmax and tmax. However, neither dmax nor tmax was affected by changes in stimulus contrast. These outcomes are entirely consonant with expectations based on the rectified elaborated Reichardt detector. Alternative explanations of the psychophysical results are discussed.

Autonomous processing in parallel distributed processing networks

Abstract This paper critically examines the claim that parallel distributed processing (PDP) networks are autonomous learning systems. A PDP model of a simple distributed associative memory is considered. It is shown that the ‘generic’ PDP architecture cannot implement the computations required by this memory system without the aid of external control. In other words, the model is not autonomous. Two specific problems are highlighted: (i) simultaneous learning and recall are not permitted to occur as would be required of an autonomous system; (ii) connections between processing units cannot simultaneously represent current and previous network activation as would be required if learning is to occur. Similar problems exist for more sophisticated networks constructed from the generic PDP architecture. We argue that this is because these models are not adequately constrained by the properties of the functional architecture assumed by PDP modelers. It is also argued that without such constraints, PDP research...

Fitting the ex-Gaussian equation to reaction time distributions

Two programs that can be used to determine the probability distributions of reaction times are detailed. The first program takes rank-ordered reaction times as input and outputs a file of quantized data. The second program uses a simplex procedure to estimate the parameters of the ex-Gaussian equation that provides the best description of the quantized data. The advantages of this type of data analysis are also discussed.

The Development of a Myoelectric Training Tool for Above-Elbow Amputees

The objective of above-elbow myoelectric prostheses is to reestablish the functionality of missing limbs and increase the quality of life of amputees. By using electromyography (EMG) electrodes attached to the surface of the skin, amputees are able to control motors in myoelectric prostheses by voluntarily contracting the muscles of their residual limb. This work describes the development of an inexpensive myoelectric training tool (MTT) designed to help upper limb amputees learn how to use myoelectric technology in advance of receiving their actual myoelectric prosthesis. The training tool consists of a physical and simulated robotic arm, signal acquisition hardware, controller software, and a graphical user interface. The MTT improves over earlier training systems by allowing a targeted muscle reinnervation (TMR) patient to control up to two degrees of freedom simultaneously. The training tool has also been designed to function as a research prototype for novel myoelectric controllers. A preliminary experiment was performed in order to evaluate the effectiveness of the MTT as a learning tool and to identify any issues with the system. Five able-bodied participants performed a motor-learning task using the EMG controlled robotic arm with the goal of moving five balls from one box to another as quickly as possible. The results indicate that the subjects improved their skill in myoelectric control over the course of the trials. A usability survey was administered to the subjects after their trials. Results from the survey showed that the shoulder degree of freedom was the most difficult to control.

Polarity matching in the Ternus configuration

Two experiments were conducted to examine the effect of changes in the sign of element contrast on perceptions of the Ternus apparent motion display. In the first experiment, the contrast polarity of all three elements in the display were alternated from the first frame of view to the second. At short durations, this increased perceptions (relative to a control condition) of simultaneity in the display, decreased perceptions of element motion, and did not significantly affect perceptions of group motion. At long durations, this manipulation did not affect performance. In a second experiment, patterns of element polarity were manipulated to favour perceptions of either element motion or of group motion relative to a control condition in which all elements had identical contrast polarity. At a long duration, this manipulation affected perceptions of the configuration; this manipulation did not affect the appearance of the display at a short duration. Together, these results are inconsistent with the predictions of Grossberg and Rudds [Psychological Review, 99, 78-121 (1992)] motion oriented contrast filter. However, they are consistent with a model of motion correspondence processing that includes a polarity matching constraint.

Myoelectric training systems

Myoelectric prostheses aim to restore the functionality of amputated limbs and improve the quality of life of amputees. Myoelectric training systems are used to train and assess the ability of amputees in how to use myoelectric technology in advance of receiving their actual myoelectric prostheses. This article describes the myoelectric training systems that have been developed over the last 20 years in both the literature and commercial industries. The results of this article will identify common features in the training systems and suggest areas for future improvement.

Using perceptrons to explore the reorientation task

The reorientation task is a paradigm that has been used extensively to study the types of information used by humans and animals to navigate in their environment. In this task, subjects are reinforced for going to a particular location in an arena that is typically rectangular in shape. The subject then has to find that location again after being disoriented, and possibly after changes have been made to the arena. This task is used to determine the geometric and featural cues that can be used to reorient the agent in the arena. The purpose of the present paper is to present several simulation results that show that a simple neural network, a perceptron, can be used to generate many of the traditional findings that have been obtained using the reorientation task. These results suggest that reorientation task regularities can be explained without appealing to a geometric module that is a component of spatial processing.