Papers & Academic Works

Studies of ongoing, rapid motor behaviors have often focused on the decision-making implicit in the task. Here, we instead study how decision-making integrates with the perceptual and motor systems and propose a framework of limited-capacity, pipelined processing with flexible resources to understand rapid motor behaviors. Results from three experiments show that human performance is consistent with our framework: participants perform objectively worse as task difficulty increases, and, surprisingly, this drop in performance is largest for the most skilled performers. As well, our analysis shows that the worst-performing participants can perform equally well under increased task demands, which is consistent with flexible neural resources being allocated to reduce bottleneck effects and improve overall performance. We conclude that capacity limits lead to information bottlenecks and that processes like attention help reduce the effects that these bottlenecks have on maximal performance.

In control theory, as in other areas of engineering research, there is an inherent tension between the breadth of a technique’s applicability and its mathematical tractability. For the area of discrete-event systems (DES), this manifested itself in a theory of supervisory control that originally provided correct-by-construction guarantees for offline solutions to a restricted kind of deterministic process. Follow-on work extended the reach of these techniques to a number of new settings, notably the development of online control without sacrificing any of the original DES performance guarantees. The ability to enact online control opened the door to applying DES techniques to the adaptive control processes presented by modern technologies: processes with dynamic and time-varying natures, whose characteristics may be understood poorly or not at all. Although many works have built on the seminal work of online control in DES, we believe that these ideas have not reached their full potential due to the difficulty in translating them to adjacent fields. In this survey, we look back at 30 years of research concerning the online control of DES and closely related limited lookahead policies with an eye to making the works accessible to practitioners in the broader control theory and artificial intelligence communities. We conclude with some thoughts on future research directions for the further development and application of online DES control techniques to problems requiring intelligent control in our modern world.

Controlling a discrete-event system commonly entails synthesizing a supervisor to ensure that the plant’s closed-loop behaviour respects a certain specification. In the traditional approach to this problem, if the desired behaviour is not controllable then the specification’s supremal controllable sublanguage is enforced instead. Here we invert the problem and formulate the Minimum Control Base Problem, with the goal of finding the minimum set of controllable events that guarantees controllability for the desired behaviour. We show that the sets of controllable events that maintain controllability for the desired behaviour form a complete lattice with respect to subset inclusion and that there is therefore a minimum capability supervisor for any desired sublanguage of the plant’s behaviour. We apply our techniques to the Small Factory problem and discuss further applications including systems design, dynamic discrete-event systems, and biological systems.

In discrete-event system control, the worst-case time complexity for computing a system’s observer is exponential in the number of that system’s states. This results in practical difficulties since some problems require calculating multiple observers for a changing system, e.g., synthesizing an opacity-enforcing supervisor. Although calculating these observers in an iterative manner allows us to synthesize an opacity-enforcing supervisor and although methods have been proposed to reduce the computational demands, room exists for a practical and intuitive solution. Here we extend the subautomaton relationship to the notion of a subobserver and demonstrate its use in reducing the computations required for iterated observer calculations. We then demonstrate the subobserver relationship’s power by simplifying state-of-the-art synthesis approaches for opacity-enforcing supervisors under realistic assumptions.

Artificial intelligence, control theory and neuroscience have a long history of interplay. An example is human motor control: optimal feedback control describes low-level motor functions and reinforcement learning explains high-level decision-making, but where the two meet is not as well understood. Here I formulate the human motor decision-making problem, describe how discrete-event systems could model it and lay out future research paths to fill in this gap in the literature.

Benchmark data sets are of vital importance in machine learning research, as indicated by the number of repositories that exist to make them publicly available. Although many of these are usable in the stream mining context as well, it is less obvious which data sets can be used to evaluate data stream clustering algorithms. We note that the classic Covertype data set’s size makes it attractive for use in stream mining but unfortunately it is specifically designed for classification. Here we detail the process of transforming the Covertype data set into one amenable for unsupervised learning, which we call the Wilderness Area data set. Our quantitative analysis allows us to conclude that the Wilderness Area data set is more appropriate for unsupervised learning than the original Covertype data set.