Key to reinforcement learning in multi-agent systems is the ability to exploit the fact that agents only directly influence only a small subset of the other agents. Such loose couplings are often modelled using a graphical model: a coordination graph. Finding an (approximately) optimal joint action for a given coordination graph is therefore a central subroutine in cooperative multi-agent reinforcement learning (MARL). Much research in MARL focuses on how to gradually update the parameters of the coordination graph, whilst leaving the solving of the coordination graph up to a known typically exact and generic subroutine. However, exact methods { e.g., Variable Elimination { do not scale well, and generic methods do not exploit the MARL setting of gradually updating a coordination graph and recomputing the joint action to select. In this paper, we examine what happens if we use a heuristic method, i.e., local search, to select joint actions in MARL, and whether we can use outcome of this local search from a previous time-step to speed up and improve local search. We show empirically that by using local search, we can scale up to many agents and complex coordination graphs, and that by reusing joint actions from the previous time-step to initialise local search, we can both improve the quality of the joint actions found and the speed with which these joint actions are found.
LINK
Introduction: To determine if athletes with coordination impairment (CI) can continue playing wheelchair rugby (WR), while an evidence-based classification system, including impairment tests for CI is not yet available. This is a defensible practise if they show similar activity limitations as athletes with other eligible impairment types (OI) within the same sports class. Methods: Standardised activities were measured in 58 elite WR athletes; 14 with CI and 44 with OI. Wheelchair activities consisted of 20-meter sprint, 12-meter sprint with full stop, intermittent sprint (3-meter sprint, stop, 3-meter sprint, stop, 6-meter sprint with full stop), sprint-curve-slalom-curve, turn on the spot 180°, turn on the spot 90°, stop, turn 90°in the same direction, X-test (short circuit with sharp turns) without the ball. Ball activities consisted of maximal throwing distance, precision throwing short (25% of maximum throw) and long (75% of maximal throw) distance and X-test with the ball (pick-up the ball and dribble whilst pushing). Descriptive statistics were used and Spearman’s Rank correlation was assessed for athletes with CI and OI for each outcome measure. Differences between athletes with CI and OI were assessed using a Mann-Whitney U test. Results: Most activities showed a high correlation with the athlete class in both athletes with CI and athletes with OI. Furthermore, outcome measures of athletes with CI overlapped with athletes with OI in the same sports class for all activities. There was a trend for worse performance in athletes with CI in turn on the spot 90°, stop, turn 90°in the same direction, the short distance one handed precision throw (P 0.11)and in the X-test with the ball (P 0.10). Discussion: Despite the current lack of evidence based impairment tests for CI, it is a defensible practise to not exclude athletes with CI from WR with the current classification system. The trends for differences in performance that were found can support athletes and coaches in optimising performance of athletes with CI.
DOCUMENT
Graphs are ubiquitous. Many graphs, including histograms, bar charts, and stacked dotplots, have proven tricky to interpret. Students’ gaze data can indicate students’ interpretation strategies on these graphs. We therefore explore the question: In what way can machine learning quantify differences in students’ gaze data when interpreting two near-identical histograms with graph tasks in between? Our work provides evidence that using machine learning in conjunction with gaze data can provide insight into how students analyze and interpret graphs. This approach also sheds light on the ways in which students may better understand a graph after first being presented with other graph types, including dotplots. We conclude with a model that can accurately differentiate between the first and second time a student solved near-identical histogram tasks.
DOCUMENT
