The investigation of relational reasoning and strategic processing possesses a rich and comprehensive history documented throughout the literature in educational psychology. Nevertheless, empirical research that examines these two constructs in conjunction remains notably scarce. Further, it is ambiguous how individuals utilize these constructs meaningfully in the context of real-world problem solving scenarios. The objective of the present talk is to review of existing literature concerning relational reasoning and strategic processing. Following this review, a case study is presented wherein a team of researchers studying human perceptions of artificial intelligence are observed across multiple meetings, with their discourse transcribed, coded for instances of both relational reasoning and strategic processing, and analyzed. This study elucidates the ways in which these two constructs influence collective understanding, decision making, and productivity as individuals engage in real-time collaborative problem solving.
As with other NP-Hard problems, the 0-1 knapsack problem can be com- putationally expensive to solve to optimality. Faced with this problem, we develop approaches to obtain estimated optimal solution values that are both very close to the optimal value and efficiently computable. Specifically, while leveraging multiple diverse datasets of various complexities, we find that, in line with previous work, we can obtain good accuracy using a randomized greedy heuristic, followed by a linear regression model which includes the mean, standard deviation, and maximum of the values obtained as predictors. We also include a predictor based on the linear programming relaxation of the integer program. We can improve accuracy slightly while maintaining reasonable computation times by applying random forest and neural network methods. Additionally, by using the information on lower and upper bounds on the optimal cost as predictors, we are able to further improve our predictions by reining in outliers. The results are especially encouraging in the context of more dif- ficult datasets. This research has potential applications including cloud computing allocation problems, multi-armed bandits, vehicle routing problems where packing problems need to be solved, and portfolio optimization, where knapsack problem solution values are needed efficiently and where fast approximate solution values are more valuable than their slow exact counterparts.