Many computer vision applications require solving multiple tasks in real-time. A neural network can be trained to solve multiple tasks simultaneously using multi-task learning. This can save computation at inference time as only a single network needs to be evaluated. Unfortunately, this often leads to inferior overall performance as task objectives can compete, which consequently poses the question: which tasks should and should not be learned together in one network when employing multi-task learning? We study task cooperation and competition in several different learning settings and propose a framework for assigning tasks to a few neural networks such that cooperating tasks are computed by the same neural network, while competing tasks are computed by different networks. Our framework offers a time-accuracy trade-off and can produce better accuracy using less inference time than not only a single large multi-task neural network but also many single-task networks.

In cooperative pathfinding problems, non-interfering paths that bring each agent from its current state to its goal state must be planned for multiple agents. We present the first practical, admissible, and complete algorithm for solving problems of this kind. First, we propose a technique called operator decomposition, which can be used to reduce the branching factors of many search algorithms, including algorithms for cooperative pathfinding. We then show how a type of independence common in instances of cooperative pathfinding problems can be exploited. Next, we take the idea of exploiting independent subproblems further by adding improvements that allow the algorithm to recognize many more cases of such independence. Finally, we show empirically that these techniques drastically improve the performance of the standard admissible algorithm for the cooperative pathfinding problem, and that their combination results in a complete algorithm capable of optimally solving relatively large problems in milliseconds.