Collaborative Research: Mutual Learning: A Systems Theoretic Investigation

相互学习可以发生在两个人之间，一个人和机器之间，或者两个机器之间。第一类是社会心理学领域的研究人员感兴趣的。人机交互的重要性在许多情况下都能感受到，最近在人类驾驶和完全自动驾驶的车辆之间的交互中也感受到了这一点。拟议的研究涉及机器-机器学习，以调查机器之间的有效合作，但也揭示了这种合作的局限性。具体地说，研究将试图回答这样的问题，即两个主体虽然单独使用将导致预期行为的方案，但是否可能通过相互学习得出错误的结论。尽管相互学习这个术语过去曾被其他研究人员使用，但我们的目标是在数学系统理论的框架内对其进行定量研究。所要研究的问题包括高维空间中的确定性优化，静态/静态环境中的随机强化学习(学习自动机)，使用确定性和随机方案的学习自动机，动态环境中的学习，如马尔可夫决策过程描述的环境中的学习，以及由确定性或随机差分和微分方程描述的动态环境中多智能体的学习/适应。在拟议的项目中获得的相互学习的结果将在国内和国际会议以及每两年一次的耶鲁适应和学习系统研讨会上广泛传播。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Mutual learning can happen between two humans, a human and machine, or between two machines. The first class is of interest to researchers in the field of social psychology. The importance of human machine interactions is being felt in many situations and most recently in the interaction between the human driven and completely autonomous vehicles. The proposed research deals with machine-machine learning to investigate efficient cooperation between machines, but also reveal the limitations of this cooperation. In particular, the research will attempt to answer questions such as whether two agents, although individually using schemes that will result in the desired behavior, may arrive at wrong conclusion using mutual learning.While the term mutual learning has been used by other investigators in the past, our objective is to investigate it in a quantitative sense within the framework of mathematical systems theory. The problems proposed for investigation include deterministic optimization in high dimensional spaces, stochastic reinforcement learning in static/stationary environments (learning automata) using both deterministic and stochastic schemes, learning in dynamic environments such as the ones described by Markov Decision Processes, and learning/adaptation by multiple agents in dynamic environments described by deterministic or stochastic difference and differential equations. The results on mutual learning obtained during the proposed project will be widely disseminated at national and international conferences as well as the bi-annual Yale Workshops on Adaptive and Learning Systems.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.