🤖 AI Summary
This work proposes an unsupervised learning approach grounded in bounded rationality, wherein agents develop internally consistent behavioral policies in the absence of explicit reward signals through a “cross-objective” mechanism that couples action prediction with outcome prediction. By integrating random walks and graph structural information, the method encourages agents to autonomously congregate within homogeneous regions during node classification tasks. Experimental results demonstrate that agents exhibit unexpected yet logically coherent behavioral patterns, successfully accomplishing classification while simultaneously revealing the critical role of observer interpretation in the emergence of such behaviors. These findings offer a novel perspective on policy self-organization in unsupervised reinforcement learning.
📝 Abstract
The key point in this note is the self-development of simple behavior strategies, consistently with the bounded rationality hypothesis. Our artificial agents adopt learning techniques, mainly unsupervised, to achieve internal consistency in their behavior, with unexpected results. Those results can be considered mainly as the effects of the observer interpretation. The first technique in use has the name Cross Targets: to train the learning agent we use data crossed between the guesses about the action to be done and the guesses about the following results. An application of the CT "blind" strategy development is then presented: random walkers solve a node classification problem on a graph, after having learnt how to remain in homogeneous regions.