HTMRL: Biologically Plausible Reinforcement Learning with Hierarchical Temporal Memory
Building Reinforcement Learning algorithms able to adapt to continuously evolving tasks is an open research challenge. One technology known to inherently handle such non-stationary patterns well is Hierarchical Temporal Memory (HTM), a model for the human neocortex. We present HTMRL, the first strictly HTM-based RL algorithm. We show that HTMRL scales to many states and actions, and demonstrate that HTM’s ability for adapting to changing patterns extends to RL.
Prof. Steven Latré, is an associate professor at the University of Antwerp and director at the research centre imec, Belgium. He is leading the IDLab Antwerp research group (85+ members), which is performing applied and fundamental research in the area of communication networks and distributed intelligence. His personal research interests cover machine learning and its application to wireless network optimization.