Remote Learning with HRI-based Tutoring Systems
Yi Zhao, Rohit Kumar, Abdul Fattah Rakib · 2024
This paper presents memristive nano-devices that naturally emulate these diverse synaptic functions. These memristors operate in a non-filamentary, low-conductance mode, ensuring stable and energy-efficient performance. Designed as adaptable hardware synapses for bio-inspired deep neural networks (DNNs), they incorporate both long- and short-term synaptic responses, enabling meta-learning—the ability to refine learning strategies over time. When applied to a demanding reinforcement learning task in a dynamic environment, the bio-inspired DNN with memristive synapses reduced energy consumption by approximately two orders of magnitude compared to traditional GPU-based implementations. This demonstrates that memristive devices, by more closely mimicking synaptic behavior, can greatly expand the potential of neuromorphic computing while enhancing the efficiency and performance of artificial intelligence systems.