Abstract
Embedded, continual learning for autonomous and adaptive behavior is a keyapplication of neuromorphic hardware. However, neuromorphic implementations ofembedded learning at large scales that are both flexible and efficient havebeen hindered by a lack of a suitable algorithmic framework. As a result, themost neuromorphic hardware is trained off-line on large clusters of dedicatedprocessors or GPUs and transferred post hoc to the device. We address this byintroducing the neural and synaptic array transceiver (NSAT), a neuromorphiccomputational framework facilitating flexible and efficient embedded learningby matching algorithmic requirements and neural and synaptic dynamics. NSATsupports event-driven supervised, unsupervised and reinforcement learningalgorithms including deep learning. We demonstrate the NSAT in a wide range oftasks, including the simulation of Mihalas-Niebur neuron, dynamic neuralfields, event-driven random back-propagation for event-based deep learning,event-based contrastive divergence for unsupervised learning, and voltage-basedlearning rules for sequence learning. We anticipate that this contribution willestablish the foundation for a new generation of devices enabling adaptivemobile systems, wearable devices, and robots with data-driven autonomy.