A Spiking Neural Network with Local Learning Rules Derived From Nonnegative Similarity Matching
This work addresses the need for principled spiking neural network algorithms for unsupervised learning, with potential applications in neuromorphic hardware, though it appears incremental as it builds on existing similarity matching frameworks.
The authors tackled the problem of designing biologically plausible spiking neural networks for unsupervised learning by deriving a spiking algorithm from the nonnegative similarity matching cost function, resulting in a network with local learning rules that demonstrated sparse feature extraction and manifold learning in simulations.
The design and analysis of spiking neural network algorithms will be accelerated by the advent of new theoretical approaches. In an attempt at such approach, we provide a principled derivation of a spiking algorithm for unsupervised learning, starting from the nonnegative similarity matching cost function. The resulting network consists of integrate-and-fire units and exhibits local learning rules, making it biologically plausible and also suitable for neuromorphic hardware. We show in simulations that the algorithm can perform sparse feature extraction and manifold learning, two tasks which can be formulated as nonnegative similarity matching problems.