Deep neural networks such as Convolutional Networks (ConvNets) and Deep Belief Networks (DBNs) represent the state-of-the-art for many machine learning and computer vision classification problems. To overcome the large computational cost of deep networks, spiking deep networks have recently been proposed, given the specialized hardware now available for spiking neural networks (SNNs). However, this has come at the cost of performance losses due to the conversion from analog neural networks (ANNs) without a notion of time, to sparsely firing, event-driven SNNs. Here we analyze the effects of converting deep ANNs into SNNs with respect to the choice of parameters for spiking neurons such as firing rates and thresholds. We present a set of optimization techniques to minimize performance loss in the conversion process for ConvNets and fully connected deep networks. These techniques yield networks that outperform all previous SNNs on the MNIST database to date, and many networks here are close to maximum performance after only 20 ms of simulated time. The techniques include using rectified linear units (ReLUs) with zero bias during training, and using a new weight normalization method to help regulate firing rates. Our method for converting an ANN into an SNN enables low-latency classification with high accuracies already after the first output spike, and compared with previous SNN approaches it yields improved performance without increased training time. The presented analysis and optimization techniques boost the value of spiking deep networks as an attractive framework for neuromorphic computing platforms aimed at fast and efficient pattern recognition.
22 July 2013 in Proceedings of the National Academy of Sciences
Significance: Neuromorphic emulations express the dynamics of neural systems in analogous electronic circuits, offering a distributed, low-power technology for constructing intelligent systems. However, neuromorphic circuits are inherently imprecise and noisy, and there has been no systematic method for configuring reliable behavioral dynamics on these substrates. We describe such a method, which is able to install simple cognitive behavior on the neuromorphic substrate. Our approach casts light on the general question of how the neuronal circuits of the brain, and also future neuromorphic technologies, could implement cognitive behavior in a principled manner.
Winner-Take-All (WTA) networks are recurrently connected populations of excitatory and inhibitory neurons that represent promising candidate microcircuits for implementing cortical computation. WTAs can perform powerful computations, ranging from signal-restoration to state-dependent processing. However, such networks require fine-tuned connectivity parameters to keep the network dynamics within stable operating regimes. In this article, we show how such stability can emerge autonomously through an interaction of biologically plausible plasticity mechanisms that operate simultaneously on all excitatory and inhibitory synapses of the network. A weight-dependent plasticity rule is derived from the triplet spike-timing dependent plasticity model, and its stabilization properties in the mean-field case are analyzed using contraction theory. Our main result provides simple constraints on the plasticity rule parameters, rather than on the weights themselves, which guarantee stable WTA behavior. The plastic network we present is able to adapt to changing input conditions, and to dynamically adjust its gain, therefore exhibiting self-stabilization mechanisms that are crucial for maintaining stable operation in large networks of interconnected subunits. We show how distributed neural assemblies can adjust their parameters for stable WTA function autonomously while respecting anatomical constraints on neural wiring.
Solving constraint satisfaction problems (CSPs) is a notoriously expensive computational task. Recently, it has been proposed that efficient stochastic solvers can be obtained through appropriately configured spiking neural networks performing Markov Chain Monte Carlo (MCMC) sampling. The possibility to run such models on massively parallel, low-power neuromorphic hardware holds great promise; however, previously proposed networks a re based on probabilistically spiking neurons, and thus rely on random number generators or external noise sources to achieve the necessary stochasticity, leading to significant overhead in the implementation. Here we show how stochasticity can be achieved by implementing deterministic models of integrate and fire neurons using subthreshold analog circuits that are affected by thermal noise. We present an efficient implementation of spike-based CSP solvers using a reconfigurable neural network VLSI device, and the device's intrinsic noise as a source of randomness. To illustrate the overall concept, we implement a generic Sudoku solver based on our approach and demonstrate its operation. We establish a link between the neuron parameters and the system dynamics, allowing for a simple temperature control mechanism.
1 January 2012 in Lecture Notes in Computer Science