Neuromorphic engineering combines the architectural and computational principles of systems neuroscience with semiconductor electronics, with the aim of building efficient and compact devices that mimic the synaptic and neural machinery of the brain. The energy consumptions promised by neuromorphic engineering are extremely low, comparable to those of the nervous system. Until now, however, the neuromorphic approach has been restricted to relatively simple circuits and specialized functions, thereby obfuscating a direct comparison of their energy consumption to that used by conventional von Neumann digital machines solving real-world tasks. Here we show that a recent technology developed by IBM can be leveraged to realize neuromorphic circuits that operate as classifiers of complex real-world stimuli. Specifically, we provide a set of general prescriptions to enable the practical implementation of neural architectures that compete with state-of-the-art classifiers. We also show that the energy consumption of these architectures, realized on the IBM chip, is typically two or more orders of magnitude lower than that of conventional digital machines implementing classifiers with comparable performance. Moreover, the spike-based dynamics display a trade-off between integration time and accuracy, which naturally translates into algorithms that can be flexibly deployed for either fast and approximate classifications, or more accurate classifications at the mere expense of longer running times and higher energy costs. This work finally proves that the neuromorphic approach can be efficiently used in real-world applications and has significant advantages over conventional digital devices when energy consumption is considered.

In visual and auditory scenes, we are able to identify shared features among sensory objects and group them according to their similarity. This grouping is preattentive and fast and is thought of as an elementary form of categorization by which objects sharing similar features are clustered in some abstract perceptual space. It is unclear what neuronal mechanisms underlie this fast categorization. Here we propose a neuromechanistic model of fast feature categorization based on the framework of continuous attractor networks. The mechanism for category formation does not rely on learning and is based on biologically plausible assumptions, for example, the existence of populations of neurons tuned to feature values, feature-specific interactions, and subthreshold-evoked responses upon the presentation of single objects. When the network is presented with a sequence of stimuli characterized by some feature, the network sums the evoked responses and provides a running estimate of the distribution of features in the input stream. If the distribution of features is structured into different components or peaks (i.e., is multimodal), recurrent excitation amplifies the response of activated neurons, and categories are singled out as emerging localized patterns of elevated neuronal activity (bumps), centered at the centroid of each cluster. The emergence of bump states through sequential, subthreshold activation and the dependence on input statistics is a novel application of attractor networks. We show that the extraction and representation of multiple categories are facilitated by the rich attractor structure of the network, which can sustain multiple stable activity patterns for a robust range of connectivity parameters compatible with cortical physiology.