First-Spike-Based Visual Categorization Using Reward-Modulated STDP

• Published in: IEEE transaction on neural networks and learning systems Vol. 29; no. 12; pp. 6178 - 6190
• Main Authors: Mozafari, Milad, Kheradpisheh, Saeed Reza, Masquelier, Timothee, Nowzari-Dalini, Abbas, Ganjtabesh, Mohammad
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.12.2018; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Animals; Cognitive science; Computer Simulation; Datasets; Distance learning; Energy efficiency; Feature extraction; Firing pattern; First-spike-based categorization; Humans; Image edge detection; Information processing; Machine learning; Models, Neurological; Nerve Net; Neural coding; Neural networks; Neuronal Plasticity - physiology; Neurons; Neurons - physiology; Neuroscience; NorB protein; Object recognition; Pattern recognition; Permutations; Reinforcement; Reinforcement learning; reinforcement learning (RL); Reward; reward-modulated spike-timing-dependent plasticity (R-STDP); Selectivity; spiking neural networks (SNNs); temporal coding; Temporal variations; Visual discrimination; visual object recognition; Visual Perception - physiology; reward-modulated spike-timing-dependent plasticity (R-STDP); temporal coding; First-spike-based categorization; visual object recognition; reinforcement learning (RL); spiking neural networks (SNNs)
• ISSN: 2162-237X; 2162-2388; 2162-2388
• DOI: 10.1109/TNNLS.2018.2826721

Reinforcement learning (RL) has recently regained popularity with major achievements such as beating the European game of Go champion. Here, for the first time, we show that RL can be used efficiently to train a spiking neural network (SNN) to perform object recognition in natural images without using an external classifier. We used a feedforward convolutional SNN and a temporal coding scheme where the most strongly activated neurons fire first, while less activated ones fire later, or not at all. In the highest layers, each neuron was assigned to an object category, and it was assumed that the stimulus category was the category of the first neuron to fire. If this assumption was correct, the neuron was rewarded, i.e., spike-timing-dependent plasticity (STDP) was applied, which reinforced the neuron's selectivity. Otherwise, anti-STDP was applied, which encouraged the neuron to learn something else. As demonstrated on various image data sets (Caltech, ETH-80, and NORB), this reward-modulated STDP (R-STDP) approach has extracted particularly discriminative visual features, whereas classic unsupervised STDP extracts any feature that consistently repeats. As a result, R-STDP has outperformed STDP on these data sets. Furthermore, R-STDP is suitable for online learning and can adapt to drastic changes such as label permutations. Finally, it is worth mentioning that both feature extraction and classification were done with spikes, using at most one spike per neuron. Thus, the network is hardware friendly and energy efficient.