Recurrently connected networks of spiking neurons underlie the astounding information processing capabilities of the brain. Yet in spite of extensive research, how they can learn through synaptic plasticity to carry out complex network computations remains unclear. We argue that two pieces of this puzzle were provided by experimental data from neuroscience. A mathematical result tells us how these pieces need to be combined to enable biologically plausible online network learning through gradient descent, in particular deep reinforcement learning. This learning method-called e-prop-approaches the performance of backpropagation through time (BPTT), the best-known method for training recurrent neural networks in machine learning. In addition, it suggests a method for powerful on-chip learning in energy-efficient spike-based hardware for artificial intelligence.

Recurrently connected networks of spiking neurons underlie the astounding information processing capabilities of the brain. But in spite of extensive research, it has remained open how learning through synaptic plasticity could be organized in such networks. We argue that two pieces of this puzzle were provided by experimental data from neuroscience. A new mathematical insight tells us how they need to be combined to enable network learning through gradient descent. The resulting learning method -called e-prop -approaches the performance of BPTT (backpropagation through time), the best known method for training recurrent neural networks in machine learning. But in contrast to BPTT, e-prop is biologically plausible. In addition, it elucidates how brain-inspired new computer chips -that are drastically more energy efficient -can be enabled to learn.

For solving tasks such as recognizing a song, answering a question, or inverting a sequence of symbols, cortical microcircuits need to integrate and manipulate information that was dispersed over time during the preceding seconds. Creating biologically realistic models for the underlying computations, especially with spiking neurons and for behaviorally relevant integration time spans, is notoriously difficult. We examine the role of spike frequency adaptation in such computations and find that it has a surprisingly large impact. The inclusion of this well-known property of a substantial fraction of neurons in the neocortex – especially in higher areas of the human neocortex – moves the performance of spiking neural network models for computations on network inputs that are temporally dispersed from a fairly low level up to the performance level of the human brain.

Recurrent networks of spiking neurons (RSNNs) underlie the astounding computing and learning capabilities of the brain. But computing and learning capabilities of RSNN models have remained poor, at least in comparison with artificial neural networks (ANNs). We address two possible reasons for that. One is that RSNNs in the brain are not randomly connected or designed according to simple rules, and they do not start learning as a tabula rasa network. Rather, RSNNs in the brain were optimized for their tasks through evolution, development, and prior experience. Details of these optimization processes are largely unknown. But their functional contribution can be approximated through powerful optimization methods, such as backpropagation through time (BPTT). A second major mismatch between RSNNs in the brain and models is that the latter only show a small fraction of the dynamics of neurons and synapses in the brain. We include neurons in our RSNN model that reproduce one prominent dynamical process of biological neurons that takes place at the behaviourally relevant time scale of seconds: neuronal adaptation. We denote these networks as LSNNs because of their Long short-term memory. The inclusion of adapting neurons drastically increases the computing and learning capability of RSNNs if they are trained and configured by deep learning (BPTT combined with a rewiring algorithm that optimizes the network architecture). In fact, the computational performance of these RSNNs approaches for the first time that of LSTM networks. In addition RSNNs with adapting neurons can acquire abstract knowledge from prior learning in a Learning-to-Learn (L2L) scheme, and transfer that knowledge in order to learn new but related tasks from very few examples. We demonstrate this for supervised learning and reinforcement learning.32nd Conference on Neural Information Processing Systems (NeurIPS 2018),

Short-term visual prediction is important both in biology and robotics. It allows us to anticipate upcoming states of the environment and therefore plan more efficiently. In theoretical neuroscience, liquid state machines have been proposed as a biologically inspired method to perform asynchronous prediction without a model. However, they have so far only been demonstrated in simulation or small scale pre-processed camera images. In this paper, we use a liquid state machine to predict over the whole  [Formula: see text]  event stream provided by a real dynamic vision sensor (DVS, or silicon retina). Thanks to the event-based nature of the DVS, the liquid is constantly fed with data when an object is in motion, fully embracing the asynchronicity of spiking neural networks. We propose a smooth continuous representation of the event stream for the short-term visual prediction task. Moreover, compared to previous works (2002 Neural Comput. 2525 282-93 and Burgsteiner H et al 2007 Appl. Intell. 26 99-109), we scale the input dimensionality that the liquid operates on by two order of magnitudes. We also expose the current limits of our method by running experiments in a challenging environment where multiple objects are in motion. This paper is a step towards integrating biologically inspired algorithms derived in theoretical neuroscience to real world robotic setups. We believe that liquid state machines could complement current prediction algorithms used in robotics, especially when dealing with asynchronous sensors.