Efficient Reward-Based Structural Plasticity on a SpiNNaker 2 Prototype

Neuromorphic computing at scale

Neuromorphic computing is a brain-inspired approach to hardware and algorithm design that efficiently realizes artificial neural networks. Neuromorphic designers apply the principles of biointelligence discovered by neuroscientists to design efficient computational systems, often for applications with size, weight and power constraints. With this research field at a critical juncture, it is crucial to chart the course for the development of future large-scale neuromorphic systems. We describe approaches for creating scalable neuromorphic architectures and identify key features. We discuss potential applications that can benefit from scaling and the main challenges that need to be addressed. Furthermore, we examine a comprehensive ecosystem necessary to sustain growth and the new opportunities that lie ahead when scaling neuromorphic systems. Our work distils ideas from several computing sub-fields, providing guidance to researchers and practitioners of neuromorphic computing who aim to push the frontier forward.

Beyond-local neural information processing in neuronal networks

While there is much knowledge about local neuronal circuitry, considerably less is known about how neuronal input is integrated and combined across neuronal networks to encode higher order brain functions. One challenge lies in the large number of complex neural interactions. Neural networks use oscillating activity for information exchange between distributed nodes. To better understand building principles underlying the observation of synchronized oscillatory activity in a large-scale network, we developed a reductionistic neuronal network model. Fundamental building principles are laterally and temporally interconnected virtual nodes (microcircuits), wherein each node was modeled as a local oscillator. By this building principle, the neuronal network model can integrate information in time and space. The simulation gives rise to a wave interference pattern that spreads over all simulated columns in form of a travelling wave. The model design stabilizes states of efficient information processing across all participating neuronal equivalents. Model-specific oscillatory patterns, generated by complex input stimuli, were similar to electrophysiological high-frequency signals that we could confirm in the primate visual cortex during a visual perception task. Important oscillatory model pre-runners, limitations and strength of our reductionistic model are discussed. Our simple scalable model shows unique integration properties and successfully reproduces a variety of biological phenomena such as harmonics, coherence patterns, frequency-speed relationships, and oscillatory activities. We suggest that our scalable model simulates aspects of a basic building principle underlying oscillatory, large-scale integration of information in small and large brains.

Ab Initio Study of Novel Phase-Change Heterostructures

Neuromorphic devices constitute a novel approach to computing that takes inspiration from the brain to unify the processing and storage units. Memories based on phase-change materials (PCMs) are potential candidates for such devices due to their non-volatility and excellent scalability, however their use is hindered by their conductance variability and temporal drift in resistance. Recently, it has been shown that the utilization of phase-change heterostructures consisting of nanolayers of the Sb2Te3 PCM interleaved with a transition-metal dichalcogenide, acting as a confinement material, strongly mitigates these problems. In this work, superlattice heterostructures made of TiTe2 and two prototypical PCMs, respectively GeTe and Ge2Sb2Te5 are considered. By performing ab initio molecular dynamics simulations, it is shown that it is possible to switch the PCMs without destroying the superlattice structure and without diffusion of the atoms of the PCM across the TiTe2 nanolayers. In particular, the model containing Ge2Sb2Te5 shows weak coupling between the two materials during the switching process, which, combined with the high stability of the amorphous state of Ge2Sb2Te5, makes it a very promising candidate for neuromorphic computing applications.

A self-adaptive hardware with resistive switching synapses for experience-based neurocomputing

Neurobiological systems continually interact with the surrounding environment to refine their behaviour toward the best possible reward. Achieving such learning by experience is one of the main challenges of artificial intelligence, but currently it is hindered by the lack of hardware capable of plastic adaptation. Here, we propose a bio-inspired recurrent neural network, mastered by a digital system on chip with resistive-switching synaptic arrays of memory devices, which exploits homeostatic Hebbian learning for improved efficiency. All the results are discussed experimentally and theoretically, proposing a conceptual framework for benchmarking the main outcomes in terms of accuracy and resilience. To test the proposed architecture for reinforcement learning tasks, we study the autonomous exploration of continually evolving environments and verify the results for the Mars rover navigation. We also show that, compared to conventional deep learning techniques, our in-memory hardware has the potential to achieve a significant boost in speed and power-saving.

E-prop on SpiNNaker 2: Exploring online learning in spiking RNNs on neuromorphic hardware

Introduction

In recent years, the application of deep learning models at the edge has gained attention. Typically, artificial neural networks (ANNs) are trained on graphics processing units (GPUs) and optimized for efficient execution on edge devices. Training ANNs directly at the edge is the next step with many applications such as the adaptation of models to specific situations like changes in environmental settings or optimization for individuals, e.g., optimization for speakers for speech processing. Also, local training can preserve privacy. Over the last few years, many algorithms have been developed to reduce memory footprint and computation.

Methods

A specific challenge to train recurrent neural networks (RNNs) for processing sequential data is the need for the Back Propagation Through Time (BPTT) algorithm to store the network state of all time steps. This limitation is resolved by the biologically-inspired E-prop approach for training Spiking Recurrent Neural Networks (SRNNs). We implement the E-prop algorithm on a prototype of the SpiNNaker 2 neuromorphic system. A parallelization strategy is developed to split and train networks on the ARM cores of SpiNNaker 2 to make efficient use of both memory and compute resources. We trained an SRNN from scratch on SpiNNaker 2 in real-time on the Google Speech Command dataset for keyword spotting.

Result

We achieved an accuracy of 91.12% while requiring only 680 KB of memory for training the network with 25 K weights. Compared to other spiking neural networks with equal or better accuracy, our work is significantly more memory-efficient.

Discussion

In addition, we performed a memory and time profiling of the E-prop algorithm. This is used on the one hand to discuss whether E-prop or BPTT is better suited for training a model at the edge and on the other hand to explore architecture modifications to SpiNNaker 2 to speed up online learning. Finally, energy estimations predict that the SRNN can be trained on SpiNNaker2 with 12 times less energy than using a NVIDIA V100 GPU.

Dendritic Computing: Branching Deeper into Machine Learning

In this paper, we discuss the nonlinear computational power provided by dendrites in biological and artificial neurons. We start by briefly presenting biological evidence about the type of dendritic nonlinearities, respective plasticity rules and their effect on biological learning as assessed by computational models. Four major computational implications are identified as improved expressivity, more efficient use of resources, utilizing internal learning signals, and enabling continual learning. We then discuss examples of how dendritic computations have been used to solve real-world classification problems with performance reported on well known data sets used in machine learning. The works are categorized according to the three primary methods of plasticity used-structural plasticity, weight plasticity, or plasticity of synaptic delays. Finally, we show the recent trend of confluence between concepts of deep learning and dendritic computations and highlight some future research directions.

A Brain-Inspired Homeostatic Neuron Based on Phase-Change Memories for Efficient Neuromorphic Computing

One of the main goals of neuromorphic computing is the implementation and design of systems capable of dynamic evolution with respect to their own experience. In biology, synaptic scaling is the homeostatic mechanism which controls the frequency of neural spikes within stable boundaries for improved learning activity. To introduce such control mechanism in a hardware spiking neural network (SNN), we present here a novel artificial neuron based on phase change memory (PCM) devices capable of internal regulation via homeostatic and plastic phenomena. We experimentally show that this mechanism increases the robustness of the system thus optimizing the multi-pattern learning under spike-timing-dependent plasticity (STDP). It also improves the continual learning capability of hybrid supervised-unsupervised convolutional neural networks (CNNs), in terms of both resilience and accuracy. Furthermore, the use of neurons capable of self-regulating their fire responsivity as a function of the PCM internal state enables the design of dynamic networks. In this scenario, we propose to use the PCM-based neurons to design bio-inspired recurrent networks for autonomous decision making in navigation tasks. The agent relies on neuronal spike-frequency adaptation (SFA) to explore the environment via penalties and rewards. Finally, we show that the conductance drift of the PCM devices, contrarily to the applications in neural network accelerators, can improve the overall energy efficiency of neuromorphic computing by implementing bio-plausible active forgetting.

Hardware Acceleration of EEG-Based Emotion Classification Systems: A Comprehensive Survey

Recent years have witnessed a growing interest in EEG-based wearable classifiers of emotions, which could enable the real-time monitoring of patients suffering from neurological disorders such as Amyotrophic Lateral Sclerosis (ALS), Autism Spectrum Disorder (ASD), or Alzheimer's. The hope is that such wearable emotion classifiers would facilitate the patients' social integration and lead to improved healthcare outcomes for them and their loved ones. Yet in spite of their direct relevance to neuro-medicine, the hardware platforms for emotion classification have yet to fill up some important gaps in their various approaches to emotion classification in a healthcare context. In this paper, we present the first hardware-focused critical review of EEG-based wearable classifiers of emotions and survey their implementation perspectives, their algorithmic foundations, and their feature extraction methodologies. We further provide a neuroscience-based analysis of current hardware accelerators of emotion classifiers and use it to map out several research opportunities, including multi-modal hardware platforms, accelerators with tightly-coupled cores operating robustly in the near/supra-threshold region, and pre-processing libraries for universal EEG-based datasets.

Adaptive Extreme Edge Computing for Wearable Devices

Wearable devices are a fast-growing technology with impact on personal healthcare for both society and economy. Due to the widespread of sensors in pervasive and distributed networks, power consumption, processing speed, and system adaptation are vital in future smart wearable devices. The visioning and forecasting of how to bring computation to the edge in smart sensors have already begun, with an aspiration to provide adaptive extreme edge computing. Here, we provide a holistic view of hardware and theoretical solutions toward smart wearable devices that can provide guidance to research in this pervasive computing era. We propose various solutions for biologically plausible models for continual learning in neuromorphic computing technologies for wearable sensors. To envision this concept, we provide a systematic outline in which prospective low power and low latency scenarios of wearable sensors in neuromorphic platforms are expected. We successively describe vital potential landscapes of neuromorphic processors exploiting complementary metal-oxide semiconductors (CMOS) and emerging memory technologies (e.g., memristive devices). Furthermore, we evaluate the requirements for edge computing within wearable devices in terms of footprint, power consumption, latency, and data size. We additionally investigate the challenges beyond neuromorphic computing hardware, algorithms and devices that could impede enhancement of adaptive edge computing in smart wearable devices.

Plasticity and Adaptation in Neuromorphic Biohybrid Systems

Neuromorphic systems take inspiration from the principles of biological information processing to form hardware platforms that enable the large-scale implementation of neural networks. The recent years have seen both advances in the theoretical aspects of spiking neural networks for their use in classification and control tasks and a progress in electrophysiological methods that is pushing the frontiers of intelligent neural interfacing and signal processing technologies. At the forefront of these new technologies, artificial and biological neural networks are tightly coupled, offering a novel "biohybrid" experimental framework for engineers and neurophysiologists. Indeed, biohybrid systems can constitute a new class of neuroprostheses opening important perspectives in the treatment of neurological disorders. Moreover, the use of biologically plausible learning rules allows forming an overall fault-tolerant system of co-developing subsystems. To identify opportunities and challenges in neuromorphic biohybrid systems, we discuss the field from the perspectives of neurobiology, computational neuroscience, and neuromorphic engineering.

Structural plasticity on an accelerated analog neuromorphic hardware system

In computational neuroscience, as well as in machine learning, neuromorphic devices promise an accelerated and scalable alternative to neural network simulations. Their neural connectivity and synaptic capacity depend on their specific design choices, but is always intrinsically limited. Here, we present a strategy to achieve structural plasticity that optimizes resource allocation under these constraints by constantly rewiring the pre- and postsynaptic partners while keeping the neuronal fan-in constant and the connectome sparse. In particular, we implemented this algorithm on the analog neuromorphic system BrainScaleS-2. It was executed on a custom embedded digital processor located on chip, accompanying the mixed-signal substrate of spiking neurons and synapse circuits. We evaluated our implementation in a simple supervised learning scenario, showing its ability to optimize the network topology with respect to the nature of its training data, as well as its overall computational efficiency.

Embodied tactile perception and learning

Various living creatures exhibit embodiment intelligence, which is reflected by a collaborative interaction of the brain, body, and environment. The actual behavior of embodiment intelligence is generated by a continuous and dynamic interaction between a subject and the environment through information perception and physical manipulation. The physical interaction between a robot and the environment is the basis for realizing embodied perception and learning. Tactile information plays a critical role in this physical interaction process. It can be used to ensure safety, stability, and compliance, and can provide unique information that is difficult to capture using other perception modalities. However, due to the limitations of existing sensors and perception and learning methods, the development of robotic tactile research lags significantly behind other sensing modalities, such as vision and hearing, thereby seriously restricting the development of robotic embodiment intelligence. This paper presents the current challenges related to robotic tactile embodiment intelligence and reviews the theory and methods of robotic embodied tactile intelligence. Tactile perception and learning methods for embodiment intelligence can be designed based on the development of new large‐scale tactile array sensing devices, with the aim to make breakthroughs in the neuromorphic computing technology of tactile intelligence.