Emre Neftci

Jason Yik, Korneel Van den Berghe, Douwe den Blanken et al. · eth-zurich

Neuromorphic computing shows promise for advancing computing efficiency and capabilities of AI applications using brain-inspired principles. However, the neuromorphic research field currently lacks standardized benchmarks, making it difficult to accurately measure technological advancements, compare performance with conventional methods, and identify promising future research directions. Prior neuromorphic computing benchmark efforts have not seen widespread adoption due to a lack of inclusive, actionable, and iterative benchmark design and guidelines. To address these shortcomings, we present NeuroBench: a benchmark framework for neuromorphic computing algorithms and systems. NeuroBench is a collaboratively-designed effort from an open community of researchers across industry and academia, aiming to provide a representative structure for standardizing the evaluation of neuromorphic approaches. The NeuroBench framework introduces a common set of tools and systematic methodology for inclusive benchmark measurement, delivering an objective reference framework for quantifying neuromorphic approaches in both hardware-independent (algorithm track) and hardware-dependent (system track) settings. In this article, we outline tasks and guidelines for benchmarks across multiple application domains, and present initial performance baselines across neuromorphic and conventional approaches for both benchmark tracks. NeuroBench is intended to continually expand its benchmarks and features to foster and track the progress made by the research community.

Nick Alonso, Beren Millidge, Jeff Krichmar et al.

Backpropagation (BP) is the most successful and widely used algorithm in deep learning. However, the computations required by BP are challenging to reconcile with known neurobiology. This difficulty has stimulated interest in more biologically plausible alternatives to BP. One such algorithm is the inference learning algorithm (IL). IL has close connections to neurobiological models of cortical function and has achieved equal performance to BP on supervised learning and auto-associative tasks. In contrast to BP, however, the mathematical foundations of IL are not well-understood. Here, we develop a novel theoretical framework for IL. Our main result is that IL closely approximates an optimization method known as implicit stochastic gradient descent (implicit SGD), which is distinct from the explicit SGD implemented by BP. Our results further show how the standard implementation of IL can be altered to better approximate implicit SGD. Our novel implementation considerably improves the stability of IL across learning rates, which is consistent with our theory, as a key property of implicit SGD is its stability. We provide extensive simulation results that further support our theoretical interpretations and also demonstrate IL achieves quicker convergence when trained with small mini-batches while matching the performance of BP for large mini-batches.

Dhireesha Kudithipudi, Anurag Daram, Abdullah M. Zyarah et al.

Lifelong learning - an agent's ability to learn throughout its lifetime - is a hallmark of biological learning systems and a central challenge for artificial intelligence (AI). The development of lifelong learning algorithms could lead to a range of novel AI applications, but this will also require the development of appropriate hardware accelerators, particularly if the models are to be deployed on edge platforms, which have strict size, weight, and power constraints. Here, we explore the design of lifelong learning AI accelerators that are intended for deployment in untethered environments. We identify key desirable capabilities for lifelong learning accelerators and highlight metrics to evaluate such accelerators. We then discuss current edge AI accelerators and explore the future design of lifelong learning accelerators, considering the role that different emerging technologies could play.

Jan Finkbeiner, Thomas Gmeinder, Mark Pupilli et al.

Current AI training infrastructure is dominated by single instruction multiple data (SIMD) and systolic array architectures, such as Graphics Processing Units (GPUs) and Tensor Processing Units (TPUs), that excel at accelerating parallel workloads and dense vector matrix multiplications. Potentially more efficient neural network models utilizing sparsity and recurrence cannot leverage the full power of SIMD processor and are thus at a severe disadvantage compared to today's prominent parallel architectures like Transformers and CNNs, thereby hindering the path towards more sustainable AI. To overcome this limitation, we explore sparse and recurrent model training on a massively parallel multiple instruction multiple data (MIMD) architecture with distributed local memory. We implement a training routine based on backpropagation through time (BPTT) for the brain-inspired class of Spiking Neural Networks (SNNs) that feature binary sparse activations. We observe a massive advantage in using sparse activation tensors with a MIMD processor, the Intelligence Processing Unit (IPU) compared to GPUs. On training workloads, our results demonstrate 5-10x throughput gains compared to A100 GPUs and up to 38x gains for higher levels of activation sparsity, without a significant slowdown in training convergence or reduction in final model performance. Furthermore, our results show highly promising trends for both single and multi IPU configurations as we scale up to larger model sizes. Our work paves the way towards more efficient, non-standard models via AI training hardware beyond GPUs, and competitive large scale SNN models.

Maxime Fabre, Lyubov Dudchenko, Younes Bouhadjar et al.

Multi-state spiking neurons combine sparse binary activations with rich second-order nonlinear recurrent dynamics, making them a promising alternative to standard deep learning models. However, gradient propagation through these dynamics often leads to instabilities that hinder scalability and performance. Inspired by the stable training and strong performance of state space models (SSMs) on long sequences, we introduce two SSM-inspired Leaky Integrate-and-Fire (SiLIF) neuron models. The first extends a two-state neuron with a learnable discretization timestep and logarithmic reparametrization, while the second additionally incorporates the initialization scheme and structure of complex-state SSMs, enabling oscillatory regimes. Our two SiLIF models achieve new state-of-the-art performance among spiking neuron models on both event-based and raw-audio speech recognition datasets. We further demonstrate a favorable performance-efficiency trade-off compared to SSMs, even surpassing them while using half the computational cost through the use of synaptic delays. Our code is available at https://github.com/Maxtimer97/SSM-inspired-LIF.

Jinwei Xing, Takashi Nagata, Xinyun Zou et al.

Although deep Reinforcement Learning (RL) has proven successful in a wide range of tasks, one challenge it faces is interpretability when applied to real-world problems. Saliency maps are frequently used to provide interpretability for deep neural networks. However, in the RL domain, existing saliency map approaches are either computationally expensive and thus cannot satisfy the real-time requirement of real-world scenarios or cannot produce interpretable saliency maps for RL policies. In this work, we propose an approach of Distillation with selective Input Gradient Regularization (DIGR) which uses policy distillation and input gradient regularization to produce new policies that achieve both high interpretability and computation efficiency in generating saliency maps. Our approach is also found to improve the robustness of RL policies to multiple adversarial attacks. We conduct experiments on three tasks, MiniGrid (Fetch Object), Atari (Breakout) and CARLA Autonomous Driving, to demonstrate the importance and effectiveness of our approach.

Nathan Leroux, Paul-Philipp Manea, Chirag Sudarshan et al.

Transformer networks, driven by self-attention, are central to Large Language Models. In generative Transformers, self-attention uses cache memory to store token projections, avoiding recomputation at each time step. However, GPU-stored projections must be loaded into SRAM for each new generation step, causing latency and energy bottlenecks. We present a custom self-attention in-memory computing architecture based on emerging charge-based memories called gain cells, which can be efficiently written to store new tokens during sequence generation and enable parallel analog dot-product computation required for self-attention. However, the analog gain cell circuits introduce non-idealities and constraints preventing the direct mapping of pre-trained models. To circumvent this problem, we design an initialization algorithm achieving text processing performance comparable to GPT-2 without training from scratch. Our architecture respectively reduces attention latency and energy consumption by up to two and five orders of magnitude compared to GPUs, marking a significant step toward ultra-fast, low-power generative Transformers.

Nathan Leroux, Jan Finkbeiner, Emre Neftci

Transformers are state-of-the-art networks for most sequence processing tasks. However, the self-attention mechanism often used in Transformers requires large time windows for each computation step and thus makes them less suitable for online signal processing compared to Recurrent Neural Networks (RNNs). In this paper, instead of the self-attention mechanism, we use a sliding window attention mechanism. We show that this mechanism is more efficient for continuous signals with finite-range dependencies between input and target, and that we can use it to process sequences element-by-element, this making it compatible with online processing. We test our model on a finger position regression dataset (NinaproDB8) with Surface Electromyographic (sEMG) signals measured on the forearm skin to estimate muscle activities. Our approach sets the new state-of-the-art in terms of accuracy on this dataset while requiring only very short time windows of 3.5 ms at each inference step. Moreover, we increase the sparsity of the network using Leaky-Integrate and Fire (LIF) units, a bio-inspired neuron model that activates sparsely in time solely when crossing a threshold. We thus reduce the number of synaptic operations up to a factor of $\times5.3$ without loss of accuracy. Our results hold great promises for accurate and fast online processing of sEMG signals for smooth prosthetic hand control and is a step towards Transformers and Spiking Neural Networks (SNNs) co-integration for energy efficient temporal signal processing.

Kenneth Stewart, Michael Neumeier, Sumit Bam Shrestha et al.

Achieving personalized intelligence at the edge with real-time learning capabilities holds enormous promise in enhancing our daily experiences and helping decision making, planning, and sensing. However, efficient and reliable edge learning remains difficult with current technology due to the lack of personalized data, insufficient hardware capabilities, and inherent challenges posed by online learning. Over time and across multiple developmental stages, the brain has evolved to efficiently incorporate new knowledge by gradually building on previous knowledge. In this work, we emulate the multiple stages of learning with digital neuromorphic technology that simulates the neural and synaptic processes of the brain using two stages of learning. First, a meta-training stage trains the hyperparameters of synaptic plasticity for one-shot learning using a differentiable simulation of the neuromorphic hardware. This meta-training process refines a hardware local three-factor synaptic plasticity rule and its associated hyperparameters to align with the trained task domain. In a subsequent deployment stage, these optimized hyperparameters enable fast, data-efficient, and accurate learning of new classes. We demonstrate our approach using event-driven vision sensor data and the Intel Loihi neuromorphic processor with its plasticity dynamics, achieving real-time one-shot learning of new classes that is vastly improved over transfer learning. Our methodology can be deployed with arbitrary plasticity models and can be applied to situations demanding quick learning and adaptation at the edge, such as navigating unfamiliar environments or learning unexpected categories of data through user engagement.

Barna Zajzon, Younes Bouhadjar, Maxime Fabre et al.

Sequential structure is a key feature of multiple domains of natural cognition and behavior, such as language, movement and decision-making. Likewise, it is also a central property of tasks to which we would like to apply artificial intelligence. It is therefore of great importance to develop frameworks that allow us to evaluate sequence learning and processing in a domain agnostic fashion, whilst simultaneously providing a link to formal theories of computation and computability. To address this need, we introduce two complementary software tools: SymSeq, designed to rigorously generate and analyze structured symbolic sequences, and SeqBench, a comprehensive benchmark suite of rule-based sequence processing tasks to evaluate the performance of artificial learning systems in cognitively relevant domains. In combination, SymSeqBench offers versatility in investigating sequential structure across diverse knowledge domains, including experimental psycholinguistics, cognitive psychology, behavioral analysis, neuromorphic computing and artificial intelligence. Due to its basis in Formal Language Theory (FLT), SymSeqBench provides researchers in multiple domains with a convenient and practical way to apply the concepts of FLT to conceptualize and standardize their experiments, thus advancing our understanding of cognition and behavior through shared computational frameworks and formalisms. The tool is modular, openly available and accessible to the research community.

Jamie Lohoff, Jan Finkbeiner, Emre Neftci

Spiking Neural Networks (SNNs) simulators are essential tools to prototype biologically inspired models and neuromorphic hardware architectures and predict their performance. For such a tool, ease of use and flexibility are critical, but so is simulation speed especially given the complexity inherent to simulating SNN. Here, we present SNNAX, a JAX-based framework for simulating and training such models with PyTorch-like intuitiveness and JAX-like execution speed. SNNAX models are easily extended and customized to fit the desired model specifications and target neuromorphic hardware. Additionally, SNNAX offers key features for optimizing the training and deployment of SNNs such as flexible automatic differentiation and just-in-time compilation. We evaluate and compare SNNAX to other commonly used machine learning (ML) frameworks used for programming SNNs. We provide key performance metrics, best practices, documented examples for simulating SNNs in SNNAX, and implement several benchmarks used in the literature.

Ni Wang, Zihan You, Emre Neftci et al.

Autonomous agents such as cars, robots and drones need to precisely localize themselves in diverse environments, including in GPS-denied indoor environments. One approach for precise localization is visual place recognition (VPR), which estimates the place of an image based on previously seen places. State-of-the-art VPR models require high amounts of memory, making them unwieldy for mobile deployment, while more compact models lack robustness and generalization capabilities. This work overcomes these limitations for robotics using a combination of event-based vision sensors and an event-based novel guided variational autoencoder (VAE). The encoder part of our model is based on a spiking neural network model which is compatible with power-efficient low latency neuromorphic hardware. The VAE successfully disentangles the visual features of 16 distinct places in our new indoor VPR dataset with a classification performance comparable to other state-of-the-art approaches while, showing robust performance also under various illumination conditions. When tested with novel visual inputs from unknown scenes, our model can distinguish between these places, which demonstrates a high generalization capability by learning the essential features of location. Our compact and robust guided VAE with generalization capabilities poses a promising model for visual place recognition that can significantly enhance mobile robot navigation in known and unknown indoor environments.

Madison Cotteret, Hugh Greatorex, Alpha Renner et al. · eth-zurich

Programming recurrent spiking neural networks (RSNNs) to robustly perform multi-timescale computation remains a difficult challenge. To address this, we describe a single-shot weight learning scheme to embed robust multi-timescale dynamics into attractor-based RSNNs, by exploiting the properties of high-dimensional distributed representations. We embed finite state machines into the RSNN dynamics by superimposing a symmetric autoassociative weight matrix and asymmetric transition terms, which are each formed by the vector binding of an input and heteroassociative outer-products between states. Our approach is validated through simulations with highly nonideal weights; an experimental closed-loop memristive hardware setup; and on Loihi 2, where it scales seamlessly to large state machines. This work introduces a scalable approach to embed robust symbolic computation through recurrent dynamics into neuromorphic hardware, without requiring parameter fine-tuning or significant platform-specific optimisation. Moreover, it demonstrates that distributed symbolic representations serve as a highly capable representation-invariant language for cognitive algorithms in neuromorphic hardware.

Sanja Karilanova, Maxime Fabre, Emre Neftci et al.

Spiking Neural Networks (SNNs) are biologically-inspired deep neural networks that efficiently extract temporal information while offering promising gains in terms of energy efficiency and latency when deployed on neuromorphic devices. However, SNN model parameters are sensitive to temporal resolution, leading to significant performance drops when the temporal resolution of target data at the edge is not the same with that of the pre-deployment source data used for training, especially when fine-tuning is not possible at the edge. To address this challenge, we propose three novel domain adaptation methods for adapting neuron parameters to account for the change in time resolution without re-training on target time-resolution. The proposed methods are based on a mapping between neuron dynamics in SNNs and State Space Models (SSMs); and are applicable to general neuron models. We evaluate the proposed methods under spatio-temporal data tasks, namely the audio keyword spotting datasets SHD and MSWC as well as the image classification NMINST dataset. Our methods provide an alternative to - and in majority of the cases significantly outperform - the existing reference method that simply scales the time constant. Moreover, our results show that high accuracy on high temporal resolution data can be obtained by time efficient training on lower temporal resolution data and model adaptation.

Jamie Lohoff, Anil Kaya, Florian Assmuth et al.

Online synaptic plasticity rules derived from gradient descent achieve high accuracy on a wide range of practical tasks. However, their software implementation often requires tediously hand-derived gradients or using gradient backpropagation which sacrifices the online capability of the rules. In this work, we present a custom automatic differentiation (AD) pipeline for sparse and online implementation of gradient-based synaptic plasticity rules that generalizes to arbitrary neuron models. Our work combines the programming ease of backpropagation-type methods for forward AD while being memory-efficient. To achieve this, we exploit the advantageous compute and memory scaling of online synaptic plasticity by providing an inherently sparse implementation of AD where expensive tensor contractions are replaced with simple element-wise multiplications if the tensors are diagonal. Gradient-based synaptic plasticity rules such as eligibility propagation (e-prop) have exactly this property and thus profit immensely from this feature. We demonstrate the alignment of our gradients with respect to gradient backpropagation on an synthetic task where e-prop gradients are exact, as well as audio speech classification benchmarks. We demonstrate how memory utilization scales with network size without dependence on the sequence length, as expected from forward AD methods.

Viet Anh Khoa Tran, Emre Neftci, Willem A. M. Wybo

Biological brains learn continually from a stream of unlabeled data, while integrating specialized information from sparsely labeled examples without compromising their ability to generalize. Meanwhile, machine learning methods are susceptible to catastrophic forgetting in this natural learning setting, as supervised specialist fine-tuning degrades performance on the original task. We introduce task-modulated contrastive learning (TMCL), which takes inspiration from the biophysical machinery in the neocortex, using predictive coding principles to integrate top-down information continually and without supervision. We follow the idea that these principles build a view-invariant representation space, and that this can be implemented using a contrastive loss. Then, whenever labeled samples of a new class occur, new affine modulations are learned that improve separation of the new class from all others, without affecting feedforward weights. By co-opting the view-invariance learning mechanism, we then train feedforward weights to match the unmodulated representation of a data sample to its modulated counterparts. This introduces modulation invariance into the representation space, and, by also using past modulations, stabilizes it. Our experiments show improvements in both class-incremental and transfer learning over state-of-the-art unsupervised approaches, as well as over comparable supervised approaches, using as few as 1% of available labels. Taken together, our work suggests that top-down modulations play a crucial role in balancing stability and plasticity.

Soikat Hasan Ahmed, Jan Finkbeiner, Emre Neftci

Event cameras offer high temporal resolution and dynamic range with minimal motion blur, making them promising for robust object detection. While Spiking Neural Networks (SNNs) on neuromorphic hardware are often considered for energy-efficient and low latency event-based data processing, they often fall short of Artificial Neural Networks (ANNs) in accuracy and flexibility. Here, we introduce Attention-based Hybrid SNN-ANN backbones for event-based object detection to leverage the strengths of both SNN and ANN architectures. A novel Attention-based SNN-ANN bridge module captures sparse spatial and temporal relations from the SNN layer and converts them into dense feature maps for the ANN part of the backbone. Additionally, we present a variant that integrates DWConvL-STMs to the ANN blocks to capture slower dynamics. This multi-timescale network combines fast SNN processing for short timesteps with long-term dense RNN processing, effectively capturing both fast and slow dynamics. Experimental results demonstrate that our proposed method surpasses SNN-based approaches by significant margins, with results comparable to existing ANN and RNN-based methods. Unlike ANN-only networks, the hybrid setup allows us to implement the SNN blocks on digital neuromorphic hardware to investigate the feasibility of our approach. Extensive ablation studies and implementation on neuromorphic hardware confirm the effectiveness of our proposed modules and architectural choices. Our hybrid SNN-ANN architectures pave the way for ANN-like performance at a drastically reduced parameter, latency, and power budget.

Djohan Bonnet, Jamie Lohoff, Jan Finkbeiner et al.

In-Context Learning (ICL) in transformers acts as an online associative memory and is believed to underpin their high performance on complex sequence processing tasks. However, in gated linear attention models, this memory has a fixed capacity and is prone to interference, especially for long sequences. We propose Palimpsa, a self-attention model that views ICL as a continual learning problem that must address a stability-plasticity dilemma. Palimpsa uses Bayesian metaplasticity, where the plasticity of each attention state is tied to an importance state grounded by a prior distribution that captures accumulated knowledge. We demonstrate that various gated linear attention models emerge as specific architecture choices and posterior approximations, and that Mamba2 is a special case of Palimpsa where forgetting dominates. This theoretical link enables the transformation of any non-metaplastic model into a metaplastic one, significantly expanding its memory capacity. Our experiments show that Palimpsa consistently outperforms baselines on the Multi-Query Associative Recall (MQAR) benchmark and on Commonsense Reasoning tasks.

Sebastian Siegel, Ming-Jay Yang, Younes Bouhadjar et al.

Structured State Space models (SSM) have recently emerged as a new class of deep learning models, particularly well-suited for processing long sequences. Their constant memory footprint, in contrast to the linearly scaling memory demands of Transformers, makes them attractive candidates for deployment on resource-constrained edge-computing devices. While recent works have explored the effect of quantization-aware training (QAT) on SSMs, they typically do not address its implications for specialized edge hardware, for example, analog in-memory computing (AIMC) chips. In this work, we demonstrate that QAT can significantly reduce the complexity of SSMs by up to two orders of magnitude across various performance metrics. We analyze the relation between model size and numerical precision, and show that QAT enhances robustness to analog noise and enables structural pruning. Finally, we integrate these techniques to deploy SSMs on a memristive analog in-memory computing substrate and highlight the resulting benefits in terms of computational efficiency.

Sven Krausse, Emre Neftci, Friedrich T. Sommer et al. · eth-zurich

The entorhinal-hippocampal formation is the mammalian brain's navigation system, encoding both physical and abstract spaces via grid cells. This system is well-studied in neuroscience, and its efficiency and versatility make it attractive for applications in robotics and machine learning. While continuous attractor networks (CANs) successfully model entorhinal grid cells for encoding physical space, integrating both continuous spatial and abstract spatial computations into a unified framework remains challenging. Here, we attempt to bridge this gap by proposing a mechanistic model for versatile information processing in the entorhinal-hippocampal formation inspired by CANs and Vector Symbolic Architectures (VSAs), a neuro-symbolic computing framework. The novel grid-cell VSA (GC-VSA) model employs a spatially structured encoding scheme with 3D neuronal modules mimicking the discrete scales and orientations of grid cell modules, reproducing their characteristic hexagonal receptive fields. In experiments, the model demonstrates versatility in spatial and abstract tasks: (1) accurate path integration for tracking locations, (2) spatio-temporal representation for querying object locations and temporal relations, and (3) symbolic reasoning using family trees as a structured test case for hierarchical relationships.

Wadjih Bencheikh, Jan Finkbeiner, Emre Neftci

Recurrent neural networks (RNNs) are valued for their computational efficiency and reduced memory requirements on tasks involving long sequence lengths but require high memory-processor bandwidth to train. Checkpointing techniques can reduce the memory requirements by only storing a subset of intermediate states, the checkpoints, but are still rarely used due to the computational overhead of the additional recomputation phase. This work addresses these challenges by introducing memory-efficient gradient checkpointing strategies tailored for the general class of sparse RNNs and Spiking Neural Networks (SNNs). SNNs are energy efficient alternatives to RNNs thanks to their local, event-driven operation and potential neuromorphic implementation. We use the Intelligence Processing Unit (IPU) as an exemplary platform for architectures with distributed local memory. We exploit its suitability for sparse and irregular workloads to scale SNN training on long sequence lengths. We find that Double Checkpointing emerges as the most effective method, optimizing the use of local memory resources while minimizing recomputation overhead. This approach reduces dependency on slower large-scale memory access, enabling training on sequences over 10 times longer or 4 times larger networks than previously feasible, with only marginal time overhead. The presented techniques demonstrate significant potential to enhance scalability and efficiency in training sparse and recurrent networks across diverse hardware platforms, and highlights the benefits of sparse activations for scalable recurrent neural network training.

Jamie Lohoff, Emre Neftci

Computing Jacobians with automatic differentiation is ubiquitous in many scientific domains such as machine learning, computational fluid dynamics, robotics and finance. Even small savings in the number of computations or memory usage in Jacobian computations can already incur massive savings in energy consumption and runtime. While there exist many methods that allow for such savings, they generally trade computational efficiency for approximations of the exact Jacobian. In this paper, we present a novel method to optimize the number of necessary multiplications for Jacobian computation by leveraging deep reinforcement learning (RL) and a concept called cross-country elimination while still computing the exact Jacobian. Cross-country elimination is a framework for automatic differentiation that phrases Jacobian accumulation as ordered elimination of all vertices on the computational graph where every elimination incurs a certain computational cost. We formulate the search for the optimal elimination order that minimizes the number of necessary multiplications as a single player game which is played by an RL agent. We demonstrate that this method achieves up to 33% improvements over state-of-the-art methods on several relevant tasks taken from diverse domains. Furthermore, we show that these theoretical gains translate into actual runtime improvements by providing a cross-country elimination interpreter in JAX that can efficiently execute the obtained elimination orders.

Kenneth Stewart, Emre Neftci

Adaptive "life-long" learning at the edge and during online task performance is an aspirational goal of AI research. Neuromorphic hardware implementing Spiking Neural Networks (SNNs) are particularly attractive in this regard, as their real-time, event-based, local computing paradigm makes them suitable for edge implementations and fast learning. However, the long and iterative learning that characterizes state-of-the-art SNN training is incompatible with the physical nature and real-time operation of neuromorphic hardware. Bi-level learning, such as meta-learning is increasingly used in deep learning to overcome these limitations. In this work, we demonstrate gradient-based meta-learning in SNNs using the surrogate gradient method that approximates the spiking threshold function for gradient estimations. Because surrogate gradients can be made twice differentiable, well-established, and effective second-order gradient meta-learning methods such as Model Agnostic Meta Learning (MAML) can be used. We show that SNNs meta-trained using MAML match or exceed the performance of conventional ANNs meta-trained with MAML on event-based meta-datasets. Furthermore, we demonstrate the specific advantages that accrue from meta-learning: fast learning without the requirement of high precision weights or gradients. Our results emphasize how meta-learning techniques can become instrumental for deploying neuromorphic learning technologies on real-world problems.

Jason K. Eshraghian, Max Ward, Emre Neftci et al.

The brain is the perfect place to look for inspiration to develop more efficient neural networks. The inner workings of our synapses and neurons provide a glimpse at what the future of deep learning might look like. This paper serves as a tutorial and perspective showing how to apply the lessons learnt from several decades of research in deep learning, gradient descent, backpropagation and neuroscience to biologically plausible spiking neural neural networks. We also explore the delicate interplay between encoding data as spikes and the learning process; the challenges and solutions of applying gradient-based learning to spiking neural networks (SNNs); the subtle link between temporal backpropagation and spike timing dependent plasticity, and how deep learning might move towards biologically plausible online learning. Some ideas are well accepted and commonly used amongst the neuromorphic engineering community, while others are presented or justified for the first time here. The fields of deep learning and spiking neural networks evolve very rapidly. We endeavour to treat this document as a 'dynamic' manuscript that will continue to be updated as the common practices in training SNNs also change. A series of companion interactive tutorials complementary to this paper using our Python package, snnTorch, are also made available. See https://snntorch.readthedocs.io/en/latest/tutorials/index.html .

Nick Alonso, Emre Neftci

Predictive coding (PC) is a general theory of cortical function. The local, gradient-based learning rules found in one kind of PC model have recently been shown to closely approximate backpropagation. This finding suggests that this gradient-based PC model may be useful for understanding how the brain solves the credit assignment problem. The model may also be useful for developing local learning algorithms that are compatible with neuromorphic hardware. In this paper, we modify this PC model so that it better fits biological constraints, including the constraints that neurons can only have positive firing rates and the constraint that synapses only flow in one direction. We also compute the gradient-based weight and activity updates given the modified activity values. We show that, under certain conditions, these modified PC networks perform as well or nearly as well on MNIST data as the unmodified PC model and networks trained with backpropagation.

Hin Wai Lui, Emre Neftci

To achieve the low latency, high throughput, and energy efficiency benefits of Spiking Neural Networks (SNNs), reducing the memory and compute requirements when running on a neuromorphic hardware is an important step. Neuromorphic architecture allows massively parallel computation with variable and local bit-precisions. However, how different bit-precisions should be allocated to different layers or connections of the network is not trivial. In this work, we demonstrate how a layer-wise Hessian trace analysis can measure the sensitivity of the loss to any perturbation of the layer's weights, and this can be used to guide the allocation of a layer-specific bit-precision when quantizing an SNN. In addition, current gradient based methods of SNN training use a complex neuron model with multiple state variables, which is not ideal for compute and memory efficiency. To address this challenge, we present a simplified neuron model that reduces the number of state variables by 4-fold while still being compatible with gradient based training. We find that the impact on model accuracy when using a layer-wise bit-precision correlated well with that layer's Hessian trace. The accuracy of the optimal quantized network only dropped by 0.2%, yet the network size was reduced by 58%. This reduces memory usage and allows fixed-point arithmetic with simpler digital circuits to be used, increasing the overall throughput and energy efficiency.

Kenneth Stewart, Andreea Danielescu, Timothy Shea et al.

Neuromorphic hardware equipped with learning capabilities can adapt to new, real-time data. While models of Spiking Neural Networks (SNNs) can now be trained using gradient descent to reach an accuracy comparable to equivalent conventional neural networks, such learning often relies on external labels. However, real-world data is unlabeled which can make supervised methods inapplicable. To solve this problem, we propose a Hybrid Guided Variational Autoencoder (VAE) which encodes event based data sensed by a Dynamic Vision Sensor (DVS) into a latent space representation using an SNN. These representations can be used as an embedding to measure data similarity and predict labels in real-world data. We show that the Hybrid Guided-VAE achieves 87% classification accuracy on the DVSGesture dataset and it can encode the sparse, noisy inputs into an interpretable latent space representation, visualized through T-SNE plots. We also implement the encoder component of the model on neuromorphic hardware and discuss the potential for our algorithm to enable real-time learning from real-world event data.

Sourav Dutta, Georgios Detorakis, Abhishek Khanna et al.

Many real-world mission-critical applications require continual online learning from noisy data and real-time decision making with a defined confidence level. Probabilistic models and stochastic neural networks can explicitly handle uncertainty in data and allow adaptive learning-on-the-fly, but their implementation in a low-power substrate remains a challenge. Here, we introduce a novel hardware fabric that implements a new class of stochastic NN called Neural-Sampling-Machine that exploits stochasticity in synaptic connections for approximate Bayesian inference. Harnessing the inherent non-linearities and stochasticity occurring at the atomic level in emerging materials and devices allows us to capture the synaptic stochasticity occurring at the molecular level in biological synapses. We experimentally demonstrate in-silico hybrid stochastic synapse by pairing a ferroelectric field-effect transistor -based analog weight cell with a two-terminal stochastic selector element. Such a stochastic synapse can be integrated within the well-established crossbar array architecture for compute-in-memory. We experimentally show that the inherent stochastic switching of the selector element between the insulator and metallic state introduces a multiplicative stochastic noise within the synapses of NSM that samples the conductance states of the FeFET, both during learning and inference. We perform network-level simulations to highlight the salient automatic weight normalization feature introduced by the stochastic synapses of the NSM that paves the way for continual online learning without any offline Batch Normalization. We also showcase the Bayesian inferencing capability introduced by the stochastic synapse during inference mode, thus accounting for uncertainty in data. We report 98.25%accuracy on standard image classification task as well as estimation of data uncertainty in rotated samples.

Jinwei Xing, Takashi Nagata, Kexin Chen et al.

Despite the recent success of deep reinforcement learning (RL), domain adaptation remains an open problem. Although the generalization ability of RL agents is critical for the real-world applicability of Deep RL, zero-shot policy transfer is still a challenging problem since even minor visual changes could make the trained agent completely fail in the new task. To address this issue, we propose a two-stage RL agent that first learns a latent unified state representation (LUSR) which is consistent across multiple domains in the first stage, and then do RL training in one source domain based on LUSR in the second stage. The cross-domain consistency of LUSR allows the policy acquired from the source domain to generalize to other target domains without extra training. We first demonstrate our approach in variants of CarRacing games with customized manipulations, and then verify it in CARLA, an autonomous driving simulator with more complex and realistic visual observations. Our results show that this approach can achieve state-of-the-art domain adaptation performance in related RL tasks and outperforms prior approaches based on latent-representation based RL and image-to-image translation.

Kenneth Stewart, Garrick Orchard, Sumit Bam Shrestha et al.

We present the Surrogate-gradient Online Error-triggered Learning (SOEL) system for online few-shot learning on neuromorphic processors. The SOEL learning system uses a combination of transfer learning and principles of computational neuroscience and deep learning. We show that partially trained deep Spiking Neural Networks (SNNs) implemented on neuromorphic hardware can rapidly adapt online to new classes of data within a domain. SOEL updates trigger when an error occurs, enabling faster learning with fewer updates. Using gesture recognition as a case study, we show SOEL can be used for online few-shot learning of new classes of pre-recorded gesture data and rapid online learning of new gestures from data streamed live from a Dynamic Active-pixel Vision Sensor to an Intel Loihi neuromorphic research processor.

Georgios Detorakis, Sourav Dutta, Abhishek Khanna et al.

Multiplicative stochasticity such as Dropout improves the robustness and generalizability of deep neural networks. Here, we further demonstrate that always-on multiplicative stochasticity combined with simple threshold neurons are sufficient operations for deep neural networks. We call such models Neural Sampling Machines (NSM). We find that the probability of activation of the NSM exhibits a self-normalizing property that mirrors Weight Normalization, a previously studied mechanism that fulfills many of the features of Batch Normalization in an online fashion. The normalization of activities during training speeds up convergence by preventing internal covariate shift caused by changes in the input distribution. The always-on stochasticity of the NSM confers the following advantages: the network is identical in the inference and learning phases, making the NSM suitable for online learning, it can exploit stochasticity inherent to a physical substrate such as analog non-volatile memories for in-memory computing, and it is suitable for Monte Carlo sampling, while requiring almost exclusively addition and comparison operations. We demonstrate NSMs on standard classification benchmarks (MNIST and CIFAR) and event-based classification benchmarks (N-MNIST and DVS Gestures). Our results show that NSMs perform comparably or better than conventional artificial neural networks with the same architecture.

Kenneth Stewart, Garrick Orchard, Sumit Bam Shrestha et al.

Recent work suggests that synaptic plasticity dynamics in biological models of neurons and neuromorphic hardware are compatible with gradient-based learning (Neftci et al., 2019). Gradient-based learning requires iterating several times over a dataset, which is both time-consuming and constrains the training samples to be independently and identically distributed. This is incompatible with learning systems that do not have boundaries between training and inference, such as in neuromorphic hardware. One approach to overcome these constraints is transfer learning, where a portion of the network is pre-trained and mapped into hardware and the remaining portion is trained online. Transfer learning has the advantage that pre-training can be accelerated offline if the task domain is known, and few samples of each class are sufficient for learning the target task at reasonable accuracies. Here, we demonstrate on-line surrogate gradient few-shot learning on Intel's Loihi neuromorphic research processor using features pre-trained with spike-based gradient backpropagation-through-time. Our experimental results show that the Loihi chip can learn gestures online using a small number of shots and achieve results that are comparable to the models simulated on a conventional processor.

Jacques Kaiser, Alexander Friedrich, J. Camilo Vasquez Tieck et al.

Spike-based communication between biological neurons is sparse and unreliable. This enables the brain to process visual information from the eyes efficiently. Taking inspiration from biology, artificial spiking neural networks coupled with silicon retinas attempt to model these computations. Recent findings in machine learning allowed the derivation of a family of powerful synaptic plasticity rules approximating backpropagation for spiking networks. Are these rules capable of processing real-world visual sensory data? In this paper, we evaluate the performance of Event-Driven Random Back-Propagation (eRBP) at learning representations from event streams provided by a Dynamic Vision Sensor (DVS). First, we show that eRBP matches state-of-the-art performance on the DvsGesture dataset with the addition of a simple covert attention mechanism. By remapping visual receptive fields relatively to the center of the motion, this attention mechanism provides translation invariance at low computational cost compared to convolutions. Second, we successfully integrate eRBP in a real robotic setup, where a robotic arm grasps objects according to detected visual affordances. In this setup, visual information is actively sensed by a DVS mounted on a robotic head performing microsaccadic eye movements. We show that our method classifies affordances within 100ms after microsaccade onset, which is comparable to human performance reported in behavioral study. Our results suggest that advances in neuromorphic technology and plasticity rules enable the development of autonomous robots operating at high speed and low energy consumption.

Jacques Kaiser, Hesham Mostafa, Emre Neftci

A growing body of work underlines striking similarities between biological neural networks and recurrent, binary neural networks. A relatively smaller body of work, however, discusses similarities between learning dynamics employed in deep artificial neural networks and synaptic plasticity in spiking neural networks. The challenge preventing this is largely caused by the discrepancy between the dynamical properties of synaptic plasticity and the requirements for gradient backpropagation. Learning algorithms that approximate gradient backpropagation using locally synthesized gradients can overcome this challenge. Here, we show that synthetic gradients enable the derivation of Deep Continuous Local Learning (DECOLLE) in spiking neural networks. DECOLLE is capable of learning deep spatio-temporal representations from spikes relying solely on local information. Synaptic plasticity rules are derived systematically from user-defined cost functions and neural dynamics by leveraging existing autodifferentiation methods of machine learning frameworks. We benchmark our approach on the MNIST and the event-based neuromorphic DvsGesture dataset, on which DECOLLE performs comparably to the state-of-the-art. DECOLLE networks provide continuously learning machines that are relevant to biology and supportive of event-based, low-power computer vision architectures matching the accuracies of conventional computers on tasks where temporal precision and speed are essential.

Georgios Detorakis, Travis Bartley, Emre Neftci

Neural networks are commonly trained to make predictions through learning algorithms. Contrastive Hebbian learning, which is a powerful rule inspired by gradient backpropagation, is based on Hebb's rule and the contrastive divergence algorithm. It operates in two phases, the forward (or free) phase, where the data are fed to the network, and a backward (or clamped) phase, where the target signals are clamped to the output layer of the network and the feedback signals are transformed through the transpose synaptic weight matrices. This implies symmetries at the synaptic level, for which there is no evidence in the brain. In this work, we propose a new variant of the algorithm, called random contrastive Hebbian learning, which does not rely on any synaptic weights symmetries. Instead, it uses random matrices to transform the feedback signals during the clamped phase, and the neural dynamics are described by first order non-linear differential equations. The algorithm is experimentally verified by solving a Boolean logic task, classification tasks (handwritten digits and letters), and an autoencoding task. This article also shows how the parameters affect learning, especially the random matrices. We use the pseudospectra analysis to investigate further how random matrices impact the learning process. Finally, we discuss the biological plausibility of the proposed algorithm, and how it can give rise to better computational models for learning.

Georgios Detorakis, Sadique Sheik, Charles Augustine et al.

Embedded, continual learning for autonomous and adaptive behavior is a key application of neuromorphic hardware. However, neuromorphic implementations of embedded learning at large scales that are both flexible and efficient have been hindered by a lack of a suitable algorithmic framework. As a result, the most neuromorphic hardware is trained off-line on large clusters of dedicated processors or GPUs and transferred post hoc to the device. We address this by introducing the neural and synaptic array transceiver (NSAT), a neuromorphic computational framework facilitating flexible and efficient embedded learning by matching algorithmic requirements and neural and synaptic dynamics. NSAT supports event-driven supervised, unsupervised and reinforcement learning algorithms including deep learning. We demonstrate the NSAT in a wide range of tasks, including the simulation of Mihalas-Niebur neuron, dynamic neural fields, event-driven random back-propagation for event-based deep learning, event-based contrastive divergence for unsupervised learning, and voltage-based learning rules for sequence learning. We anticipate that this contribution will establish the foundation for a new generation of devices enabling adaptive mobile systems, wearable devices, and robots with data-driven autonomy.

Emre Neftci, Charles Augustine, Somnath Paul et al.

An ongoing challenge in neuromorphic computing is to devise general and computationally efficient models of inference and learning which are compatible with the spatial and temporal constraints of the brain. One increasingly popular and successful approach is to take inspiration from inference and learning algorithms used in deep neural networks. However, the workhorse of deep learning, the gradient descent Back Propagation (BP) rule, often relies on the immediate availability of network-wide information stored with high-precision memory, and precise operations that are difficult to realize in neuromorphic hardware. Remarkably, recent work showed that exact backpropagated weights are not essential for learning deep representations. Random BP replaces feedback weights with random ones and encourages the network to adjust its feed-forward weights to learn pseudo-inverses of the (random) feedback weights. Building on these results, we demonstrate an event-driven random BP (eRBP) rule that uses an error-modulated synaptic plasticity for learning deep representations in neuromorphic computing hardware. The rule requires only one addition and two comparisons for each synaptic weight using a two-compartment leaky Integrate & Fire (I&F) neuron, making it very suitable for implementation in digital or mixed-signal neuromorphic hardware. Our results show that using eRBP, deep representations are rapidly learned, achieving nearly identical classification accuracies compared to artificial neural network simulations on GPUs, while being robust to neural and synaptic state quantizations during learning.

S. Burc Eryilmaz, Emre Neftci, Siddharth Joshi et al.

Current large scale implementations of deep learning and data mining require thousands of processors, massive amounts of off-chip memory, and consume gigajoules of energy. Emerging memory technologies such as nanoscale two-terminal resistive switching memory devices offer a compact, scalable and low power alternative that permits on-chip co-located processing and memory in fine-grain distributed parallel architecture. Here we report first use of resistive switching memory devices for implementing and training a Restricted Boltzmann Machine (RBM), a generative probabilistic graphical model as a key component for unsupervised learning in deep networks. We experimentally demonstrate a 45-synapse RBM realized with 90 resistive switching phase change memory (PCM) elements trained with a bio-inspired variant of the Contrastive Divergence (CD) algorithm, implementing Hebbian and anti-Hebbian weight updates. The resistive PCM devices show a two-fold to ten-fold reduction in error rate in a missing pixel pattern completion task trained over 30 epochs, compared to untrained case. Measured programming energy consumption is 6.1 nJ per epoch with the resistive switching PCM devices, a factor of ~150 times lower than conventional processor-memory systems. We analyze and discuss the dependence of learning performance on cycle-to-cycle variations as well as number of gradual levels in the PCM analog memory devices.

Bruno U. Pedroni, Sadique Sheik, Siddharth Joshi et al.

Spike-timing-dependent plasticity (STDP) incurs both causal and acausal synaptic weight updates, for negative and positive time differences between pre-synaptic and post-synaptic spike events. For realizing such updates in neuromorphic hardware, current implementations either require forward and reverse lookup access to the synaptic connectivity table, or rely on memory-intensive architectures such as crossbar arrays. We present a novel method for realizing both causal and acausal weight updates using only forward lookup access of the synaptic connectivity table, permitting memory-efficient implementation. A simplified implementation in FPGA, using a single timer variable for each neuron, closely approximates exact STDP cumulative weight updates for neuron refractory periods greater than 10 ms, and reduces to exact STDP for refractory periods greater than the STDP time window. Compared to conventional crossbar implementation, the forward table-based implementation leads to substantial memory savings for sparsely connected networks supporting scalable neuromorphic systems with fully reconfigurable synaptic connectivity and plasticity.

Peter U. Diehl, Guido Zarrella, Andrew Cassidy et al.

In recent years the field of neuromorphic low-power systems that consume orders of magnitude less power gained significant momentum. However, their wider use is still hindered by the lack of algorithms that can harness the strengths of such architectures. While neuromorphic adaptations of representation learning algorithms are now emerging, efficient processing of temporal sequences or variable length-inputs remain difficult. Recurrent neural networks (RNN) are widely used in machine learning to solve a variety of sequence learning tasks. In this work we present a train-and-constrain methodology that enables the mapping of machine learned (Elman) RNNs on a substrate of spiking neurons, while being compatible with the capabilities of current and near-future neuromorphic systems. This "train-and-constrain" method consists of first training RNNs using backpropagation through time, then discretizing the weights and finally converting them to spiking RNNs by matching the responses of artificial neurons with those of the spiking neurons. We demonstrate our approach by mapping a natural language processing task (question classification), where we demonstrate the entire mapping process of the recurrent layer of the network on IBM's Neurosynaptic System "TrueNorth", a spike-based digital neuromorphic hardware architecture. TrueNorth imposes specific constraints on connectivity, neural and synaptic parameters. To satisfy these constraints, it was necessary to discretize the synaptic weights and neural activities to 16 levels, and to limit fan-in to 64 inputs. We find that short synaptic delays are sufficient to implement the dynamical (temporal) aspect of the RNN in the question classification task. The hardware-constrained model achieved 74% accuracy in question classification while using less than 0.025% of the cores on one TrueNorth chip, resulting in an estimated power consumption of ~17 uW.

Peter U. Diehl, Bruno U. Pedroni, Andrew Cassidy et al.

We present an approach to constructing a neuromorphic device that responds to language input by producing neuron spikes in proportion to the strength of the appropriate positive or negative emotional response. Specifically, we perform a fine-grained sentiment analysis task with implementations on two different systems: one using conventional spiking neural network (SNN) simulators and the other one using IBM's Neurosynaptic System TrueNorth. Input words are projected into a high-dimensional semantic space and processed through a fully-connected neural network (FCNN) containing rectified linear units trained via backpropagation. After training, this FCNN is converted to a SNN by substituting the ReLUs with integrate-and-fire neurons. We show that there is practically no performance loss due to conversion to a spiking network on a sentiment analysis test set, i.e. correlations between predictions and human annotations differ by less than 0.02 comparing the original DNN and its spiking equivalent. Additionally, we show that the SNN generated with this technique can be mapped to existing neuromorphic hardware -- in our case, the TrueNorth chip. Mapping to the chip involves 4-bit synaptic weight discretization and adjustment of the neuron thresholds. The resulting end-to-end system can take a user input, i.e. a word in a vocabulary of over 300,000 words, and estimate its sentiment on TrueNorth with a power consumption of approximately 50 uW.

Maruan Al-Shedivat, Emre Neftci, Gert Cauwenberghs

The goal of a generative model is to capture the distribution underlying the data, typically through latent variables. After training, these variables are often used as a new representation, more effective than the original features in a variety of learning tasks. However, the representations constructed by contemporary generative models are usually point-wise deterministic mappings from the original feature space. Thus, even with representations robust to class-specific transformations, statistically driven models trained on them would not be able to generalize when the labeled data is scarce. Inspired by the stochasticity of the synaptic connections in the brain, we introduce Energy-based Stochastic Ensembles. These ensembles can learn non-deterministic representations, i.e., mappings from the feature space to a family of distributions in the latent space. These mappings are encoded in a distribution over a (possibly infinite) collection of models. By conditionally sampling models from the ensemble, we obtain multiple representations for every input example and effectively augment the data. We propose an algorithm similar to contrastive divergence for training restricted Boltzmann stochastic ensembles. Finally, we demonstrate the concept of the stochastic representations on a synthetic dataset as well as test them in the one-shot learning scenario on MNIST.

Emre Neftci, Srinjoy Das, Bruno Pedroni et al.

Restricted Boltzmann Machines (RBMs) and Deep Belief Networks have been demonstrated to perform efficiently in a variety of applications, such as dimensionality reduction, feature learning, and classification. Their implementation on neuromorphic hardware platforms emulating large-scale networks of spiking neurons can have significant advantages from the perspectives of scalability, power dissipation and real-time interfacing with the environment. However the traditional RBM architecture and the commonly used training algorithm known as Contrastive Divergence (CD) are based on discrete updates and exact arithmetics which do not directly map onto a dynamical neural substrate. Here, we present an event-driven variation of CD to train a RBM constructed with Integrate & Fire (I&F) neurons, that is constrained by the limitations of existing and near future neuromorphic hardware platforms. Our strategy is based on neural sampling, which allows us to synthesize a spiking neural network that samples from a target Boltzmann distribution. The recurrent activity of the network replaces the discrete steps of the CD algorithm, while Spike Time Dependent Plasticity (STDP) carries out the weight updates in an online, asynchronous fashion. We demonstrate our approach by training an RBM composed of leaky I&F neurons with STDP synapses to learn a generative model of the MNIST hand-written digit dataset, and by testing it in recognition, generation and cue integration tasks. Our results contribute to a machine learning-driven approach for synthesizing networks of spiking neurons capable of carrying out practical, high-level functionality.