Recent advances in retinal neuroscience inspired the development of hardware and software systems that leverage evolutionarily derived retinal computations for real-world computer vision applications. In this work, we propose a novel, reconfigurable CMOS circuit designed specifically for Looming Detection (LD), a key retinal computation associated with detecting rapidly approaching objects and potential threats. We analyze the circuit's performance using real-world data from controlled laboratory environments and hardware-aware algorithmic simulations, demonstrating its accuracy in identifying potential collisions and avoiding false-positives from mundane object movements. Furthermore, we evaluate the CMOS hardware characteristics using GlobalFoundries' 22nm FD-SOI technology, exhibiting a 0.36 pJ energy consumption per looming spike for a unit kernel size of 5 × 5 pixels. This work contributes a foundational approach to adaptive hardware-software co-design by integrating advances in retinal neuroscience with modern CMOS technology to address real-time LD with in-sensor computations.

As machine learning models increasingly process sensitive data, understanding their vulnerability to privacy attacks is vital. Membership Inference Attack (MIA), which infer whether specific data points were used during training, is one such privacy risk. Previous work suggests that Spiking Neural Networks (SNNs), which rely on event-driven computation and discrete spike-based encoding, exhibit greater resilience to MIAs compared to Artificial Neural Networks (ANNs). This resilience is attributed to their non-differentiable activations and inherent stochasticity, which reduce the correlation between model responses and individual training samples. To further enhance privacy in SNNs, we explore two techniques: Quantization and Surrogate Gradients. Quantization, which reduces model precision to limit information leakage, has been shown to improve privacy resilience in ANNs. Since SNNs exhibit sparse and irregular activations, quantization may have an even stronger effect on disrupting the activation patterns exploited by MIAs. In this study, we compare the vulnerability of SNNs and ANNs to MIAs under weight and activation quantization across multiple datasets. We evaluate privacy vulnerability using the attack model’s Receiver Operating Characteristic (ROC) curve’s Area Under the Curve (AUC) metric, where lower values indicate stronger privacy protection, and assess model accuracy to quantify the privacy-accuracy trade-off. Our results show that quantization enhances privacy in both architectures with minimal performance degradation, but full-precision SNNs remain more resilient than even quantized ANNs. Additionally, we examine the impact of surrogate gradients on privacy in SNNs. Among the five surrogate gradients evaluated, Spike Rate Escape provides the best privacy-accuracy trade-off, while Arctangent (aTan) increases vulnerability to MIAs. These findings reinforce SNNs’ inherent privacy advantages and demonstrate that both quantization and surrogate gradient selection can further influence privacy-accuracy trade-offs in SNNs.

While machine learning (ML) models are becoming mainstream, including in critical application domains, concerns have been raised about the increasing risk of sensitive data leakage. Various privacy attacks, such as membership inference attacks (MIAs), have been developed to extract data from trained ML models, posing significant risks to data confidentiality. While the predominant work in the ML community considers traditional Artificial Neural Networks (ANNs) as the default neural model, neuromorphic architectures, such as Spiking Neural Networks (SNNs), have recently emerged as an attractive alternative mainly due to their significantly low power consumption. These architectures process information through discrete events, i.e., spikes, to mimic the functioning of biological neurons in the brain. While the privacy issues have been extensively investigated in the context of traditional ANNs, they remain largely unexplored in neuromorphic architectures, and little work has been dedicated to investigating their privacy-preserving properties. In this paper, we investigate the question of whether SNNs have inherent privacy-preserving advantages. Specifically, we investigate SNNs’ privacy properties through the lens of MIAs across diverse datasets, in comparison with ANNs. We explore the impact of different learning algorithms (surrogate gradient and evolutionary learning), programming frameworks (snnTorch, TENNLab, and LAVA), and various parameters on the resilience of SNNs against MIA. Our experiments reveal that SNNs demonstrate consistently superior privacy preservation compared to ANNs, with evolutionary algorithms further enhancing their resilience. For example, on the CIFAR-10 dataset, SNNs achieve an AUC as low as 0.59 compared to 0.82 for ANNs, and on CIFAR-100, SNNs maintain a low AUC of 0.58, whereas ANNs reach 0.88. Furthermore, we investigate the privacy-utility trade-off through Differentially Private Stochastic Gradient Descent (DPSGD), observing that SNNs incur a notably lower accuracy drop than ANNs under equivalent privacy constraints.

Graph neural networks have emerged as a specialized branch of deep learning, designed to address problems where pairwise relations between objects are crucial. Recent advancements utilize graph convolutional neural networks to extract features within graph structures. Despite promising results, these methods face challenges in real-world applications due to sparse features, resulting in inefficient resource utilization. Recent studies draw inspiration from the mammalian brain and employ spiking neural networks to model and learn graph structures. However, these approaches are limited to traditional Von Neumann-based computing systems, which still face hardware inefficiencies. In this study, we present a fully neuromorphic implementation of spiking graph neural networks designed for Loihi 2. We optimize network parameters using Lava Bayesian Optimization, a novel hyperparameter optimization system compatible with neuromorphic computing architectures. We showcase the performance benefits of combining neuromorphic Bayesian optimization with our approach for citation graph classification using fixed-precision spiking neurons. Our results demonstrate the capability of integer-precision, Loihi 2 compatible spiking neural networks in performing citation graph classification with comparable accuracy to existing floating point implementations.

Performing optimization with event-based asynchronous neuromorphic systems presents significant challenges. Intel’s neuromorphic computing framework, Lava, offers an abstract application programming interface designed for constructing event-based computational graphs. In this study, we introduce a novel framework tailored for asynchronous Bayesian optimization that is also compatible with Loihi 2. We showcase the capability of our asynchronous optimization framework by connecting it with a graph-based satellite scheduling problem running on physical Loihi 2 hardware.

With the mainstream integration of machine learning into security-sensitive domains such as healthcare and finance, con-cerns about data privacy have intensified. Conventional artificial neural networks (ANNs) have been found vulnerable to several attacks that can leak sensitive data. Particularly, model inversion (MI) attacks enable the reconstruction of data samples that have been used to train the model. Neuromorphic architectures have emerged as a paradigm shift in neural computing, enabling asynchronous and energy-efficient computation. However, little to no existing work has investigated the privacy of neuromorphic architectures against model inversion. Our study is motivated by the intuition that the non-differentiable aspect of spiking neural networks (SNNs) might result in inherent privacy-preserving properties, especially against gradient-based attacks. To investigate this hypothesis, we propose a thorough exploration of SNNs' privacy-preserving capabilities. Specifically, we develop novel inversion attack strategies that are comprehensively designed to target SNNs, offering a comparative analysis with their conventional ANN counterparts. Our experiments, conducted on diverse event-based and static datasets, demonstrate the effectiveness of the proposed attack strategies and therefore questions the assumption of inherent privacy-preserving in neuromorphic architectures.

Despite significant efforts to enhance the resilience of computer systems against malware attacks, the abundance of exploitable vulnerabilities remains a significant challenge. While preventing compromises is difficult, traditional signature-based static analysis techniques are susceptible to bypassing through metamorphic/polymorphic malware or zero-day exploits. Dynamic detection techniques, particularly those utilizing machine learning (ML), have the potential to identify previously unseen signatures by monitoring program behavior. However, classical ML models are power and resource intensive and may not be suitable for devices with limited budgets. This constraint creates a challenging tradeoff between security and resource utilization, which cannot be fully addressed through model compression and pruning. In contrast, neuromorphic architectures offer a promising solution for low-power brain-inspired systems. In this work, we explore the novel use of neuromorphic architectures for malware detection. We accomplish this by encoding sub-semantic micro-architecture level features in the spiking domain and proposing a Spiking Neural Network (SNN) architecture for hardware-aware malware detection. Our results demonstrate promising malware detection performance with an 89% F1-score. Ultimately, this work advocates that neuromorphic architectures, due to their low power consumption, represent a promising candidate for malware detection, especially for energy-constraint processors in IoT and Edge devices.

Graph convolutional networks leverage both graph structures and features on nodes and edges for improved learning performance in comparison with classical machine learning approaches. Spiking neuromorphic computers natively implement network-like computation and have been shown to be successful at implementing graph learning without features. Incorporating graph features brings the challenge of efficient feature representation and balancing the contribution of topology and features in learning. In this work, we present our design of a simulated network of spiking neurons to perform semi-supervised learning on graph data using both the graph structure and the node features. We explore various design choices, present preliminary results, and discuss the opportunities for using neuromorphic computers for this task in the future.

Abstract Asynchronous evolutionary algorithms are becoming increasingly popular as a means of making full use of many processors while solving computationally expensive search and optimization problems. These algorithms excel at keeping large clusters fully utilized, but may sometimes inefficiently sample an excess of fast‐evaluating solutions at the expense of higher‐quality, slow‐evaluating ones. We have previously introduced a steady‐state parent selection strategy, SWEET (“Selection whilE EvaluaTing”), that sometimes selects individuals that are still being evaluated and allows them to reproduce early. We perform a takeover‐time analysis that confirms that this strategy gives slow‐evaluating individuals that have higher fitnesses an increased ability to multiply in the population. We also find that SWEET appears effective at improving optimization performance on problems in which solution quality is positively correlated with evaluation time. We evaluate our approach on six simulated real‐valued optimization problems and three real‐world applications: an autonomous vehicle controller problem that involves tuning a spiking neural network and two adversarial EA problems. We further evaluate SWEET versus a basic asynchronous process in a simulated setting. We present evidence that SWEET outperforms basic asynchronous processes in a use‐case in which performance is positively correlated with evaluation time, and performs comparably (and often better) than basic asynchronous processes in several use‐cases where performance is negatively correlated with evaluation time. That said, in the cases where performance and evaluation time are negatively correlated the variance of outcomes for SWEET is notably high.

Neuroevolution has had significant success over recent years, but there has been relatively little work applying neuroevolution approaches to spiking neural networks (SNNs). SNNs are a type of neural networks that include temporal processing component, are not easily trained using other methods because of their lack of differentiable activation functions, and can be deployed into energy-efficient neuromorphic hardware. In this work, we investigate two evolutionary approaches for training SNNs. We explore the impact of the hyperparameters of the evolutionary approaches, including tournament size, population size, and representation type, on the performance of the algorithms. We present a multi-objective Bayesian-based hyperparameter optimization approach to tune the hyperparameters to produce the most accurate and smallest SNNs. We show that the hyperparameters can significantly affect the performance of these algorithms. We also perform sensitivity analysis and demonstrate that every hyperparameter value has the potential to perform well, assuming other hyperparameter values are set correctly.

There are a wide variety of training approaches for spiking neural networks for neuromorphic deployment. However, it is often not clear how these training algorithms perform or compare when applied across multiple neuromorphic hardware platforms and multiple datasets. In this work, we present a software framework for comparing performance across four neuromorphic training algorithms across three neuromorphic simulators and four simple classification tasks. We introduce an approach for training a spiking neural network using a decision tree, and we compare this approach to training algorithms based on evolutionary algorithms, back-propagation, and reservoir computing. We present a hyperparameter optimization approach to tune the hyperparameters of the algorithm, and show that these optimized hyperparameters depend on the processor, algorithm, and classification task. Finally, we compare the performance of the optimized algorithms across multiple metrics, including accuracy, training time, and resulting network size, and we show that there is not one best training algorithm across all datasets and performance metrics.

Training neural networks for neuromorphic deployment is non-trivial. There have been a variety of approaches proposed to adapt back-propagation or back-propagation-like algorithms appropriate for training. Considering that these networks often have very different performance characteristics than traditional neural networks, it is often unclear how to set either the network topology or the hyperparameters to achieve optimal performance. In this work, we introduce a Bayesian approach for optimizing the hyperparameters of an algorithm for training binary communication networks that can be deployed to neuromorphic hardware. We show that by optimizing the hyperparameters on this algorithm for each dataset, we can achieve improvements in accuracy over the previous state-of-the-art for this algorithm on each dataset (by up to 15 percent). This jump in performance continues to emphasize the potential when converting traditional neural networks to binary communication applicable to neuromorphic hardware.

Though robustness and resilience are commonly quoted as features of neuromorphic computing systems, the expected performance of neuromorphic systems in the face of hardware failures is not clear. In this work, we study the effect of failures on the performance of four different training algorithms for spiking neural networks on neuromorphic systems: two back-propagation-based training approaches (Whetstone and SLAYER), a liquid state machine or reservoir computing approach, and an evolutionary optimization-based approach (EONS). We show that these four different approaches have very different resilience characteristics with respect to simulated hardware failures. We then analyze an approach for training more resilient spiking neural networks using the evolutionary optimization approach. We show how this approach produces more resilient networks and discuss how it can be extended to other spiking neural network training approaches as well.

Designing spiking neural networks for neuromorphic deployment is a non-trivial task. It is further complicated when there are resource constraints for the neuromorphic implementation, such as size or power constraints, that may be present in edge applications. In this work, we utilize a previously presented approach, EONS, to design spiking neural networks for a memristive neuromorphic implementation for scientific data applications. We specifically use a multi-objective approach in EONS to maximize network accuracy on the scientific data application task, but also to minimize network size and energy. We illustrate that EONS determines both the network structure and the parameters, removing the burden from the user on determining the appropriate spiking neural network structure, and we show that the resulting networks are very different from the layered structure of typical neural networks. Finally, we show that the multi-objective approach produces smaller, more energy efficient networks than the original EONS approach and produces comparable accuracy to a back-propagation style training approach.

Designing a neuromorphic computing system involves selection of several hyperparameters that not only affect the accuracy of the framework, but also the energy efficiency and speed of inference and training. These hyperparameters might be inherent to the training of the spiking neural network (SNN), the input/output encoding of the real-world data to spikes, or the underlying neuromorphic hardware. In this work, we present a Bayesian-based hyperparameter optimization approach for spiking neuromorphic systems, and we show how this optimization framework can lead to significant improvement in designing accurate neuromorphic computing systems. In particular, we show that this hyperparameter optimization approach can discover the same optimal hyperparameter set for input encoding as a grid search, but with far fewer evaluations and far less time. We also show the impact of hardware-specific hyperparameters on the performance of the system, and we demonstrate that by optimizing these hyperparameters, we can achieve significantly better application performance.

As deep neural networks have been deployed in more and more applications over the past half decade and are finding their way into an ever increasing number of operational systems, their energy consumption becomes a concern whether running in the datacenter or on edge devices. Hyperparameter optimization and automated network design for deep learning is a quickly growing field, but much of the focus has remained only on optimizing for the performance of the machine learning task. In this work, we demonstrate that the best performing networks created through this automated network design process have radically different computational characteristics (e.g. energy usage, model size, inference time), presenting the opportunity to utilize this optimization process to make deep learning networks more energy efficient and deployable to smaller devices. Optimizing for these computational characteristics is critical as the number of applications of deep learning continues to expand.

The ever increasing computational cost of Deep Neural Networks (DNN) and the demand for energy efficient hardware for DNN acceleration has made accuracy and hardware cost co-optimization for DNNs tremendously important, especially for edge devices. Owing to the large parameter space and cost of evaluating each parameter in the search space, manually tuning of DNN hyperparameters is impractical. Automatic joint DNN and hardware hyperparameter optimization is indispensable for such problems. Bayesian optimization-based approaches have shown promising results for hyperparameter optimization of DNNs. However, most of these techniques have been developed without considering the underlying hardware, thereby leading to inefficient designs. Further, the few works that perform joint optimization are not generalizable and mainly focus on CMOS-based architectures. In this work, we present a novel pseudo agent-based multiobjective hyperparameter optimization (PABO) for maximizing the DNN performance while obtaining low hardware cost. Compared to the existing methods, our work poses a theoretically different approach for joint optimization of accuracy and hardware cost and focuses on memristive crossbar based accelerators. PABO uses a supervisor agent to establish connections between the posterior Gaussian distribution models of network accuracy and hardware cost requirements. The agent reduces the mathematical complexity of the co-optimization problem by removing unnecessary computations and updates of acquisition functions, thereby achieving significant speed-ups for the optimization procedure. PABO outputs a Pareto frontier that underscores the trade-offs between designing high-accuracy and hardware efficiency. Our results demonstrate a superior performance compared to the state-of-the-art methods both in terms of accuracy and computational speed (~100x speed up).