As one of the most complex systems known to science, modeling brain behavior and function is both fascinating and extremely difficult. Empirical data is increasingly available from ex vivo human brain organoids and surgical samples, as well as in vivo animal models, so the problem of modeling the behavior of large-scale neuronal systems is more relevant than ever. The statistical physics concept of a mean-field model offers a tractable way to bridge the gap between single-neuron and population-level descriptions of neuronal activity, by modeling the behavior of a single representative neuron and extending this to the population. However, existing neural mean-field methods typically either take the limit of small interaction sizes, or are applicable only to the specific neuron models for which they were derived. This paper derives a mean-field model by fitting a transfer function called Refractory SoftPlus, which is simple yet applicable to a broad variety of neuron types. The transfer function is fitted numerically to simulated spike time data, and is entirely agnostic to the underlying neuronal dynamics. The resulting mean-field model predicts the response of a network of randomly connected neurons to a time-varying external stimulus with a high degree of accuracy. Furthermore, it enables an accurate approximate bifurcation analysis as a function of the level of recurrent input. This model does not assume large presynaptic rates or small postsynaptic potential size, allowing mean-field models to be developed even for populations with large interaction terms.

Abstract Coastline detection is vital for coastal management, involving frequent observation and assessment to understand coastal dynamics and inform decisions on environmental protection. 
Continuous streaming of high-resolution images demands robust data processing and storage solutions to manage large datasets efficiently, posing challenges that require innovative solutions for real-time analysis and meaningful insights extraction.
This work leverages low-latency event-based vision sensors coupled with neuromorphic hardware in an attempt to decrease a two-fold challenge, reducing the computational burden to ~0.375 mW whilst obtaining a coastline detection map in as little as 20 ms.
The proposed Spiking Neural Network (SNN) runs on the SpiNNaker neuromorphic platform using a total of 18040 neurons reaching 98.33% accuracy.
The model has been characterised and evaluated by computing the accuracy of Intersection over Union (IoU) scores over the ground truth of a real-world coastline dataset across different time windows. The system's robustness was further assessed by evaluating its ability to avoid coastline detection in non-coastline profiles and funny shapes, achieving a success rate of 97.3%.

Abstract Given the rapidly growing scale and resource requirements of machine learning applications, the idea of building more efficient learning machines much closer to the laws of physics is an attractive proposition. One central question for identifying promising candidates for such neuromorphic platforms is whether not only inference but also training can exploit the physical dynamics. In this work, we show that it is possible to successfully train a system of coupled phase oscillators—one of the most widely investigated nonlinear dynamical systems with a multitude of physical implementations, comprising laser arrays, coupled mechanical limit cycles, superfluids, and exciton-polaritons. To this end, we apply the approach of equilibrium propagation, which permits to extract training gradients via a physical realization of backpropagation, based only on local interactions. The complex energy landscape of the XY/Kuramoto model leads to multistability, and we show how to address this challenge. Our study identifies coupled phase oscillators as a new general-purpose neuromorphic platform and opens the door towards future experimental implementations.

Abstract Electrochemical organic neuromorphic devices (ENODes) are rapidly developing as platforms for computing, automation, and biointerfacing. Resembling short- and long-term synaptic plasticity is a key characteristic in creating functional neuromorphic interfaces that showcase spiking activity and learning capabilities. This potentially enables ENODes to couple with biological systems, such as living neuronal cells and ultimately the brain. Before coupling ENODes with the brain, it is worth investigating the neuromorphic behavior of ENODes when they interface with electrolytes that have a consistency similar to brain tissue in mechanical properties, as this can affect the modulation of ion and neurotransmitter diffusion. Here, we present ENODEs based on different PEDOT:PSS formulations with various geometries interfacing with gel-electrolytes loaded with a neurotransmitter to mimic brain-chip interfacing. Short-term plasticity and neurotransmitter-mediated long-term plasticity have been characterized in contact with diverse gel electrolytes. We found that both the composition of the electrolyte and the PEDOT:PSS formulation used as gate and channel material play a crucial role in the diffusion and trapping of cations that ultimately modulate the conductance of the transistor channels. It was shown that paired pulse facilitation can be achieved in both devices, while long-term plasticity can be achieved with a tissue-like soft electrolyte, such as agarose gel electrolyte, on spin-coated ENODes. Our work on ENODe-gel coupling could pave the way for effective brain interfacing for computing and neuroelectronic applications.

Abstract Spiking neural networks (SNNs), inspired by the neural circuits of the brain, are promising in achieving high computational efficiency with biological fidelity. Nevertheless, it is quite difficult to optimize SNNs because the functional roles of their modelling components remain unclear. By designing and evaluating several variants of the classic model, we systematically investigate the functional roles of key modelling components, leakage, reset, and recurrence, in leaky integrate-and-fire (LIF) based SNNs. Through extensive experiments, we demonstrate how these components influence the accuracy, generalization, and robustness of SNNs. Specifically, we find that the leakage plays a crucial role in balancing memory retention and robustness, the reset mechanism is essential for uninterrupted temporal processing and computational efficiency, and the recurrence enriches the capability to model complex dynamics at a cost of robustness degradation. With these interesting observations, we provide optimization suggestions for enhancing the performance of SNNs in different scenarios. This work deepens the understanding of how SNNs work, which offers valuable guidance for the development of more effective and robust neuromorphic models.

Abstract Resistively switching electrochemical metallization memory cells are gaining huge interest since they are seen as promising candidates and basic building blocks for future computation-in-memory applications. However, especially filamentary-based memristive devices suffer from inherent variability, originating from their stochastic switching behavior. A variability-aware compact model of electrochemical metallization memory cells is presented in this study and verified by showing a fit to experimental data. It is an extension of the deterministic model. Since this extension consists of several different features allowing for a realistic variability-aware fit, it depicts a unique model comprising physics-based, stochastically and experimentally originating variabilities and reproduces them well. In addition, a physics-based model parameter study is executed, which enables a comprehensive view into the device physics and presents guidelines for the compact model fitting procedure.

Abstract Both biological and artificial neural networks inherently balance their performance with their operational cost, which characterizes their computational abilities. Typically, an efficient neuromorphic neural network is one that learns representations that reduce the redundancies and dimensionality of its input. For instance, in the case of sparse coding (SC), sparse representations derived from natural images yield representations that are heterogeneous, both in their sampling of input features and in the variance of those features. Here, we focused on this notion, and sought correlations between natural images’ structure, particularly oriented features, and their corresponding sparse codes. We show that representations of input features scattered across multiple levels of variance substantially improve the sparseness and resilience of sparse codes, at the cost of reconstruction performance. This echoes the structure of the model’s input, allowing to account for the heterogeneously aleatoric structures of natural images. We demonstrate that learning kernel from natural images produces heterogeneity by balancing between approximate and dense representations, which improves all reconstruction metrics. Using a parametrized control of the kernels’ heterogeneity of a convolutional SC algorithm, we show that heterogeneity emphasizes sparseness, while homogeneity improves representation granularity. In a broader context, this encoding strategy can serve as inputs to deep convolutional neural networks. We prove that such variance-encoded sparse image datasets enhance computational efficiency, emphasizing the benefits of kernel heterogeneity to leverage naturalistic and variant input structures and possible applications to improve the throughput of neuromorphic hardware.

Auto-encoders are capable of performing input reconstruction, denoising, and classification through an encoder-decoder structure. Spiking Auto-Encoders (SAEs) can utilize asynchronous sparse spikes to improve power efficiency and processing latency on neuromorphic hardware. In our work, we propose an efficient SAE trained using only Spike-Timing-Dependant Plasticity (STDP) learning. Our auto-encoder uses the Time-To-First-Spike (TTFS) encoding scheme and needs to update all synaptic weights only once per input, promoting both training and inference efficiency due to the extreme sparsity. We showcase robust reconstruction performance on the Modified National Institute of Standards and Technology (MNIST) and Fashion-MNIST datasets with significantly fewer spikes compared to state-of-the-art SAEs by 1–3 orders of magnitude. Moreover, we achieve robust noise reduction results on the MNIST dataset. When the same noisy inputs are used for classification, accuracy degradation is reduced by 30%–80% compared to prior works. It also exhibits classification accuracies comparable to previous STDP-based classifiers, while remaining competitive with other backpropagation-based spiking classifiers that require global learning through gradients and significantly more spikes for encoding and classification of MNIST/Fashion-MNIST inputs. The presented results demonstrate a promising pathway towards building efficient sparse spiking auto-encoders with local learning, making them highly suited for hardware integration.

Neuromorphic perception with event-based sensors, asynchronous hardware, and spiking neurons shows promise for real-time, energy-efficient inference in embedded systems. Brain-inspired computing aims to enable adaptation to changes at the edge with online learning. However, the parallel and distributed architectures of neuromorphic hardware based on co-localized compute and memory imposes locality constraints to the on-chip learning rules. We propose the event-based three-factor local plasticity (ETLP) rule that uses the pre-synaptic spike trace, the post-synaptic membrane voltage and a third factor in the form of projected labels with no error calculation, that also serve as update triggers. ETLP is applied to visual and auditory event-based pattern recognition using feedforward and recurrent spiking neural networks. Compared to back-propagation through time, eProp and DECOLLE, ETLP achieves competitive accuracy with lower computational complexity. We also show that when using local plasticity, threshold adaptation in spiking neurons and a recurrent topology are necessary to learn spatio-temporal patterns with a rich temporal structure. Finally, we provide a proof of concept hardware implementation of ETLP on FPGA to highlight the simplicity of its computational primitives and how they can be mapped into neuromorphic hardware for online learning with real-time interaction and low energy consumption.