Spiking Neural Networks (SNNs) are one of the most promising bio-inspired neural networks models and have drawn increasing attention in recent years. The event-driven communication mechanism of SNNs allows for sparse and theoretically low-power operations on dedicated neuromorphic hardware. However, the binary nature of instantaneous spikes also leads to considerable information loss in SNNs, resulting in performance degradation. To address this issue, multi-level spiking neuron models have been proposed to provide both low-quantization error and minimal inference latency while approaching the performance of full precision Artificial Neural Networks (ANNs). In this paper, we provide an in depth analysis of the different trade-offs between quantization, generalization performance, and energy efficiency between binary SNNs, multi-level SNNs and ANNs for two different applications scenarios: image classification and image denoising. Our goal is to determine analytical thresholds on SNNs activity to understand when SNNs can be more energy efficient than ANNs but also to explain why SNNs cannot provide energy gains in other scenarios. Our analysis is based on an hardware execution model, which allows us to take into account the energy costs related to synaptic operations as well as memory accesses. Our experimental results show that multi-level spiking neurons provide better information compression, allowing therefore a reduction in latency without performance loss for classification tasks. When compared to binary SNNs on image classification scenarios, multi-level SNNs indeed allow reducing by 2 to 3 times the energy consumption depending on the number of quantization intervals. On neuromorphic data, multi-level SNNs allows us to drastically reduce the inference latency to 1 timestep, which corresponds to a compression factor of 10 compared to previously published results on binary SNNs. However, for the image denoising task we observe that neither the binary nor the multi-level SNNs can match the ANNs energy efficiency and performance. Based on a hardware-aware execution model of a neuromorphic SNN accelerator, we then provide analytical upper-bounds on SNNs activity and energy-efficiency. These theoretical bounds are then compared against the experimental results and allows us to understand, from a system level point of view, the relation between energy efficiency, quantization schemes and finally the performance of the SNNs.

Human vision achieves remarkable perceptual performance while operating under strict metabolic constraints. A key ingredient is the selective attention mechanism, driven by rapid saccadic eye movements that constantly reposition the high-resolution fovea onto task-relevant locations, unlike conventional AI systems that process entire images with equal emphasis. Our work aims to draw inspiration from the human visual system to create smarter, more efficient image processing models. Using DINO, a self-supervised Vision Transformer that produces attention maps strikingly similar to human gaze patterns, we explore a saccade inspired method to focus the processing of information on key regions in visual space. To do so, we use the ImageNet dataset in a standard classification task and measure how each successive saccade affects the model's class scores. This selective-processing strategy preserves most of the full-image classification performance and can even outperform it in certain cases. By benchmarking against established saliency models built for human gaze prediction, we demonstrate that DINO provides superior fixation guidance for selecting informative regions. These findings highlight Vision Transformer attention as a promising basis for biologically inspired active vision and open new directions for efficient, neuromorphic visual processing.

Energy efficiency remains a fundamental challenge in Wireless Sensor Networks (WSNs) due to the limited power supply of sensor nodes, which directly constrains network life-time. Prediction-based approaches introduce embedded intelligence to reduce data transmissions and improve energy efficiency. However, uneven energy consumption among nodes can limit overall network lifetime. This paper presents a dynamic node role rotation mechanism that combines clustering strategies with embedded data prediction to achieve balanced energy consumption while preserving prediction performance. The approach is deployed on a real LoRa-based testbed integrating lightweight neural networks for on-device temperature forecasting. Experimental results show that the proposed method balances per-node energy consumption, maintains prediction accuracy below 2% MAPE, and extends network lifetime up to 2.5 years depending on the clustering interval. Compared to static prediction schemes, the system achieves 60–62% transmission reduction with limited overhead. These results validate the proposed approach as a hardware-feasible and energy-aware solution for predictive WSNs in real-world conditions.

In the quest for energy-efficient neural network architectures, Spiking Neural Networks (SNNs) have emerged as a promising alternative to traditional Artificial Neural Networks (ANNs). However, the inherent SNN training complexity and the high spiking activity often limit their practical application. This paper explores hybrid neural networks (HNNs) that integrate ANN and SNN layers to optimize the trade-off between performance and energy efficiency. Our study systematically explores nearly exhaustive configurations of ANNs and SNNs on two datasets (CIFAR-10 and GSC V2) and five architectures (VGG16, VGG11, ResNet18, ResNet8, and a 4-layer CNN). Depending on the proportion of SNN and ANN layers, our hybrid models can reduce up to 2.45 times the energy consumption while maintaining the same ANN baseline accuracy, and even exceeding it for some models. These results highlight the potential of hybrid models in the development of energy-efficient neural networks and paves the way for future hardware implementations.

With the growing interest in on On-orbit servicing (OOS) and Active Debris Removal (ADR) missions, spacecraft poses estimation algorithms are being developed using deep learning to improve the precision of this complex task and find the most efficient solution. With the advances of bio-inspired low-power solutions, such a spiking neural networks and event-based processing and cameras, and their recent work for space applications, we propose to investigate the feasibility of a fully event-based solution to improve event-based pose estimation for spacecraft. In this paper, we address the first event-based dataset SEENIC with real event frames captured by an event-based camera on a testbed. We show the methods and results of the first event-based solution for this use case, where our small spiking end-to-end network (S2E2) solution achieves interesting results over 21cm position error and 14degree rotation error, which is the first step towards fully event-based processing for embedded spacecraft pose estimation.

Despite their remarkable capabilities, state-of-the-art AI models rely on deeply parameterized architectures that require extensive labeled datasets and multiple training epochs, revealing significant inefficiencies compared to biological intelligence in terms of data utilization. While self-supervised pre-training techniques have advanced the field, these approaches still demand considerable amount of data to achieve high classification accuracy. Biological neural systems, in contrast, demonstrate remarkable efficiency through local learning rules and multi-modal integration. Drawing inspiration from these principles, we present BrAMA (Brain-inspired Architecture for Multimodal Association), a novel framework that constructs meaningful data representations by associating symbolic representation of multimodal signals, drawing inspiration from cognitivist principles. Our key contributions include novel Hebbian local learning rules and architectural improvements to existing brain-inspired approaches. Through parameterized Hebbian connections between self-organizing maps, our enhanced learning mechanisms and its semi-supervision learning capabilities, BrAMA achieves superior accuracy compared to state-of-the-art approaches while requiring significantly fewer training examples with a single epoch.

In this paper we propose a Hardware-Aware Neural Architecture Search method focused on memory constrained neural networks accelerators. Our solution is based on the observation that minimizing data transfers to and from the accelerator while maximizing the number of operations can lead to efficient network design. We introduce an optimized search space integrated into a NAS algorithm providing networks that improve the latency while minimizing the energy consumption. Those networks also achieve a better accuracy when compared to manually designed mobile architectures. Our experimental results, performed on Imagenet, show that the accuracy can be improved by 1.3% for a specific range of latencies compared to different versions of MobileNetV2 rescaled by hand.

Spiking Neural Networks (SNNs) are one of the most promising bio-inspired neural networks models and have drawn increasing attention in recent years. The event-driven communication mechanism of SNNs allows for sparse and theoretically low-power operations on dedicated neuromorphic hardware. However, the binary nature of instantaneous spikes also leads to considerable information loss in SNNs, resulting in accuracy degradation. To address this issue, we propose a multi-level spiking neuron model able to provide both low-quantization error and minimal inference latency while approaching the performance of full precision Artificial Neural Networks (ANNs). Experimental results with popular network architectures and datasets, show that multi-level spiking neurons provide better information compression, allowing therefore a reduction in latency without performance loss. When compared to binary SNNs on image classification scenarios, multi-level SNNs indeed allow reducing by 2 to 3 times the energy consumption depending on the number of quantization intervals. On neuromorphic data, our approach allows us to drastically reduce the inference latency to 1 timestep, which corresponds to a compression factor of 10 compared to previously published results. At the architectural level, we propose a new residual architecture that we call Sparse-ResNet. Through a careful analysis of the spikes propagation in residual connections we highlight a spike avalanche effect, that affects most spiking residual architectures. Using our Sparse-ResNet architecture, we can provide state-of-the-art accuracy results in image classification while reducing by more than 20

Achieving optimal semantic segmentation with frame-based vision sensors poses significant challenges for real-time systems like UAVs and self-driving cars, which require rapid and precise processing. Traditional frame-based methods often struggle to balance latency, accuracy, and energy efficiency. To address these challenges, we leverage event streams from event-based cameras-bio-inspired sensors that trigger events in response to changes in the scene. Specifically, we analyze the number of events triggered between successive frames, with a high number indicating significant changes and a low number indicating minimal changes. We exploit this event information to solve the semantic segmentation task by employing a Spiking Neural Network (SNN), a bio-inspired computing paradigm known for its low energy consumption. Our experiments on the DSEC dataset show that our approach significantly reduces latency with only a limited drop in accuracy. Additionally, by using SNNs, we achieve low power consumption, making our method suitable for energy-constrained real-time applications. To the best of our knowledge, our approach is the first to effectively balance reduced latency, minimal accuracy loss, and energy efficiency using events stream to enhance semantic segmentation in dynamic and resource-limited environments.

The limited energy resources in wireless sensor networks (WSNs) is a critical challenge to long-term operation, particularly in remote areas where recharging batteries is not easily feasible. This article proposes an energy-efficient approach based on convolutional neural networks (CNNs) for on-device data prediction at the cluster head (CH) level. By substituting certain data transmissions with predicted sensor values, this method minimizes radio communication, the most energy-intensive task in sensor nodes (SNs), thereby extending the network's lifetime. We evaluate the performance of this approach by analyzing two neural network structures and operational modes in an environmental monitoring context. Simulations with real-world sensor data show that the CNN model accurately predicts sensor values with minimal error. In addition, the solution is validated through real-world deployment in a WSN, with real-time monitoring, energy-efficient communication, and frame management. A detailed energy profile and consumption assessment is provided, alongside an estimation of network autonomy. Results indicate a reduction in data transmissions by over 50%, enabling approximate to 34.3% of energy savings and approximate to 57.0% improvement in network autonomy without compromising accuracy. This solution offers a practical approach for sustainable, long-term WSN deployments.

Interest in spiking neural networks (SNNs) has been growing steadily, promising an energy efficient alternative to formal neural networks (FNNs), commonly known as artificial neural networks (ANNs). Despite increasing interest, especially for Edge applications, these eventdriven neural networks suffered from their difficulty to be trained compared to FNNs. To alleviate this problem, a number of innovative methods have been developed to provide performance more or less equivalent to that of FNNs. However, the spiking activity of a network during inference is usually not considered. While SNNs may usually have performance comparable to that of FNNs, it is often at the cost of an increase of the network's activity, thus limiting the benefit of using them as a more energy efficient solution. In this paper, we propose to leverage Knowledge Distillation (KD) for SNNs training with surrogate gradient descent in order to optimize the trade-off between performance and spiking activity. Then, after understanding why KD led to sparsity increase, we also explored Activations regularization and proposed a novel method with Logits Regularization. These approaches, validated on several datasets, clearly show a reduction in network spiking activity (- 26.73% on GSC and - 14.32% on CIFAR-10) while preserving accuracy.

Wireless sensor nodes are increasingly deployed to monitor and collect data in remote and inaccessible locations. These nodes transmit the collected information to base stations for further processing. However, one of the major challenges they face is the limited energy resource, which significantly restricts their operational lifetime and, consequently, the reliability and efficiency of the network. Addressing this critical issue, this paper presents the use of convolutional neural networks in a wireless sensor network at a node-level to enhance autonomy and extend the network lifetime by replacing certain data communications with sensor data prediction. We evaluate the performance of the proposed method by examining different neural network struc-tures and their operating modes in the context of environmental monitoring. Through many simulations using real-world sensor data, we demonstrate that, the neural network is well capable of learning to predict data with an acceptable margin of error reducing the need for data transmissions by at least 50 %. This reduction in communication leads to energy savings and extends the network's operational lifetime. Moreover, the performances in terms of memory footprint, and energy consumption on board make it possible to extend the sleep mode of the sensor nodes to increase the wireless sensor network lifetime.

This study addresses the classification of objects using their electromagnetic signatures with Convolutional Neural Networks (CNNs) trained on noiseless data. The singularity expansion method (SEM) was applied to establish a compact model that accurately represents the ultra-wideband scattered field of an object, independently of its orientation and observation angle. To perform the classification, we used a CNN associated with a noise-robust SEM technique to classify different objects based on their characteristic parameters. To validate this approach, we compared the performance of the classifier with and without SEM pre-processing of the scattered field for different noise levels and for object sizes not present in the training set. Moreover, we propose a procedure that determines the direction of the receiving antenna and orientation of an object based on the residues associated with each complex natural resonance. This classification procedure using pre-processed SEM data is promising, especially when generalizing to object sizes not included in the training set.

Addressing key challenges in wireless sensor networks (WSNs), such as network lifetime and energy balance, this letter intro- duces the semi-decentralized prediction method (SDPM) for energy- efficient wireless sensor networks. This approach enhances energy efficiency by combining clustering principles with data prediction for smart cluster-head (CH) selection. SDPM facilitates the periodic election of an effective CH from among the cluster-nodes, who then predicts data for the nodes within the cluster, thereby reducing transmission and conserving energy. Our findings demonstrate SDPM's significant impact on reducing energy consumption, promising for real-world WSNs to achieve longer network lifetime and better energy management.

The complexity of event-based object detection (OD) poses considerable challenges. Spiking Neural Networks (SNNs) show promising results and pave the way for efficient event-based OD. Despite this success, the path to efficient SNNs on embedded devices remains a challenge. This is due to the size of the networks required to accomplish the task and the ability of devices to take advantage of SNNs benefits. Even when "edge" devices are considered, they typically use embedded GPUs that consume tens of watts. In response to these challenges, our research introduces an embedded neuromorphic testbench that utilizes the SPiking Low-power Event-based ArchiTecture (SPLEAT) accelerator. Using an extended version of the Qualia framework, we can train, evaluate, quantize, and deploy spiking neural networks on an FPGA implementation of SPLEAT. We used this testbench to load a state-of-the-art SNN solution, estimate the performance loss associated with deploying the network on dedicated hardware, and run real-world event-based OD on neuromorphic hardware specifically designed for low-power spiking neural networks. Remarkably, our embedded spiking solution, which includes a model with 1.08 million parameters, operates efficiently with 490 mJ per prediction.

Spiking Neural Networks are a type of neural networks where neurons communicate using only spikes. They are often presented as a low-power alternative to classical neural networks, but few works have proven these claims to be true. In this work, we present a metric to estimate the energy consumption of SNNs independently of a specific hardware. We then apply this metric on SNNs processing three different data types (static, dynamic and event-based) representative of real-world applications. As a result, all of our SNNs are 6 to 8 times more efficient than their FNN counterparts.

In this paper, a miniature communicating terminal with Embedded artificial Intelligence capabilities for Internet of Things experiment is presented. This platform can communicate to terrestrial and satellite gateways using LoRaWan protocol. The terminal integrates multiple sensors and Machine Learning algorithms that can be executed on the low-power Microcontroller to detect patterns and provide meta-data to the network. A model of this terminal based on OMNET++ is presented to simulate a large scale deployment.