A5036123913- Advanced Neural Network Applications
- Neural Networks and Applications
- Anomaly Detection Techniques and Applications
- Context-Aware Activity Recognition Systems
- Advanced Vision and Imaging
- Robotics and Sensor-Based Localization
- Advanced Image and Video Retrieval Techniques
- CCD and CMOS Imaging Sensors
- Video Coding and Compression Technologies
- Advanced Memory and Neural Computing
- Neural Networks and Reservoir Computing
- Advanced Data Compression Techniques
- Machine Learning and ELM
- Sensor Technology and Measurement Systems
- IoT and Edge/Fog Computing
- Speech and Audio Processing
- Advanced Battery Technologies Research
- Inertial Sensor and Navigation
- Music and Audio Processing
- Parallel Computing and Optimization Techniques
- Computer Graphics and Visualization Techniques
- Indoor and Outdoor Localization Technologies
- Advanced MEMS and NEMS Technologies
- Adversarial Robustness in Machine Learning
- Non-Invasive Vital Sign Monitoring
STMicroelectronics (Italy)
2016-2025
STMicroelectronics (Germany)
2022-2023
STMicroelectronics (Czechia)
2012-2023
Institute of Electrical and Electronics Engineers
2023
Università Cattolica del Sacro Cuore
2023
STMicroelectronics (Switzerland)
2000-2021
University of Pavia
2018
University of Cyprus
2017
Polytechnic University of Turin
2012
Mayo Clinic
2012
Recent advancements in ultra-low-power machine learning (TinyML) hardware promises to unlock an entirely new class of smart applications. However, continued progress is limited by the lack a widely accepted benchmark for these systems. Benchmarking allows us measure and thereby systematically compare, evaluate, improve performance systems therefore fundamental field reaching maturity. In this position paper, we present current landscape TinyML discuss challenges direction towards developing...
Identifying diseases from images of plant leaves is one the most important research areas in precision agriculture. The aim this paper to propose an image detector embedding a resource constrained convolutional neural network (CNN) implemented low cost, power platform, named OpenMV Cam H7 Plus, perform real-time classification disease. CNN so obtained has been trained on two specific datasets for detection, ESCA-dataset and PlantVillage-augmented dataset, low-power, low-cost Python...
Lithium-ion (Li-Ion) batteries are rechargeable which can maximize battery lifespan thanks to their chemical abilities, at the same time increasing power energy density. For these reasons, Li-Ion have earned considerable popularity, and they widely used both in mobile computing devices (e.g. smartphones smartwatches) automotive systems hybrid electric vehicles). A fundamental parameter for health monitoring is State of Health (SoH), computed from maximum releasable capacity, represents...
Esca is one of the most common disease that can severely damage grapevine. This disease, if not properly treated in time, cause vegetative stress or death attacked plant, with consequence losses production as well a rising risk propagation to closer grapevines. Nowadays, detection carried out manually through visual surveys usually done by agronomists, requiring enormous amount time. Recently, image processing, computer vision and machine learning methods have been widely adopted for plant...
Unintentional falls are the leading cause of fatal injuries and nonfatal trauma among older adults. An automated monitoring system that detects occurring issues remote notifications will prove very valuable for improving level care could be provided to people at higher risk. The work presented focuses on design embedded software wearable devices connected in wireless mode a system. implementation recurrent neural networks (RNNs) architectures micro controller units (MCU) fall detection with...
The deployment of neural networks on resource-constrained micro-controllers has gained momentum, driving many advancements in Tiny Neural Networks. This paper introduces a tiny feed-forward network, TinyFC, integrated into the Field-Oriented Control (FOC) Permanent Magnet Synchronous Motors (PMSMs). Proportional-Integral (PI) controllers are widely used FOC for their simplicity, although limitations handling nonlinear dynamics hinder precision. To address this issue, lightweight 1,400...
The integration of large language models (LLMs) on low-power edge devices such as Raspberry Pi, known (ELMs), has introduced opportunities for more personalized, secure, and low-latency intelligence that is accessible to all. However, the resource constraints inherent in lack robust ethical safeguards raise significant concerns about fairness, accountability, transparency model output generation. This paper conducts a comparative analysis text-based bias across deployments edge, cloud,...
Esca is one of the most common grape leaf diseases that seriously affect yield, causing a loss global production in range 20%–40%. Therefore, timely and effective identification disease could help to develop an early treatment approach control its spread while reducing economic losses. For this purpose use computer vision machine learning techniques for recognizing plant have been extensively studied recent years. The aim paper propose image detector based on high-performance convolutional...
In this work, a new custom design of an anomaly detection and classification system is proposed.It composed convolutional Auto-Encoder (AE) hardware to perform which cooperates with mixed HW/SW Convolutional Neural Network (CNN) the detected anomalies.The AE features partial binarization, so that weights are binarized while activations, associated some selected layers, nonbinarized.This has been necessary meet severe area energy constraints allow it be integrated on same die as MEMS sensors...
Large Language Models achieve state of art performances on a broad variety Natural Processing tasks. In the pervasive IoT era, their deployment edge devices is more compelling than ever. However, gigantic model footprint has hindered on-device learning applications which enable AI models to continuously learn and adapt changes over time. Back-propagation, in use by majority deep frameworks, computationally intensive requires storing intermediate activations into memory cope with model’s...
Cyber-Physical Systems (CPSs) represent the technological asset of Industry 4.0. This paper introduces a novel generation CPSs, called Intelligent able to integrate intelligent functionalities such as fault prediction, autonomous behavior and self-adaptation directly at CPS units. Such will increase autonomy, reduce required bandwidth energy-efficiency CPSs making them fully address challenging needs increasing performance in 4.0 (as well other relevant scenarios, e.g., smart...
Collecting vast amount of data and performing complex calculations to feed modern Numerical Weather Prediction (NWP) algorithms require centralize intelligence into some the most powerful energy resource hungry supercomputers in world. This is due chaotic nature atmosphere which interpretation virtually unlimited computing storage resources. With Machine Learning (ML) techniques, a statistical approach can be designed order perform weather forecasting activity. Moreover, recently growing...
In this paper, an energy efficient HW accelerator for AI edge-computing in Human Activity Recognition is proposed. The system processes samples from a tri-axial accelerometer and classifies the human activities by using novel Hybrid Neural Network (HNN) topology, which has been designed to reduce computational complexity of while preserving its accuracy. design improves characteristics HNN means architecture that aimed allocated physical resources memory accesses. While accuracy measured on...
Human Activity Recognition requires very high accuracy to be effectively employed into practical applications, ranging from elderly care microsurgical devices. The highest accuracies are achieved by Deep Learning models, but these not easily deployable in handheld or wearable devices with constrained resources. We therefore present a new HAR system suitable for compact FPGA implementation. A Binarized Neural Network (BNN) architecture achieves the classification based on data single...
On-device learning challenges the restricted memory and computation requirements imposed by its deployment on tiny devices. Current training algorithms are based backpropagation which requires storing intermediate activations to compute backward pass update weights into memory. Recently "Forward-only algorithms" have been proposed as biologically plausible alternatives backpropagation. At same time, they remove need store potentially lower power consumption due read write operations, thus,...
The challenge of deploying neural network (NN) learning workloads on ultralow power tiny devices has recently attracted several machine researchers the Tiny community. A typical on-device session processes real-time streams data acquired by heterogeneous sensors. In such a context, this letter proposes Restricted Coulomb energy (TinyRCE), forward-only approach based hyperspherical classifier, which can be deployed microcontrollers and potentially integrated into sensor package. TinyRCE is...
A custom Human Activity Recognition system is presented based on the resource-constrained Hardware (HW) implementation of a new partially binarized Hybrid Neural Network. The processes data in real-time from single tri-axial accelerometer, and able to classify between 5 different human activities with an accuracy 97.5% when Output Data Rate sensor set 25 Hz. Network (HNN) has binary weights (i.e. constrained +1 or -1) but uses non-binarized activations for some layers. This, conjunction...
MEMS pressure sensors are widely used in several application fields, such as industrial, medical, automotive, etc, where they required to be increasingly accurate and reliable. However, these very sensitive mechanical temperature variations. For example, the soldering process, which involves significant thermal stress, causes drift sensor accuracy. This article introduces a digital circuit implementing tiny neural network able compensate for measurement real time. The is capable of...