Marcello Coppola’s research while affiliated with STMicroelectronics and other places

... The definition of our model is mostly empirical, as the current pre-defined neural networks are mainly intended for very complex problems, and these large models are simply not appropriate for inference on small electronic devices. More details about our particular defined model can be found in [9]. ...

Reference:

Inside the AI Accelerators: From High Performance to Energy Efficiency

... However, a failure in the cloud narrows down to limitations and latency in connectivity and security. Edge computing governances closer data to the substantial physical areas, avoiding the disvalued servitude of centralized servers [3,4]. Advanced AI models, in particular, are preferred for realtime industrial operations that burden with very less latency and very much happening privacy preservation. ...

Reference:

Generative AI for Edge-Based Predictive Maintenance in Smart Factories

... While these techniques provide a strong foundation, studies such as Mohimont et al. [47] indicate an ongoing gap, noting that only 19% of grapes within a cluster are typically visible, which highlights the need for enhanced occlusion adjustment techniques. Additionally, Zabawa et al. [6] report frequent under-segmentation in dense grape clusters, pointing to the difficulties in achieving accurate yield estimates without advanced occlusion handling. ...

Reference:

GCNet: A Deep Learning Framework for Enhanced Grape Cluster Segmentation and Yield Estimation Incorporating Occluded Grape Detection with a Correction Factor for Indoor Experimentation

... Their optimized model (RTSD-Net) with TensorRT achieved about 25.20 fps and performed 15% faster than the original YOLOv4-tiny model on Jetson Nano without significant loss of accuracy. Other promising applications of Edge-AI are air temperature forecasting [109], environment monitoring [110], autonomous navigation systems [111] and so on. ...

Reference:

Sensing and Automation Technologies for Ornamental Nursery Crop Production: Current Status and Future Prospects

... Artificial Intelligence (AI) is a computational method that tried to imitate, in a simplified form, human intelligence to solve engineering problems that have escaped normal computer techniques [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20]. According to Russell and Norvig [2], it has been argued that AI can be linked to systems that able to think like people, behave like humans whereas AI able to think logically and act rationally. ...

Reference:

Dimensions Affecting Consumer Acceptance towards Artificial Intelligence (AI) Service in the Food and Beverage Industry in Klang Valley

... Authors argue that self-supervised training methods improve classification performance when datasets with several unlabelled data are used, which is the case of the WM-811K dataset. [13] presents experimentations with simplified AlexNet, MobiletNetV1, and VGG, so as to limit the number of parameters they require. This latter architecture leads to the best accuracy while requiring the least parameters. ...

Reference:

Power Optimized Wafermap Classification for Semiconductor Process Monitoring

... Although the number of parameters has considerably reduced, these architectures are still too large to be implemented in low-power hardware devices. Andrade et al. (2021) presents experimentations with simplified AlexNet, MobiletNetV1, and VGG, so as to limit the number of parameters they require. This latter architecture leads to the best accuracy while requiring the least parameters. ...

Reference:

Accurate and energy efficient ad-hoc neural network for wafer map classification

... In the logistics sector, it is easy to see how optimizing such a tool can lead to efficient warehouse management-thanks to its ability to predict demand and customer behavior, it can optimize route selection or warehouse management (Vermesan et al., 2022). ...

Reference:

Framework of IoT, Blockchain, Digital Twins, and Artificial Intelligence Solutions in Support of the Digital Business Transformation of Logistics 5.0

... Dalam beberapa tahun terakhir, kemajuan teknologi pemantauan berbasis Supervisory Control and Data Acquisition (SCADA) telah banyak diterapkan Muhammad Irfan / Sistem Early Warning Deteksi Malfungsi pada Motor Listrik 3 Fasa Menggunakan Mikrokontroler Berbasis Lora Terintegrasi Scada untuk meningkatkan keandalan dan efisiensi pemeliharaan motor listrik industri (Chukwunweike et al., 2024;Stankov, 2024). SCADA memungkinkan pemantauan kondisi operasional secara real-time, di mana parameter seperti tegangan, arus, suhu, dan kecepatan putaran dapat dikontrol dari pusat kendali (Kaittan & Mohammed, 2024;Vermesan et al., 2022). Dengan sistem ini, deteksi dini terhadap potensi kerusakan motor menjadi lebih akurat, sehingga membantu mengurangi risiko downtime serta meningkatkan efisiensi pemeliharaan (Akbar et al., 2023;Kaittan & Mohammed, 2024). ...

Reference:

Sistem Early Warning Deteksi Malfungsi pada Motor Listrik 3 Fasa Menggunakan Mikrokontroler Berbasis Lora Terintegrasi Scada

... It also makes the entire system more resilient to attacks and disruptions. Finally, the system is designed to comply with all relevant data protection and cybersecurity regulations, ensuring it meets the highest standards for security and privacy [5][6][7]. ...

Reference:

ERATOSTHENES Project Integrated Solution