Machine Learning
Machine Learning and Deep Learning for embedded, edge systems is a very interesting technology. No need for internet connections – enabling decisions to be made right at the sensor which in turn reduces latency and downtime. We have strong competence in Machine Learning on the edge and our services covers any part of the development phase, from feasibility studies and algorithm development to accelerated implementations in hardware such as FPGA, GPU or CPU.
Edge Deep Learning
As the hardware capabilities have grown, so has the size of the neural networks. But in recent years there has been a shift in focus to make the networks as small as possible. This means that the power of Deep Learning has become available for mobile phones, IoT devices and other embedded systems.
Synective Labs take part in the edge deep learning research by combining our knowledge of FPGA development with the latest deep learning architectures. We have also worked with deep learning on embedded CPUs such as the Raspberry Pi platform. Edge deep learning requires algorithmic insights as well as the know-how of how to implement, and Synective stands firmly with one foot in each world.
Real-Time Sensor and Image Analysis with Edge Machine Learning
At Synective Labs, we understand the need for real-time and accurate analysis of sensor and image data. Running machine learning models on the cloud can result in slow response times, making it difficult to perform real-time analysis. To address this challenge, we offer advanced machine learning techniques tailored for edge computing, closer to where the sensor data is being collected and reducing cloud computation costs.
Our expertise in using tools such as PyTorch, TensorFlow and Vitis AI allows us to develop models that can run on the edge, making the analysis of sensor and image data faster and more efficient. By leveraging IoT-to-Cloud communications, we can automate data collection and model training in the cloud, enabling real-time analysis and faster response times
Data collection and processing
Data collection is the first crucial step to any machine learning project. We leverage IoT-to-Cloud communications to automate data collection, making it easier and more efficient.
Optimisation and deployment to edge device
Our solutions utilize cutting-edge tools such as TensorFlow Lite Micro to optimize our trained models for deployment on microcontrollers, while utilizing FPGA-vendor toolchains or hand-coded RTL for FPGA implementations.
Edge inference
The optimized model is deployed to the edge device, where it can run in real-time and provide accurate and efficient analysis of electronic device wear, sensor, and image data.
Contact us to learn how our machine learning techniques can be tailored to improve real-time sensor and image analysis for your application.
We offer services:
Edge Convolutional Neural Networks
Deep Learning on FPGA
Feasibility Studies and Consultation
Algorithm Development
Hardware Acceleration of Machine Learning Algorithms
Training Data Collection
Get in touch
Deep Learning on FPGA
Synective has developed a demo system to showcase the power of embedded FPGA Deep Learning. It was able to classify hand-written digits using nothing but a tiny, low-end FPGA and a camera.
Finger Recognition on Raspberry Pi
Synective has designed a CNN that is optimized with respect to memory footprint and inference speed. The network is implemented and runs on the ARM processor of the Raspberry Pi.
Embedded CNN Master Thesis 2019
Synective continually take on thesis workers that improve our embedded Deep Learning system.
High Performance Embedded
Synective Labs provides high performance embedded solution to a wide range of customers, with a focus on computational and data intense problems.
FPGA/ASIC design
Synective Labs is a leading design house and consulting company within FPGA and ASIC design in the Nordic region.
IP block
The Synective CAN 2.0/CAN FD IP core implements a complete CAN controller for integration into FPGAs and ASICs.
Image and Signal Processing
Synective Labs specializes in algorithm development for image and signal processing, with many years of experience from advanced camera, radar and sensor systems.
Server Based Acceleration
Synective Labs work with some of the world’s leading providers of hardware for server-based acceleration, in order to reach state of the art results.
Machine Learning
Synective Labs is actively researching and implementing machine learning with deep neural networks designed to fit on small edge FPGAs and other low power embedded systems.