In today's world of computer vision AI, there is a significant challenge of running AI models on the limited compute power of edge device chipsets, as they are typically too large, process heavy and consume significant power. Furthermore, there is the rising costs of IT infrastructure (e.g. cloud) to support AI models, which is not green-tech friendly, as well as the costly process of spending months to manually optimizing AI models, while trying to maintain accuracy.
To break down these barriers, Deeplite created its unique software platform called Deeplite Neutrino, which is an intelligent optimization engine for Deep Neural Networks (DNNs) that can be deployed on constrained edge-based products and more efficiently within cloud servers. Deeplite Neutrino helps AI experts automatically optimize high-performance DNN models to satisfy target computation constraints. The engine inputs large, trained DNN models to produce smaller, highly-efficient models while preserving the accuracy. This brings significant value to manufacturing sectors which typically have a variety of edge devices, such as cameras for automated optical inspection and robotics arms for rapid product sorting and packaging, that are powered by different processors (e.g. CPU, MCU and RISC-V). In order to run the AI with improved inference and operational efficiency, the DNN models need to be optimized, with enhanced throughput, while maintaining accuracy.
Deeplite Neutrino provides user-friendly software to generate compact and efficient AI models
Deeplite, the Canadian start-up, provides a software-based AI-Driven Optimizer to make DNNs faster, smaller and energy-efficient to run on edge device and more cost effectively in a cloud environment. Today, there is a significant challenge of running machine vision and perception AI applications on the limited compute power of edge devices and their associated chipsets.
The company's software product named Deeplite Neutrino provides a unique on-premise tool for AI engineering and data science teams, which accelerates their respective machine learning operations workflow for deep learning models such as image classification, object detection and segmentation. With an easy-to-use optimization solution, OEMs and chip manufacturers can leverage Deeplite Neutrino to rapidly produce highly compact, high performance models to deploy in production on a multitude of hardware platforms. The automated functionality of Deeplite Neutrino can be as simple as the following workflow:
*Input the pretrained model, associated dataset and specific performance KPIs, e.g. desired model accuracy and size.
*Press start and watch Neutrino output a highly optimized model with acceptable accuracy.
*Deploy your optimized model to your target hardware devices via the associated compiler and inference engine.
Showcasing of automatic optical inspection and robotic fast-time-to-classifying systems
The current popular Industrial 4.0 use cases powered by Deeplite Neutrino are computer vision-based for a variety of edge devices such as cameras, robotics and drones. One of the most popular AI applications is automatic optical inspection which will accelerate quality control inspection during the manufacturing process. The need to have faster inference at the camera to identify faults in the production line will save significant operational costs, increase customer orders and enhance customer satisfaction. Customers will utilize Deeplite Neutrino to drive efficiencies for compressing and deploying object detection and segmentation AI models for the quality control application on low power devices e.g. cameras. The same applies to fast-time-to-classifying images via robots for rapid identification of sorting packages on a manufacturing conveyor belt.
Deeplite is currently working with customers in North America, Taiwan, Japan and select countries in Europe. Thanks to the close collaboration with the Canadian Trade Office in Taipei, Deeplite is focused on engaging with Taiwan customers. As a world leader in hardware manufacturing and a strong foundation for intelligent technology development, particularly in ICT and semiconductors, Taiwan is a great fit for Deeplite's market focus. Furthermore, Taiwan is seeking to develop a world-leading AI on Device solution, which incorporates both hardware and software collaboration. This aligns extremely well with Deeplite's Neutrino software value to chip manufacturers, edge device manufacturers (OEMs) and within manufacturing processes. Deeplite looks forward to building a strong market presence in Taiwan.
Charles Marsh, CCO of Deeplite
DIGITIMES' editorial team was not involved in the creation or production of this content. Companies looking to contribute commercial news or press releases are welcome to contact us.
