Earlier this month, we participated in the annual Accelerating AI Workshop that highlighted the challenges and opportunities of AI acceleration from the cloud to the edge. Every year, this event brings together experts from across the industry and academia to share the latest trends and innovations, identify new challenges and opportunities, explore collaborations, and identify common infrastructure requirements. The full workshop recording can be viewed here.
The workshop is hosted by our partner CMC Microsystems, a not-for-profit organization accelerating research and innovation in Canada with a mission to lower barriers to technology adoption by creating and sharing platform technologies. This year’s workshop was very insightful and technical, and brought together many speakers from Deeplite, as well as other innovative companies and academia. They shared their latest achievements and innovations in the field of AI and machine learning algorithms, software, and hardware from the cloud to deeply embedded machine learning systems. CMC Microsystems also manages Canada’s National Design Network, which is a country-wide collaboration between 65 universities and colleges to connect 10,000 academic participants with 1000 companies to design, fabricate and test micro and nanosystem prototypes.
This community atmosphere is essential for the continued development and application of AI, and creates a place where engineers, developers and others in the space can learn from each other and exchange ideas. Deeplite’s CEO, Nick Romano, spoke during the workshop about our platform, products and how members of CMC Microsystems can take advantage of our optimization tools right now.
Making AI accessible in the real world
Nick took attendees through an overview of what Deeplite does, and how our optimization products work. We’ve learned that productizing computer vision on AI at the edge is difficult so Deeplite has made its solutions easy to use so more organizations can get products out of research labs and take advantage of AI at the edge.
Right now, Deeplite Neutrino is available on the CMC Microsystems platform. Neutrino is an automated optimization engine, which allows users to optimize their AI models at architectural levels, as well as quantize applications down to low bit precisions. Additionally, our Deeplite Runtime (Deeplite RT) engine is designed to run models quickly on low-cost, low-power hardware. Lastly, our newest solution is DeepliteZoo, which is being populated with high-performance models based on specific use cases, such as person-detection.
One of the most important pieces of the Deeplite suite is that our solutions don’t compromise accuracy. This is why they work so well in edge devices – they use up less compute power so not only are they energy-efficient, but even the largest, slowest AI models can quickly be optimized so they are more accessible to use.
Deeplite in action
During the presentation, Nick gave a quick demo of how users can input their large AI models into Deeplite’s optimization engine. It’s a very command-driven process – the user sets certain parameters for the job (such as accuracy needs) as it’s inputted. Then, there’s a one-line command that will launch the process and run the optimization. It’s a very automated, hands-off process in which the work is done behind the scenes. The results of the optimization are then returned to the user.
![Deeplite's optimization command line](https://blog.deeplite.ai/hs-fs/hubfs/command%20line.png?width=2100&name=command%20line.png)
![Results of using Deeplite Neutrino](https://blog.deeplite.ai/hs-fs/hubfs/Results.png?width=1618&name=Results.png)
Deeplite's hands-off optimization tool requires one command line which will then launch the process.
For example, you could have started with a large and slow Pytorch model, but then running it through Deeplite’s engine, you’ll get a highly optimized model with full precision. Then, that model can be run through whatever third-party tool the user wants, that are running on GPUs or Intel CPUs.
Coming together as a community
Overall, the Workshop really showcased the sense of community in the AI and computer vision space. AI innovations are pushed forward by bringing great minds together, which is why it’s important to have events like these. This community of engineers and others allows us all to collaborate, learn and exchange information, which is what CMC and the community versions of our products are all about.
Deeplite Neutrino was just released into the CMC Community, so you can visit here to get it now. Coming soon to CMC’s platform is DeepliteRT, and we are currently working with our custom AI accelerator partners to get Deeplite’s quantization capabilities on the CMC platform too. View the full recording, including Nick’s presentation, here.