Deeplite Raises $6M Seed Funding - Edge ML with Reduced Compute
Optimizing Machine Learning Model Deployment with Deeplite
Deploying machine learning applications often presents challenges due to high costs and substantial computational demands. Deeplite, a Montreal-based startup, is addressing this issue by offering solutions to reduce model size. This allows for execution on hardware with limited resources.
Recent Funding and Investors
The company recently secured $6 million in seed funding. PJC, a venture capital firm located in Boston, spearheaded the investment round. Additional support came from Innospark Ventures, Differential Ventures, and Smart Global Holdings.
Somel Investments, BDC Capital, and Desjardins Capital also contributed to this funding initiative.
Deeplite’s Core Technology
According to Nick Romano, CEO and co-founder of Deeplite, the company’s focus is on enhancing the efficiency of complex deep neural networks. These networks typically require significant computing power, consume substantial memory, and rapidly deplete battery life.
“Our platform facilitates the transformation of these models into a more suitable format for deployment on resource-constrained edge devices,” Romano stated.
Applications and Device Compatibility
These devices can range in size from smartphones and drones to compact single-board computers like the Raspberry Pi. This expanded compatibility enables developers to implement AI in scenarios previously considered impractical.
Introducing Neutrino
Deeplite has developed a product named Neutrino. This tool allows users to define their deployment requirements and specify the desired level of model compression. The goal is to minimize both model size and the resources needed for production execution.
The platform aims to enable machine learning applications to operate with an exceptionally small footprint.
The Compression Process
Davis Sawyer, chief product officer and co-founder, explains that Deeplite’s solution is applied after a model has been constructed, trained, and prepared for deployment. Users provide the model and its associated dataset.
Subsequently, they can determine the optimal level of compression, potentially accepting a slight reduction in accuracy if permissible. The primary focus is on minimizing the model’s size.
“Compression significantly reduces the model size, enabling deployment on less expensive processors. We’ve observed reductions from 200 megabytes to 11 megabytes, and even from 50 megabytes to 100 kilobytes,” Davis clarified.
Investor Perspective
Rob May, leading the investment for PJC, expressed his admiration for the Deeplite team and their innovative technology.
“The deployment of AI, particularly deep learning, on devices with limited resources is a widespread industry challenge, hampered by a shortage of skilled AI professionals. Deeplite’s automated software solution promises substantial economic benefits as Edge AI becomes a dominant computing paradigm,” May commented.
Company Origins and Growth
The company’s origins lie within the TandemLaunch incubator in Montreal. Deeplite officially launched in mid-2019 and currently employs 15 individuals.
The company plans to expand its workforce to 30 employees by the end of the current year.
Commitment to Diversity and Inclusion
Romano emphasizes that the founders are dedicated to fostering a diverse and inclusive organizational culture.
“Our strategy prioritizes attracting the best talent while ensuring a diverse and inclusive environment. This commitment is integral to the company’s core values,” he stated.
Future Workplace Plans
Deeplite intends to establish offices in both Montreal and Toronto as employee hubs once it is safe to return to in-person work.
However, there will be no mandatory requirement for employees to work from the office.
“We are adopting a flexible approach, allowing employees to work remotely as they prefer. We anticipate a reduced office footprint compared to traditional models,” Romano concluded.