Ben Assaf spent several years building development infrastructure at Mobileye, the self-driving startup acquired by Intel in 2017, while working on methods to accelerate training of AI models at Hebrew University.
An expert in MLOps (Machine Learning Operations) – tools to simplify the process of taking AI models into production and then maintaining and monitoring them – Asaf was inspired to launch a company that removed major barriers for software engineers and companies deploying AI models into production.
“When I first had the idea of starting a company, a lot of companies and people didn’t have the industry experience building or implementing an MLOps practice in their AI development pipeline,” Assaf told TechCrunch in an email interview. “I thought we could make AI ‘small’ so that it would be lighter, faster and less expensive to produce and market.”
In 2021, Asaf teamed up with his wife, Naeul Kim, who was working as a digital transformation consultant for organizations, to found Clicka, one of the startups competing in TechCrunch Disrupt’s Startup Battlefield 200 competition. Clika provides a toolkit for companies to automatically “downsize” internally developed AI models, reducing the amount of computing power they consume and, as an added benefit, speeding up their processes Inference.
“With Clika, you can simply plug in your pre-trained AI models and get a model that is ‘magically’ compact and fully compatible with the target device – server, cloud, edge or on-premises device,” Assaf said.
To achieve this feat, Klika relies on techniques like quantization, which essentially reduces the number of bits — the smallest increments of data on a computer — in the form required to represent the information. By sacrificing some accuracy, quantization can shrink the model without interfering with its ability to perform a particular task, such as identifying different breeds of dogs.
Clika also breeds Report with Potential improvements or changes Which can be done for MOdile to to improve performance.
Interest in ways to make models more efficient is growing as the AI industry faces supply chain issues related to the hardware needed to run these models. Microsoft recently to caution In an earnings report, Azure customers may experience service outages due to a shortage of AI hardware. Meanwhile, Nvidia’s top-performing AI chipset – the H100 series GPU – is It said Sold until 2024.
Of course, Clika isn’t the only startup pursuing AI model compression methods. There’s Deci, which is backed by Intel; OctoML, which, like Clika, automatically optimizes and populates models for a range of different devices; and CoCoPie, a startup that creates a platform for optimizing AI models specifically for edge devices.
But Assaf says Clika has a technological advantage.
“While other solutions use rule-based techniques for compression, Clika’s compression engine uses it [an] “An AI approach to understanding the structures of different AI models and applying the best compression method for each unique AI model,” he said. “We have the world’s best compression toolkit for vision AI, outperforming the current solution developed by Meta and Nvidia.”
“Best in the world” is quite the claim. But Clika has managed to attract investors for its value — raising $1.1 million in a pre-funding round last year with participation from Kimsiga Lab, Dodam Ventures, D-Camp, and angel investor Lee Sanghee.
Assaf wasn’t ready to talk about customer momentum yet, which is Clika’s momentum Currently pre-revenue, running a closed beta For “a few companies.” but Clica plans to pursue seed funding sometime “soon,” he said.