Abacus.AI, the two-year-old startup that develops deep learning “hybrid” neural network forms, announced Wednesday that the company has secured $ 50 million in venture capital funding as part of a Series C round of funding, led by private equity firm Tiger Global. Management.

The company has now received $ 90.3 million in funding. Tiger Global is joined by returning investors Coatue Management and Index Ventures, as well as Alkeon.

“A lot of it will go to R&D, engineering and science,” said Bindu Reddy, the company’s co-founder and CEO, in an interview with ZDNet via Zoom. “We continue to want to be the best of the breed when it comes to AI and ML platforms.” The other main use of capital will be marketing, including building the sales team.

Also: AI startup Abacus.ai activates real-time deep learning system for businesses

Abacus currently has 45 people, including four in sales and marketing. Reddy plans to increase the company’s total workforce to 80 by the end of this year.

Along with the fundraising news, Abacus.ai unveiled a computer vision application as a service. The company had started with tabular data applications and then earlier this year with natural language processing applications.

“The main difference between us and anyone else in the market today is the hybrid models,” Reddy said. “If you look at someone like OpenAI or even Google, they have language models, they have vision models, but they are pure language or vision models,” in which the function, such as the classification of ‘images, is closely matched to the particular data type. .

abacus-team-photo-2021.png

Abacus.ai has secured $ 90.3 million to help companies bring AI deep learning forms to production. Seen here, clockwise from left, are the co-founders of Bindu Reddy, previously responsible for “AI Verticals” for Amazon’s AWS; Arvind Sundararajan, previously engineering manager for Google’s ad serving technology; and Siddartha Naidu, previously Senior Engineer for Amazon’s Order Fulfillment Team and also BigQuery Software Developer at Google.

Abacus.ai

“What we do is support hybrid models where you can combine language, vision, and structured data to get better results on your models.” An example, Reddy said, would be finding the price of a house, not only on the basis of characteristic attributes, such as the number of bedrooms, but also on the natural language description and then on photographic data that shows the qualities of the house.

Also: AI Startup Abacus.ai Secures $ 22 Million Series B Funding To Automate Deep Learning Model Building

“Adding this language and vision signal into a predictive model is what we focus on, and what we end up being very good at.”

The customer’s example is reminiscent of what Opendoor does with deep learning, for example. Reddy confirmed that the door-to-door sales app is a real customer app, but declined to identify the customer.

To date, more than 10,000 customers have used Abacus.AI to train more than 30,000 models, the company says, “and several of them, including 1-800-Flowers, Flex, Recorded Books, Daily Look and Prodege , use Abacus.AI in production. for many of their AI use cases. ”

The appeal of hybrid models is twofold, said Reddy: they are more resource efficient than traditional deep learning approaches involving a very large number of parameters; and most business problems really look like hybrid approaches, she said.

“We’re a startup, so making pure vision or pure language models from scratch takes a lot more money,” Reddy observed.

“And honestly, most business use cases are hybrid – we’re more or less applied AI.” Feature sets are more often organized by an enterprise user rather than massive amounts of feature discoveries.

Also: AI Startup Abacus Launches Commercial Deep Learning Service, Secures $ 13 Million Series A Funding

“Our goal, at a very high level, is to say that we have extracted all the information from this data.”

Abacus.ai is cheaper than cloud providers for AI applications, argues Reddy, due to the targeted nature of enterprise use cases. “The problem with most cloud platforms is that you spend a lot of money experimenting, more than putting things into production,” she said. “The experimentation required to bring things into production is a lot less” with the Abacus.ai apps, she said.

Abacus.ai uses a consumption-based pricing system, where you pay according to the number of predictions made. It’s similar to Snowflake in the data warehousing market. “The problem with licensing is that they tend not to be value driven, and that stops adoption in a way, telling people they have to spend more money to adopt the product. . ”

In the same way that Snowflake optimizes data compression, said Abacus.ai, Abacus can reduce costs for customers. “We do a whole bunch of optimizations on the models we run so that we don’t have to spend too much on the computation.”

“You can group models together so they’re on the same server, you can run a cluster of access points for data transformations,” are examples of cost savings.

In addition to running a commercial enterprise, Abacus.ai researchers continue to publish in academic circles on deep learning in computing. Abacus.ai has various research papers that have been accepted for conference publication. Two relate to the company’s core competence in “neural architecture research” to automatically discover the optimal architecture for a neural network. These papers were accepted at this year’s NeurIPS AI conference.

Other articles concern explainability, Reddy said, including one accepted by NeurIPS regarding benchmarks for explainability; and another that deals specifically with tabular data, called “Regularization is all you need”.

Another document under development will focus on the hybrid approach.

When asked what kind of business Reddy hopes Abacus.ai will be, she replied, “I think we want to be a cross between Google and Amazon, if that’s possible.”

“Amazon has a great culture and Google is great in terms of technology,” Reddy explained. “If we could combine the two cultures and get the best of both worlds, that would be great.”


Source link