AI startup Abacus.ai nabs $22 million in Series B funding to automate creation of deep learning models

Abacus.ai, the year-and-a-half-old San Francisco startup that seeks to automate deep learning models for customers, said Wednesday it has received $22 million in financing in a Series B round led by venture capital firm Coatue, bringing the company's total funding to $40.5 million. 

Coatue joins former investors Index Partners, which participated in a Series A investment round totaling $13 million in July, and new investor Decibel Ventures. The firm has an impressive list of individual investors among the technorati, including Google CEO Eric Schmidt, investor and former Amazon executive Ram Shriram, and Yahoo! co-founder and onetime CEO Jerry Yang. 

Coatue has funded a variety of firms across the tech landscapes, including Databricks and DoorDash.

Abacus, which began life as Reality Engines, is competing in a vibrant market for industrial AI, the application of machine to business problems. 

The company employs a variety of approaches to deep learning, including, in particular, generative adversarial networks, or GANs, to offer a kind of push-button service in the cloud that lets companies train and test and deploy novel AI programs without the hassle of traditional laboratory deep learning work. 

The approach is known as neural architecture search, where the deep learning program performs a search through possible neural network architectures to find one that is optimal for a given task, rather than a person designing the architecture by hand. The service went live this summer. 

Along with the funding announcement on Wednesday, Abacus.ai announced it will be offering parts of its technology as modules, for companies that want to pick and choose which parts they want. One module is to host and monitor machine learning models. They can be models developed with Abacus.ai's help, or models brought to Abacus.ai that the customer has already developed. 

A second module can be used to de-bias a machine learning model, to remove age, gender or race bias. And a third module, the Feature Store, is a facility for companies that stage and prep data for their models. 

Abacus calls the modular offering collectively by the name Abacus.ai Deconstructed. 

Abacus notes the same modules have always been part of the automated neural architecture search service. "Deconstructed simply makes available these modules to our customers, in case, they want to use one or more modules to help them accelerate the models they train in-house to production," the company said. 

Abacus.ai does fundamental work on the science of deep learning to advance its tools. The company will present two papers next month at the NeurIPS conference on machine learning. One, titled, "A Study on Encodings for Neural Architecture Search," proposes a way to encode a neural network architecture so as to be efficiently manipulated by the search algorithm. Another, titled "Intra-Processing Methods for Debiasing Neural Networks," is the basis for the de-biasing service. 

Abacus.ai currently employs 32 people and chief executive Bindu Reddy says the company "plans to hire aggressively to hit 50 by the end of the year."