Abacus.ai, the year-and-a-half-old San Francisco startup that seeks to automate deep studying fashions for patrons, mentioned Wednesday it has obtained $22 million in financing in a Collection B spherical led by enterprise capital agency Coatue, bringing the corporate’s whole funding to $40.5 million.
Coatue joins former buyers Index Companions, which participated in a Collection A funding spherical totaling $13 million in July, and new investor Decibel Ventures. The agency has a powerful listing of particular person buyers among the many technorati, together with Google CEO Eric Schmidt, investor and former Amazon govt Ram Shriram, and Yahoo! co-founder and onetime CEO Jerry Yang.
Coatue has funded a wide range of companies throughout the tech landscapes, together with Databricks and DoorDash.
Abacus, which started life as Actuality Engines, is competing in a vibrant marketplace for industrial AI, the applying of machine to enterprise issues.
The corporate employs a wide range of approaches to deep studying, together with, specifically, generative adversarial networks, or GANs, to supply a sort of push-button service within the cloud that lets firms practice and check and deploy novel AI packages with out the trouble of conventional laboratory deep studying work.
The method is named neural structure search, the place the deep studying program performs a search via attainable neural community architectures to seek out one that’s optimum for a given activity, relatively than an individual designing the structure by hand. The service went stay this summer season.
Together with the funding announcement on Wednesday, Abacus.ai introduced it will likely be providing components of its expertise as modules, for firms that need to decide and select which components they need. One module is to host and monitor machine studying fashions. They are often fashions developed with Abacus.ai’s assist, or fashions dropped at Abacus.ai that the shopper has already developed.
A second module can be utilized to de-bias a machine studying mannequin, to take away age, gender or race bias. And a 3rd module, the Characteristic Retailer, is a facility for firms that stage and prep information for his or her fashions.
Abacus calls the modular providing collectively by the title Abacus.ai Deconstructed.
Abacus notes the identical modules have all the time been a part of the automated neural structure search service. “Deconstructed merely makes accessible these modules to our clients, in case, they need to use a number of modules to assist them speed up the fashions they practice in-house to manufacturing,” the corporate mentioned.
Abacus.ai does basic work on the science of deep studying to advance its instruments. The corporate will current two papers subsequent month on the NeurIPS convention on machine studying. One, titled, “A Research on Encodings for Neural Structure Search,” proposes a approach to encode a neural community structure in order to be effectively manipulated by the search algorithm. One other, titled “Intra-Processing Strategies for Debiasing Neural Networks,” is the premise for the de-biasing service.
Abacus.ai at present employs 32 folks and chief govt Bindu Reddy says the corporate “plans to rent aggressively to hit 50 by the tip of the yr.”