GraphLab’s machine learning platform now supports deep learning

Seattle-based startup GraphLab has released a new version of its machine learning platform that, among other things, lets users build their own custom deep learning networks.

According to a press release:

Combined with GraphLab Create image analysis tools, the Deep Learning package enables accurate and in-depth understanding of images and videos. The GraphLab Create image analysis package makes quick work of importing and preprocessing millions of images as well as numeric data. It is built on the latest architectures including Convolution Layer, Max, Sum, Average Pooling and Dropout. The available API allows for extensibility in building user custom neural networks. Applications include image classification, object detection and image similarity.

[company]GraphLab[/company], which was founded by a group of University of Washington researchers, started as a commercial version of a open source graph-analysis project of the same name in 2013. In July of this year, the company publicly unveiled its new, broader focus as well as the first version of the GraphLab Create platform.

The GraphLab platform, which also now supports Spark as a data source. Source: GraphLab

The GraphLab platform, which also now supports Spark as a data source. Source: GraphLab

The company is now one of a handful of startups promising to let users build their own deep neural networks, joining Ersatz Labs and Skymind. Other startups, including AlchemyAPI, Clarifai and a growing list of text-analysis startups use deep learning techniques under the covers to power APIs or enterprise software that can do things such as image classification or sentiment analysis. A startup called Nervana Systems is building specialized hardware systems for deep learning models.

Deep learning is a field of artificial intelligence that has garnered a lot of attention over the past couple years, thanks in large part to the efforts of huge web companies such as Google, Microsoft, Facebook, Baidu and even Yahoo. There are various types of algorithms and models that fall under the deep learning umbrella, but at a high level they all work by breaking down data inputs into sets of features that can be used to determine what something is or which things are closely related (e.g., words that are used similarly, or songs that sound the same).

Gigaom held a deep learning meetup in September that featured a number of experts and entrepreneurs in the space. You can watch all the sessions here, and Baidu Chief Scientist (and Stanford professor) Andrew Ng’s talk is embedded below.