Artistic Algorithms, Deep Learning Shift to Cloud
Application builder Algorithmia has expanded its marketplace to include a cloud-based platform for creating deep learning environments for training style transfer models, a type of filter that can for example convert images to match a particular artist’s work.
Among other things, the tools provide an algorithmic understanding of how humans create and perceive artistic imagery so that trained models can covert photos and images into familiar artistic styles. The Seattle-based startup’s tools also target smart applications.
The startup has released a machine image running on the Amazon Web Services (NASDAQ: AMZN) cloud along with a pipeline for creating style transfer models. The tool allows users to set up a deep learning environment in the cloud, train models and release them as a scalable REST-based API. The platform allows users to train models in about a day using GPU-based Amazon Elastics Compute Cloud instances.
The Amazon Machine Image (AMI) pipeline provides access to 38 different filters that can be used by developers for new applications and services, the company said. All are available via a company API. The AMI pipeline also handles installation of necessary software to the AWS cloud instance, including a deep learning library called Torch and a model dubbed TextureNet.
The three-year-old company also unveiled an interactive demonstration of style transfer techniques that includes five filters that can be used for free.
The Algorithmia marketplace is intended to bring algorithm developers together “to share knowledge, test algorithms and run them directly in their applications,” the company notes. The marketplace provides access to all of its algorithms—including facial, text, and speech recognition; sentiment analyses; recommendation systems; time-series forecasting; and natural language processing–via a REST-based API.
The tool rollouts also reflect the shift within the machine intelligence field toward leveraging GPUs “on-demand” running in the cloud as a way of reducing tedious tasks. That approach frees application development teams to experiment with deep learning algorithms and models that can be trained in a day or so at a fraction of the cost of traditional approaches. Algorithmia estimates the cost at about $25 per model using its open source Amazon Machine Image for training style transfer models.