Following examples set by the likes of Google and others, Amazon has made its Deep Scalable Sparse Tensor Network Engine (DSSTNE) generally available to researchers, developers and everyone else. The engine, which is used to provide product recommendations to Amazon shoppers — usually under the “You may also be interested in” — is now available on Github.
The package includes examples, instructions for setup, FAQs, User guide and holds a business-friendly Apache 2.0 license.
Speaking on the topic, Amazon said,
We are releasing DSSTNE as open source software so that the promise of deep learning can extend beyond speech and language understanding and object recognition to other areas such as search and recommendations. We hope that researchers around the world can collaborate to improve it. But more importantly, we hope that it spurs innovation in many more areas.
Corporations seem to have realized that with more people working on the technologies they usually hold so close to their heart, chances of boosting development and arriving at breakthroughs is drastically increased. Which is why we have a spree of new posts on Github, that contain technologies from the likes of Google, Microsoft, Yahoo and Amazon.
However, packages containing machine learning platforms are something which are still relatively scarce on the website — more because so few companies actually work in the field than because of any unwillingness to share. After all, the harder a thing is, the more likely you are to seek help for it. Google for example, released it’s Tenserflow machine learning platform a few months ago, a launch that was closely followed by Microsoft’s release of DMTK.
Well, Amazon has jumped the bandwagon now. The company also claims that its recently released platform is significantly better than Google’s Tensorflow. According to Amazon, DSSTNE — read “Destiny” — is not only capable of outperforming Google’s machine learning platform when there is less data to crunch, but can scale better across multiple machines and is significantly easier to deploy.
Amazon claims that DSSTNE can solve recommendation problems and perform natural language understanding tasks almost twice as fast as compared to Google’s library.
DSSTNE is much faster than any other DL package (2.1x compared to TensorFlow in 1 g2.8xlarge) for problems involving sparse data, which includes recommendations problems and many natural language understanding (NLU) tasks.
The company also said that it meant DSSTNE as a real world tool and thus focused more upon the production deployment of real-world deep learning applications, emphasizing upon speed and scale over experimental flexibility.
Some of the DSSTNE’s other capabilities include:
- Multi-GPU Scale: Training and prediction both scale out to use multiple GPUs, spreading out computation and storage in a model-parallel fashion for each layer.
- Large Layers: Model-parallel scaling enables larger networks than are possible with a single GPU.
- Sparse Data: DSSTNE is optimized for fast performance on sparse datasets. Custom GPU kernels perform sparse computation on the GPU, without filling in lots of zeroes.
In a practical situation, this equates up to being able to build recommendations systems that are able to model ten million unique products or even perform Natural Language Understanding tasks with very large vocabularies. While other packages would need to revert to CPU computation for the sparse data of this size which would equate to a decrease in performance by about an order of magnitude.
While the platform does not yet come with the convolution layers needed for image processing, and has only limited support for recurrent layers needed for many NLU and speech recognition tasks, Amazon said that both of these features are on the way.
Meanwhile, DSSTNE is on Github folks, so in case you want to take a look — and even suggest improvements — at the workings of the system that recommends products to you while you are making a purchase, feel free to go and take a look.