Today the AWS Artificial Intelligence Bootcamp took place at the VW:Datalab in Munich. We were of course on-site and learned a lot about Deep Learning with Amazon Web Services. In addition to the AWS AI services Polly, Lex, Rekognition, and the Machine Learning platform itself, much was discussed about the technology behind the services. Amazon’s recommendation system is also based on modern deep learning algorithms. Like other major vendors, Amazon also has an open source framework for this task.

MXNet as Amazon’s AI Driver

Amazon is the largest contributor to Apache MXNet. MXNet is a scalable Deep Learning framework that is strongly reminiscent of TensorFlow, Caffe, and others. A Deep Learning framework is typically characterized by its ability to handle and compute matrices or “tensors” particularly well. MXNet, similar to Spark, also optimizes the computation graph and improves runtime many times over.

Artificial Intelligence Replaces Machines

Overall, the day in Munich was a complete success for us. Parts of the newly acquired knowledge can be directly applied in our platform croGoDeal, even though it is/will be programmed in Scala rather than Python.

To conclude the event, one of the speakers from Amazon Web Services showed interesting calculations regarding artificial intelligence and the singularity:

A human can perform approximately 10^15 operations/s. To book this capacity with a Deep Learning network on AWS would cost roughly $105,000/h. Training a good Deep Learning model alone would require about 3 weeks. This example clearly shows that humans will not be replaced by machines anytime soon. However, one should not forget “Moore’s Laws.”