Cloud Machine Learning: AWS v/s GCP v/s Azure

Cloud services are everywhere. The next big bet for Amazon, Microsoft and Google has been cloud for a while now. Winning this bet would establish the next technology leader.

Market share

Amazon being an early entrant, in 2006, made cloud computing an industry wide term. Since then it has bulldozed its way into acquiring new customers and holding approximately 40% of the market share. Microsoft and Google have followed in Amazon’s footsteps and hold approximately 12% and 5% market share respectively. (Source: Synergy Research Group, Feb 2018)

Today, companies rely on cloud services for more than just storage and compute. Companies expect to be able to exploit Alexa’s language skills, Google Photo’s identification skills, etc. The list is endless and these intelligent services are what will differentiate the otherwise commoditized cloud storage services.

Competitive Landscape

In the Cloud Machine Learning space, these three companies battle on - enabling platforms for model deployment, natural language and vision. Google stands out here, as it also provides TPU (hardware) and transfer learning services.

Amazon, Machine Learning:

1. Deploying Models: Sagemaker, Apache MXNet, Tensorflow on Amazon, Deep Learning on EC2: Machine Learning and Deep Learning platforms, to deploy scalable machine learning solutions.
2. Language Skills: Comprehend, Lex, Polly, Recognition, Translate and Transcribe: Here AWS has leveraged its progress on Alexa and exposed these capabilities to users.
3. AWS DeepLens: Only vision product on the Amazon portfolio. You could deploy your ML vision algorithms to a wireless camera.

Google, Cloud AI:

1. TPU: With 180 Teraflops per-second designed to work exclusively with tensors, TPUs are set to revolutionize machine learning.
2. Cloud AutoML: Google’s transfer learning would mean companies with limited data could build on pre-trained models by Google. Imagine an online retailer being able to classify shoes by type based on computer vision. Traditionally you would need 10’s of thousands of examples to train an algorithm like this with low accuracy. With transfer learning 100 examples could yield better results.
3. Deploying Models: Machine Learning as a Service: Which includes pre-built models and flexibility to build and deploy your own models.
4. Language Skills: DialogFlow, Speech-to-text, Text-to-speech, Natural Language, Translation: Here like Amazon, Google exposes its language skills from youtube, google assistant, translate, etc. to its users.
5. Image and Video Analysis: Identifying objects in images and videos and understand the content or context of the video or image.

Azure, Machine Learning:

1. Machine Learning: Environments for data scientists to run machine learning experiments.
2. Language, Azure Bot, Knowledge, Search, Language, Speech: Conversational service, helps build automated responses to customer queries and natural language capabilities for text analysis.
3. Vision: Identify, caption and moderate your pictures

Based on the services offered and the complexity of machine learning capabilities exposed, Google is the clear winner. Amazon’s offerings in the speech and machine learning are comprehensive. However, Amazon falls short on vision, custom hardware support and provides no transfer learning. Microsoft lacks the most, with limited development in all three areas and limited pre-trained models it can expose to users.

As for Google, with TPU accelerators, extensive language and vision skills and transfer learning (my favorite), Google has taken an ambitious path to providing machine learning cloud solutions. It is yet to be seen if these differentiating forces would help Google compete with Amazon’s dominant position. However the challenge and promise of the future is exciting.