Snapshot is based on recent experience and what I like about them.

Azure - Auto-ML and Power BI link up really compresses cycle time from data to ML and visualize the result. Assume data prep is done, and for test drive case data stored in Excel, running simple Auto-ML classification from setup to execution in 15 min. No programming in R or Python required (think NASCAR, stock cars on steroid).

AWS - 1) Alexa really opens the door to AI without explicitly learning how to program in NLP (natural language processing). Like writing a screenplay, “utterance” and “intent” (in English or language supported by AWS) are what brings the voice interface to live. 2) picture the ability to work remotely, e.g., conduct ML experiments while watching sunset on a beach. AWS WorkSpaces makes that possible (e.g. chromebook / browser with access to internet). https://clients.amazonworkspaces.com/
Its companion WorkDocs facilitates collaboration, upload data / download results. "Total" productivity.

GCP (Google Cloud Platform) - for deep learning (i.e., deep neural network DNN), tensorflow estimator is a great tool with minimal coding. In its purest, it takes one line of code to specify simultaneously the width and depth of DNN configuration. For 158-layer DNN, simply type in a list [n1, n2, n3, n4, …., n158] where n1 is number of neurons in NN layer 1, n2 in layer 2 etc. For full throttle, GCP offers up tensor processing unit (TPU) which runs faster than CPU and GPU. Think Formula One race car.

The rate of innovation is always greater than any organization doing it alone. Cloud is the real equalizer for small team. Predict by 2020, kubernetes will afford true interoperability among different cloud platforms. Mitigate the risk of vendor lock.

Article of interest: https://www.cnbc.com/2019...s-at-build-2019.html


p.s. in my humble opinion, with 80/20 rule thinking about ML, Azure makes a good bet to get the job done. Alexa helps bring others along in using AI from operations control to the C-suites. GCP is great for exploring DNN, worthy of investigation what TPU can do in real-time critical application.

Let's share and hear from the community what other lessons learned running business in the cloud.