Which cloud offers better AI tools?

This is a review of the artificial intelligence and machine learning offering of the top three clouds [Amazon, Google and Microsoft].

Google AI

Google is known for its AI and big data skills. Google has taken Jupyter Notebooks and made an easier version called Google Colab[oratory]. By mixing instructions and guides with running the AI code on Google Cloud, AI data scientists, who have been trained on Kaggle, will find this a very familiar environment. Google also invented Hadoop and other large scale cloud storage. Their AI tools are integrated into their versions of large scale storage, compute and analytics like Big Query. The main AI service Colab uses is called Vertex. Additionally you have Vertex Workbench which is a single development environment for the entire data science workflow [datasets, feature store, labelling tasks, workbench, pipelines, training, experiments, model registry, endpoints, batch predictions, metadata, matching engine, marketplace,...]. Google offers access to optimised virtual servers with GPUs and TPUs [tensor processing units]. Google also offers AI solutions to non-AI experts in the form of AutoML. A series of task specific AI products allow for different solutions from processing text / sentiment analysis [Natural Language AI], transcribe recordings into text [Speech-to-Text], as well as the reverse [Text-to-Speech], translations [Translation AI], Document Management [DocAI Warehouse], image classification/object detection/vision product search/… [Vision AI] and conversational AI [Dialogflow]. Google also has vertical AI solutions, e.g. Retail API, Talent Solutions [HR / Job postings],...

Advantages

• Easy to use and get started solutions for data scientists, e.g. Colab.
• Simple to make complex ML pipelines and train on armies of GPU/TPU servers.
• Close integration with Keras, Tensorflow,...

Disadvantages

• As with any big organisation, there are parts that are considered “legacy”. When going into Natural Language AI, Video AI, Vision AI,... you are asked to migrate to Vertex AI instead.
• It is very clear that Dialogflow was acquired and integrated into Google Cloud instead of a natively built service. It is behind the others.
• Not a lot of vertical services.

Amazon AI

AWS ‘ main AI product is SageMaker to build, train and deploy machine learning models at scale. SageMaker Studio is a dark mode and enhanced version of Jupyter Labs. Amazon offers a lot of other AI services. Augmented AI involves human reviews. Amazon Comprehend focuses on natural language processing and text analytics.? AWS DeepComposer is all about generative AI, generative adversarial networks, autoregressive convolutional neural networks, transformers,... all applied to music. AWS DeepRacer is similar but to train a robot car to drive autonomously. Amazon DevOps Guru allows machine learning to be applied to DevOps. Amazon Forecast is an ML service to provide predictions and forecasts. Amazon Fraud Detector enables easy detection of fake account creation, online payment fraud,... Amazon Kendra provides AI powered enterprise search with natural language processing. Amazon Lex is providing the power of Alexa to anybody. Anomaly detection is done via Amazon Lookout. Equipment uses sensors in equipment, Metrics bases analysis on business metrics where Vision uses computer vision. Amazon Personalize enables the Amazon Marketplace personalisation and product recommendation power to be used by anybody. Amazon Polly is text to speech. Amazon Rekognition is for object detection inside images and videos. Amazon Textract is optical character recognition and scanned document reading as a service. Amazon Transcribe is speech recognition. Amazon Translate is AI-powered translation. AWS has also industry specific solutions, e.g. Amazon Comprehend Medical [NLP and text analytics for healthcare], Amazon HealthLake [analyse, transform, query, … health data], Amazon Omics [genomics data analysis], Amazon Monitron [monitor industrial equipment]. Amazon is also strong in Internet of Things, e.g. AWS Panorama [computer vision for cameras].

Advantages

• AWS ‘ Jupyter Lab Notebook comes with a very long list of sample SageMaker notebooks.
• Lots of pre-trained solutions to solve very concrete AI tasks from recommendations, to monitoring, anomaly detection,...
• AWS DeepComposer helps people new to AI to understand how generative AI can be applied to music. AWS DeepRacer teaches the same for autonomous driving robots.
• Strong in healthcare, industrial and financial services.

Disadvantages

• When using the SageMaker setup, AWS assumes a data scientist knows a lot about cloud policies, VPCs, encryption, IAM roles, ARNs, … SageMaker is not a tool for AI beginners.
• Amazon is great for specific tasks but as a general AI platform it is overly complex.

Microsoft Azure AI

Azure Synapse Analytics is a fully managed data warehouse for enterprises. Azure DataBricks is a hosted Apache Spark. Azure Machine Learning is the equivalent of AWS SageMaker and Google Vertex. It contains Notebooks, Automated ML and Designer as well as all the other stuff a data scientist needs. Azure Open Datasets gives users access to public datasets. Data Science Virtual Machines are pre-configured machines with development tools, deep learning / ML / AI tools, data platforms and ingestion tools. Azure Kinect DK is for the famous Kinect 3D camera. Azure also has healthcare services [Microsoft Genomics, Health Bot,...] and industrial / IoT solutions [Project Bonsai allows for industrial control system simulations, Azure Percept for edge AI].??

Cognitive Services brings text-to-speech, speech-to-text, translation, computer vision, custom vision, face API, anomaly detection, content moderation, personalizer [recommendations], Azure OpenAI,... all together. Applied AI Services includes Form Recognizer [object detection], Immersive Reader [OCR], Metrics Advisor [metrics monitoring and anomaly detection], Video Analyzer / Video Indexer, Cognitive Search [natural language and other types of searches], Bot Services [like DialogFlow and Flex but on top of Microsoft Bot Framework].

Advantages

• Relatively simple to use AI/ML services.?
• The power of OpenAI?
• Unique services like Kinect DK.

Disadvantages

• Lack of onboarding functionality when using new services and no embedded tutorials. Several services do not even have a simple explanation of what they are for and want you to immediately launch a new paying service.

Which is best for whom?

If you are a general or a novice data scientist you are probably best to start with Google. Google makes it super simple to go from Google Colab [a better Jupyter Lab Notebook which is similar to Kaggle] on a single machine to distributed GPU/TPU clustered model training and inference. Microsoft would be a second choice and Amazon SageMaker would be a distant third.

An API focused team will probably like all three. If you are in healthcare then Amazon and Azure are best. In IoT and industrial Amazon and Microsoft are very strong. Fraud detection on Amazon. Human Augmented AI, DeepRacer and DeepComposer are innovative Amazon services. Microsoft beats everybody with OpenAPI and 3D cameras [Kinect]. Google and Amazon are strong in retail. Amazon used to be ahead with Alexa but Microsoft Bot Framework and OpenAPI has changed this. Google’s Dialog Flow is running behind. Microsoft is great if you want standard data and analytical infrastructure, e.g. Spark, whereas the others have built custom solutions, e.g. Google’s Big Query.?