Cloud AI Service Survey
This is a survey of artificial intelligence capabilities provided by Amazon Web Services, Google Cloud, IBM, and Microsoft.
Automatic Machine Learning
Google Cloud Auto ML
Machine Learning Model Training and Deployment
Amazon SageMaker
Jupyter Notebook, API
Amazon SageMaker is an online environment for defining, training, testing, and deploying machine learning models. It gives you a Jupyter notebook from which to do development, and a library to import. The library provides
-
good integration with Amazon’s Elastic Cloud Compute (EC2) service so that as your models can be run on a variety of hardware and software infrastructures, whether during development or deployment.
-
hyperparameter tuning, so that you need not resort to brute-force grid searches through the space of hyperparameters
-
good integration with Amazon’s Simple Storage Service (S3), where many data scientists store training and testing data anyway.
-
a number of standard machine learning algorithms, but you can also develop your own using TensorFlow or MxNet.
Amazon Machine Learning
Web Application, API
For a lightweight introduction to machine learning, there is Amazon Machine Learning. You specify that it should import data from AWS data sources including S3, Redshift, or RDS. You get some visual tools for previewing the data.
You specify one of the columns in your data as a target attribute that the system will learn to predict. It automatically divides your data into a training set and a test set. It uses the training set to train the model, and then tells you how the model did on the test set, and lets you set a prediction threshold. You need not deal with any complexities of the model itself.
Once you’re happy with the model, you can deploy it for real-time (on-demand) predictions through the web interface, or for batch prediction on data in a file or in S3.
Google Cloud Machine Learning (ML) Engine
Command-line interface, Tensorboard
Google Cloud ML Engine includes freely downloadable components for developing machine learning models on a development system. The models can use scikit-learn, XGBoost, Keras, or TensorFlow. Once your models are running locally, you can copy your data into the cloud, and train your model on a cloud instance or in distributed fashion across multiple cloud instances.
Cloud ML Engine can also run in a mode that tunes hyperparameters on your model.
Once the model is trained, you can deploy it to perform prediction on new data. Cloud ML Engine supports “online” prediction via a REST API, or batch prediction.
IBM Machine Learning
Command-line interface, REST API
IBM Watson Studio
Web application
IBM Watson Studio lets you upload data, cleanse and refine it, and visualize it to discover patterns and trends.
Jupyter notebooks or RStudio let you analyze the data.
Built-in models let you classify image or natural language data. A graphical model builder lets you define a Spark ML model.
The service lets you run experiments in parallel and automates evaluation of model performance under various hyperparameter configurations.
It lets you accelerate training by distributing models across multiple servers and using multiple GPUs.
Image Processing (Computer Vision)
Google Cloud Vision
REST API
When you submit an image through its REST API, Google Cloud Vision
- classifies it into thousands of categories, providing scores for the most likely ones,
- detects faces within it,
- detects topical entities including celebrities and logos,
- determines the likelihood of several types of inappropriate content, and
- reads printed words contained within it
Google Cloud AutoML Vision lets you upload images and labels and uses this data to train a recognition model. There is also an option to upload images and have human beings label them.
Amazon Rekognition
REST API
When you submit an image through its REST API, Amazon Rekognition will identify objects, celebrities, text, and activities, as well as detect any inappropriate content. It also provides face recognition and analysis (sex, eyes open/closed, glasses, facial hair, happiness and age range).
Microsoft Azure Computer Vision
Microsoft Azure Custom Vision
Microsoft Azure Face
Microsoft Azure Content Moderator
IBM Visual Recognition
Video Intelligence (Computer Vision)
Google Cloud Video Intelligence
Amazon Rekognition
REST API
When you submit a video through its REST API, Amazon Rekognition will identify objects, people, their paths, celebrities, scenes, and activities, as well as detect any inappropriate content. It also provides face recognition and analysis (sex, eyes open/closed, glasses, facial hair, happiness and age range).