AWS Machine Learning: Part 1

Machine learning has grown to be quite the popular trendsetter in recent years for different organizations. Hence, it is unsurprising to see major cloud companies coming up with custom cloud solutions to support data scientists in every way possible.

The growing trends of shifting data storage to cloud, scalability, data security, maintenance, and deriving insights from data have given rise to MLaaS at a federated cost. Machine learning is not just restricted to building the model, rather it consists of many complex components such as data ingestion, data preparation, model training, hyperparameter tuning, model deployment, model monitoring, explainability, and many more. It also requires collaboration across teams, from Data Engineering to Data Science to ML Engineering. With machine learning as a service, data scientists can manage these intricacies.

What is Machine Learning as a Service?

Machine Learning as a Service (MLaaS) is a collection of services that provide machine learning tools as part of cloud computing services. It is based on machine learning technology that implies teaching machines to recognize patterns. A typical MLaaS will consist of services that will cover most of the aspects of a machine learning life cycle such as:

• Data preprocessing,
• Model training, tuning, and evaluation
• Data visualization
• Data orchestration and pipelines
• Model deployment

MLaaS services are often ready-made services that can be adapted by any organization as a part of their working needs. The key is in the fact that the organizations do not need to worry about the actual infrastructure and computations. The providers’ data centers manage them remotely.

Benefits of using MLaaS

Let’s understand some benefits of these services:

High computational power: We can get access to highly optimized CPUs and GPUs to implement computationally intensive data processing and modelling.

Dynamic scalability: We can scale applications on the fly with zero downtime.

Security and transparency: Cloud providers take care of security attacks, downtime, and recovery as well. With automated monitoring and visibility, the service usages can be tracked any day.

Affordability: Applications can be built to adhere pay as per usage.

Constant availability:  Use of multiple servers, storage protection, and geo-redundant storing can ensure that the application is up and always running.

MLaaS platforms offer these solutions and many more. Let’s now look at a brief overview of some platforms offering these MLaaS solutions.

Key Players in the Market

These are some of the leading providers of MLaaS services:

Source – https://fullstackfeed.com/machine-learning-as-a-service/

Following are some of the MLaaS services offered by the key players in the market:

1. Computer vision: Amazon Rekognition, Azure Computer Vision Service, Google Cloud Vision API
2. Natural language processing: Amazon Comprehend, Azure Cognitive Service for Language, Google Cloud Natural Language API
3. Speech recognition: Amazon Transcribe, Azure Custom Speech Service, Google Dialogflow Enterprise Edition
4. AI platforms: Amazon Sagemaker, Azure Machine Learning Studio, Google Cloud Machine Learning Engine

In this blog, we will focus mainly on AWS machine learning services, as it captures 70% of the market as of today. AWS is the oldest of its peers and being the oldest, it has already established trust and reliability factors. Read on for more details…

Amazon Machine Learning services 

Amazon-tailored machine learning services are one of the best automated solutions in the market. Amazon Web Services offers machine learning via two ways. The first one is through pretrained models which are ready-to-use services which you can use without any machine learning expertise and the second one is via a machine learning development IDE called Amazon Sagemaker which you can use to build powerful machine learning models.

Source- https://www.infoworld.com/article/3608413/review-aws-ai-and-machine-learning-stacks-up-and-up.html

A few important ready-to-use services are as follows:

Amazon Rekognition- Amazon Rekognition offers capabilities to extract insights from your images and videos related tasks such as custom object detection and object classification, face recognition, text detection, and PPE kit detection.

Amazon Lex- The Lex API has been created to embed chatbots in your applications as it contains automatic speech recognition and natural language processing capacities.

Amazon Transcribe- Transcribe has been created solely for recognizing spoken text. There is another variant called Medical Transcribe which recognizes medical text.

Amazon Polly- It is an automatic text-to-speech translation service.

Amazon Kendra- Kendra API helps websites and applications improve their search service by using machine learning.

Amazon Comprehend- Comprehend carries out a set of NLP-related tasks such as named entity recognition, Language Recognition, Sentiment Analysis, and much more.

Amazon Translate- As the name says, the Translate API translates texts i.e., from one language to another.

Amazon Textract- Textract detects text from data sources such as invoices or receipts, medical records, tax forms, and driving licenses.

Amazon Forecast- Forecast is a machine learning service which provides time series forecasting ability.

AWS ML Hardware- Amazon has launched a few AI-ML physical products. AWS DeepRacer, AWS DeepLens, AWS DeepComposer, and AWS Inferentia are some of the services.

These services contain quite exhaustive implementation documentation that is simple to follow and apply.

Now let’s take a look at some of the disadvantages of MLaaS.

Disadvantages of MLaaS

Well, there aren’t any major disadvantages associated with these services, but if your data needs to be in premise and there are security issues, then you shouldn’t move to MLaaS. Also sometimes, you want to train a model in local and then deploy the model. In this case at least you can save on the training part, and you shouldn’t be using MLaaS.

There are a lot of complexities involved in building a machine learning application end to end, and hence MLaaS is helping us to do so with just a few clicks. But we cannot deny the fact that in order to use these services we will need a bit of data science or domain expertise. In the next series, we will dig into one of the most important services of Amazon – AWS Sagemaker – where we will train and deploy a model. Deep learning has made a lot of practical applications of machine learning possible!

Reach out to us at Nitor Infotech if you’d like to know more about the pioneering technology we leverage to partner businesses like yours through the digital transformation journey. Do check out this blog to discover an automated machine learning approach with H2O.ai. You will also find this blog about Artificial Intelligence engineering interesting.