Cloud machine learning consulting at Fast Data Science

azure data scientist associate 600x600It is not always possible or practical for a company to run machine learning on their own machines. This requires considerable investment in hardware, and it is often easier to rent computing power rather than buying equipment which may become obsolete or be unwanted in the near future. For that reason, many of our clients prefer us to build machine learning models on rented servers, known as the cloud. This is our cloud machine learning consulting offer.

We are able to work with any cloud ML environment to develop and deploy machine learning models to production. Fast Data Science is in the Microsoft Partner Network and the director, Thomas Wood, is certified by Microsoft as an Azure Data Scientist Associate.

We can provide cloud machine learning consulting on any platform. The main cloud machine learning providers are:

When we have compared costs between these providers, in general they turn out to be quite competitive between each other.

It can be quite daunting to choose a cloud machine learning solution, and it is particularly difficult to plan costs. In addition, it is quite easy to leave a cloud server turned on accidentally, and run up a large bill at the end of the month. As part of our cloud machine learning consulting offer, we will assist with these tasks and set up cost monitoring on the cloud platform, so that you are always in control of how much money you are spending on your cloud machine learning applications.

NLP on Cloud Machine Learning

Since at Fast Data Science we focus on Natural Language Processing applications, this article addresses cloud machine learning from a text processing angle.

Natural Language Processing tasks are often resource-intensive and require high-performance computing environment such as GPUs, if they rely on deep learning technology.

It is impractical for every organisation to purchase and maintain the necessary hardware and for this reason cloud machine learning has become essential for natural language processing applications.

There are a number of text analytics vendors on the market offering both pre-trained NLP models and the ability to build your own custom model. As of 2020, the main competitors in the market are

cuadro cloudml

Due to market forces, prices and offerings tend to be broadly similar across vendors. At Fast Data Science we are prepared to look into and evaluate any cloud NLP provider according to your organisation’s needs.

What is the Cloud?

With all the talk of ‘the cloud’, you could be forgiven for thinking that machine learning has finally learnt to control the weather! In reality, ‘the cloud’ refers to ‘cloud computing’, which is simply the idea of using somebody else’s computers on the internet for running your programs or storing information. The cloud is everywhere, accessible from anywhere, and you often don’t need to care which country the computer you are using is located in. In fact, if you use Google Docs, Gmail, Onedrive or Zoom, the program you are using relies on the cloud. Some cloud applications, such as Google Docs, are free or have a freemium version, but for commercial applications of machine learning, you will probably need to pay.

Why run AI and NLP on cloud machine learning?

Traditionally, if you wanted to train and deploy a machine learning model online, for example, a price prediction model on a website, you would have had to rent or buy a server and invest a lot of time maintaining it. This was both expensive and troublesome.

Cloud machine learning, or cloud AI, means the practice of renting somebody else’s computers (‘the cloud’) to train and deploy machine learning models. You pay for the usage of the cloud provider’s infrastructure instead of making large upfront investments, and the computations take place in the cloud. Examples include Google AI, Amazon AWS, and Microsoft Azure.

Machine Learning Consultant
2-min
MANO2-compressed
3-min

Cloud machine learning for healthcare

Cloud solutions are becoming increasingly used to bridge the final mile between patients and healthcare providers, as it is hard to deploy machine learning solutions on-premise at all locations where patients will receive treatment. The offerings of cloud machine learning are improving access to care in the case of geographically distributed populations.

Because medical data is highly sensitive, it is important to ensure that data does not leave the jurisdiction of the patient. For example, Microsoft Azure has a system allowing machine learning solutions to be deployed across regions without sending data between regions, and similar solutions are available with competing cloud providers.

Advantages of Cloud Machine Learning versus On-Premise Solutions

There are several advantages of using a cloud machine learning environment to build your models:

  • you don’t have huge upfront costs of buying hardware that you may later need to upgrade or which may become obsolete.
  • somebody else takes care of servers, IP addresses, SSL certificates, DNS, and so on, so you don’t have to.
  • you have the flexibility of being able to upgrade or upscale your hardware such as GPUs at the click of a mouse when demand grows, similarly you can downsize your solution when demand shrinks.

For this reason many organisations, including most UK public sector bodies, prefer to work with cloud providers such as Microsoft, Google and Amazon when they commission AI models.