Several businesses struggle with getting machine learning into production. Although many have completed their pilot efforts, businesses actually found it difficult to implement it actually into production.
Even though the industry seems to be interested in machine learning, according to a CIO survey, the deployment started growing at a slower rate between last year and this. The report stated that, in 2017, 4% of organisations had deployed AI in production, however, in 2018, 14% had deployed to production, which has become a 19% deployment by 2019.) Experts believe that businesses which have already experimented with open source technologies in their pilot efforts will likely to be interested and curious about machine learning platforms. Recognising the need in the market, AWS has recently rolled out a number of machine learning offerings that claim to help organisations to get ready for the deployment efforts.
Sign up for your weekly dose of what's up in emerging technology.
In a chat with Denis Batalov, Worldwide Technical Leader, ML and AI, Amazon Web Services, at the AI Conclave 2019, he spoke about the artificial intelligence (AI) services which are designed to put machine learning in the hands of end-users — with no machine learning experience.
AIM – AI and ML have always been lumped together, what is the connection between the two? Are they interdependent?
AI and ML are often lumped together. At the end of the day, machine learning is a technique that enables AI applications. In the early days of AI, people were trying to manually cram all the knowledge and all the rules about the world into the system, and let the system use those rules. Quickly, the realisation came that it’s impossible to do that as there are so many facts, so many rules, and they change as well, and this wasn’t the way to build intelligent applications. So machine learning was a way to fix that – as techniques for machines to automatically acquire knowledge of rules and at the same time achieve results that we want.
If you look at AWS’s offerings, Amazon SageMaker is our platform service for ML. At the higher level of our three-layer stack is our AI services. For those services at the top layer, you don’t need to manually build ML models yourself, as the models are built for you. So the machine learning is almost taken care for you, and you’re just using the intelligence in those applications. Amazon SageMaker is the most flexible technology to build custom algorithms and deploy them into production.
AIM – What are the major AI /ML trends and what are your customers telling you?
Well, the number one observation is that machine learning is no longer an aspirational technology, it’s rather pervasive, as every single industry is incorporating it. We at AWS see a broad range of adoption and use cases of ML applications. For example – Intuit – which is a tax software company – is using SageMaker and machine learning to discover tax savings by automatically processing lots of financial records, and that is of course, very useful to people.
You can look at key applications in the medical and healthcare industry, which are actually affecting human lives. Cerner has built a model for congestive heart failure, and using that model, they can detect or predict 15 months in advance something before clinical manifestation. For example, Formula One or various sports, these are important for a lot of people, and it’s changing their lives. In India for example, we have clients like Shaadi.com, they want to ensure that the pictures that are available are appropriate, so they are using Amazon Rekognition for image moderation to make sure only appropriate information shows up.
The key thing to notice is that the adoption of this technology is across the board. If we take a specific use case of predictive maintenance, or churn prediction, and understand how it works and proves the algorithms on it, we will see more and more managed services appear with ready-made solutions for specific use cases. So instead of saying there’s a platform that enables people to build any machine learning application, here are some specific services – for example, we launched Amazon Kendra for ML-powered search; we launched Amazon CodeGuru, which is a machine learning-powered software development process; Amazon Personalize for recommendations; and Amazon Fraud Detector, and so on for specific use cases. So we should see more and more of these ready-made services and solutions, and that is quite clear as a trend.
AIM – A brief introduction to Amazon’s AI/ ML services. Let’s talk about SageMaker and SageMaker Studio.
The beauty of SageMaker is that it’s an open platform, which will help you pick and choose the required tools one need to support their ML journey. It removes the heavy lifting from each step of the machine learning process to make it easier to develop high-quality models.
SageMaker is being used by thousands of customers to accelerate their machine learning deployments. We have regularly added new capabilities to it, including Amazon SageMaker Ground Truth to build highly accurate annotated training datasets, SageMaker RL to help developers use a powerful training technique called reinforcement learning, and SageMaker Neo, which gives developers the ability to train an algorithm once and deploy on any hardware. These capabilities have helped many more developers build custom machine learning models.
With regard to SageMaker Studio, the best way to think about it is to compare it to a software development process. The machine learning process is not so different, and that’s what SageMaker Studio is all about. It’s offering a single suite of tools that enable the daily work of machine learning practitioners and data scientists.
With SageMaker Studio, developers can write code, track experiments, visualise data, and perform debugging and monitoring all within a single, integrated visual interface, which significantly boosts the developer’s productivity. Additionally, since all these steps of the ML workflow are tracked within the environment, developers can quickly move back and forth between steps, and also clone, tweak, and replay them. This gives developers the ability to make changes quickly, observe outcomes, and iterate faster, reducing the time to market for high-quality ML solutions.
Another exciting new functionality we launched is SageMaker AutoPilot, which can be used by somebody who does not need to know the details of machine learning algorithms and can just provide data in a tabular format or CSV files. In turn, the system automatically figures out what pre-processing steps are needed for the data, what algorithms need to be chosen and how these algorithms need to be configured and provides a result by choosing the best model out of all the different algorithms is made. No machine learning experience is needed in that case, and that’s very powerful.
AIM – Comparison of the Indian market with the global one, in terms of ML implementation.
The real difference is often bigger among different industries, instead of countries. We live in a globalised world where these new technologies are accessible anywhere, therefore it won’t be completely right to say that there is any major difference in terms of adoption of ML services in India with respect to the global economy.
There’s definitely lots of interest and lots of work being done in India – redBus for example, is using SageMaker to categorize reviews on their platform; FreshWorks, for example, has built 33,000 machine learning models for different customer interactions using SageMaker and is able to reduce their training time from 33 hours to 27 minutes. The kind of questions that people ask – they’re no different from anywhere else in the world. I think it all depends on the individual company’s journey in adopting AI and ML technology. There’s lots of interest; some are starting and some are already quite advanced.
This is what we’re trying to do at AWS, this is why we have the layered approach – an experienced practitioner can look at SageMaker, and a talented developer can just call an API and they can get the intelligence added to the application that way.
The key difference between businesses in adopting ML models or any other newer technologies is the ‘company culture’, and the ability to transform the company culture. Definitely, there are a lot of startups who are born in the cloud and think about ML from the very inception. Sometimes, their whole business is made on machine learning. However, there are big traditional companies as well who are recognising that ML is transformative, and they need to be able to apply it and it’s a matter of execution.
Machine learning is not a one-off project. Companies need to invest in a data strategy, ensure that the data is collected, and is easily accessible to people who want to build machine learning models. They also need to invest in educating and training people, which is why we have various programs offered by our training and certification team, and various training partners.
AIM – With the emergence of newer technologies, how important is upskilling for you?
With so many technological advancements, the pace is really high, and it is true that people need to be on the lookout for upskilling themselves or follow the latest technology.
Even in terms of courses, if one semester students are being taught Java, the next semester the course needs to be changed because things and trends must have shifted during those six months. So upskilling has always been a trend in the technology industry where things are changing rapidly. What is more important is to remember that, while it’s easy to get drowned in the information and updates available these days, one has to remember that, the key skill is, ‘learn and be curious’, which is the leadership principle of Amazon, which means that one has to be open-minded to accept new technologies and understand what they are. People need to pick one technology, which can be their niche and should aim at learning that thoroughly. There are plenty of materials out there which can help people to gain their knowledge.