How AWS’s five tenets of innovation lend themselves to machine learning

As machine learning disrupts more and more industries, it has demonstrated its potential to reduce time spent by employees on manual tasks. However, training machine learning models can take months to achieve, creating excessive costs.

With this in mind, AWS vice-president of machine learning, Swami Sivasubramanian used his keynote speech at AWS re:Invent to announce new tools that aim to speed up operations and save costs. Sivasubramanian went through five tenets for machine learning that AWS observes, which acted as vessels for further explanations of use cases for the new tools.

Providing firm foundations

Firstly, Sivasubramanian explained the importance of providing firm foundations, vital for freedom of creativity. The technology has provided foundations for autonomous vehicles and robotic communication, among other budding spaces. One drawback of machine learning, however, is that a single framework is yet to be established for all practitioners, with Tensorflow, Pytorch and Mxnet being the main three.

AWS SageMaker, the cloud service provider’s machine learning service, has been able to speed up training processes. During the keynote, availability of faster distribution training on Amazon SageMaker was announced, which is predicted to complete training up to 40% faster than before and can allow for completion in the space of a few hours.

How Kubernetes extends to machine learning (ML)

This article explores the ways in which Kubernetes enhances the use of machine learning (ML) within the enterprise. Read here

The shortest possible path to success

From preparing and optimising data and algorithms to training and deployment, machine learning training can be time-consuming and costly. AWS released SageMaker in 2017 to break down barriers for budding data engineers.

Following its predecessor, SageMaker, Data Wrangler was launched during re:Invent to accelerate data preparation, which commonly takes up most of the time spent on training machine learning algorithms. This tool allows for the preparation of data from multiple sources without the need to write code. With more than 300 data transformations, Data Wrangler can cut the time taken to aggregate and prepare data from weeks to minutes.

To then make it even easier for builders to reach their project goals in the quickest time possible, the Sagemaker Feature Store was launched, which allows features to stay in sync with each other and aggregate data faster.

Sagemaker Pipelines is another new tool which allows developers to leverage end-to-end continuous integration and delivery.

There is also a need to understand and eradicate biases, and in response to this, AWS announced Sagemaker Clarify. This tool works in four steps; by detecting bias during analyses with algorithms before delivering a report which allows steps to be taken; models are checked for unbalanced data, and once deployed, a report is given for each input for prediction, which helps to provide information to customers. Bias detection can be carried out over time, with notifications being given if any bias is found.

AI bias: Why it happens and how companies can address it

As artificial intelligence becomes more prevalent throughout business and society, companies need to be mindful of human bias creeping into their machine models. Richard Downs, UK director at Applause discusses how businesses can use the wisdom of crowds to source the diverse set of data and inputs needed to train algorithms. Read here

John Loughlin, chief technologist in data and analytics at Cloudreach, said: “The Clarify product really caught my eye, because bias is an important problem that we need to address, so that people maintain their trust in these kinds of technology. We don’t want adoption to be impeded because models aren’t doing what they’re supposed to.”

Also announced during the keynote was deep profiling for Sagemaker Debugger, which allows builders to monitor performance in order to move the training process along faster.

Expanding to more builders

With the aim of making machine learning accessible to as many builders as possible, SageMaker Autopilot was introduced last year to provide recommendations on the best models for any project. The tool features added visibility, showing users how models are built, and ranking models using a leaderboard, before one is decided on.

Integration of this kind of technology for databases, data warehouses, data lakes and business intelligence (BI) tools were referred to as future frontiers that customers have been demanding, and machine learning tools were announced for Redshift and Neptune during the keynote. While capabilities for Redshift make it possible to get predictions for data warehouses starting from a SQL query, ML for Neptune can make predictions for connected datasets without the need for prior experience in using the technology.

Brad Campbell, chief technologist in platform development at Cloudreach, said: “What stands out when I look at ML for Redshift is that what you have in Redshift, which you don’t get in other data sources, is the true composite of your business’s end-to-end value chain in one place.

“Typically when I’ve worked in Redshift, there was a lot of ETL work to be done, but with ML, this can really unlock value for people who have all this end-to-end value chain data coalesced in a data warehouse.”

Another recently launched tool, Amazon Quicksight ML, provides stories of data dashboards in natural language, cutting the time spent on gaining business intelligence information from days or weeks to seconds. The tool takes into consideration the different terms that various departments within an organisation may use, meaning that the tool can be used by any member of staff, regardless of the department they work in.

Kevin Davis, cloud strategist at Cloudreach, said: “There is another push in this area to lower the bar of entry for ML consumption in the business space. There is a broadening of scope for people who can implement these services, and a lot of horizontal integration for ML capabilities, along with some deep vertical implementation capabilities.”

How artificial intelligence and machine learning changed the SaaS industry

Yair Green, CTO at GlobalDots, explains how artificial intelligence and machine learning changed the Software-as-a-Service industry. Read here

Solving business problems

Without considering problems that the business needs to solve, no project can be truly successful. According to Sivasubramanian, any good machine learning problem to focus on is rich in data, impacts the business, but can’t be solved using traditional methods.

AI-powered tools such as Code Guru, DevOps Guru, Connect and Kendra from AWS allow staff to quickly solve business problems that arise within DevOps, call centres and intelligent search services, which can range from performance issues to customer complaints.

During the keynote, the launch of Amazon Lookout for Metrics was announced, which will allow developers to find anomalies within their machine learning models, with the tool ranking them according to severity. This ensures that models are working as they should be.

“The caveat I have around Lookout for Metrics is that it’s clearly directed, and intended to look at the most common business insights,” said Davis.

“In terms of generally lowering the bar of entry, you can potentially put this in the hands of business analysts that are familiar enough with SQL queries, and allow them to directly pull insights or anomalies from business data stores.”

For the healthcare sector, AWS also announced the launch of Amazon Healthlake, which provides an analysis of patient data that would otherwise be difficult to make conclusions on due to its usually unstructured nature.

Commenting on the release of Amazon Healthlake, Samir Luheshi, chief technologist in application modernisation at Cloudreach, said: “Healthlake stands out as very interesting. There are a lot of challenges around managing HIPAA and EU GDPR, and it’s not an easy lift, so I’d be interested to see how extra layers can be applied to this to make it suitable for consumption in Europe.”

RPA and the rise of intelligent automation in healthcare

Andrew Pellegrino, director of intelligent automation at DataRobot, analyses RPA and the rise of intelligent automation in healthcare. Read here

Learn continuously

Just as algorithms need to be learned so that tasks can be automated effectively, the final tenet of ML discussed by Sivasubramanian calls for companies that deploy machine learning to encourage their engineers to continuously learn new skills and technologies, if they aren’t doing so already.

AWS has been looking to educate the next generation of builders through its own Machine Learning University, which offers solution-based machine learning training and certification, and where budding builders can learn from AWS practitioners. Learners can also develop skills specific to a particular job role, such as a cloud architect or cloud developer.

Furthermore, AWS DeepRacer, the cloud service provider’s 3D racing simulator, allows developers of any skill level to learn the essentials of reinforcement learning, and submit models in an aim to win races. The decision making of models can be evaluated with the aid of a 1/18th scale car that’s driven by machine learning.

Related Topics

AWS
Machine Learning