Skip to main content

Once more into the deep

(Image credit: Image Credit: John Williams RUS / Shutterstock)

Hellerstein explained: “The story we have seen over the last couple of years has been the practical success of Machine Learning at a surprisingly wide range of tasks. One fascinating aspect in that story is the use of custom hardware to accelerate it, including the adaptation of the use of graphics processing units plus open source software such as Google Tensorflow, Berkeley’s Caffe and Facebook Torch.”

The combination of human and artificial intelligence (Machine Learning) working together like this in enterprise settings has, as a result, seen major breakthroughs in data wrangling – the formerly laborious, time consuming and therefore pretty costly process of converting raw data into forms which can then be analysed.

Hellerstein gave an example of one of Trifacta’s customers, PepsiCo:

“Their analysts used to spend literally months manually building a single forecasting report to determine the right quantity and mix of product needed to ship to retailers.

“Trifacta, which relies on Machine Learning to continually learn and then guide analysts with increasingly intelligent suggestions, has helped PepsiCo’s reduce preparation time by up to 90%.

“So now, instead of spending the bulk of their time standardising and piecing data together, analysts focus on the more interesting, challenging issues of analysing the data and telling the right story. This in turn has driven new insights such as more effective shipping use from better analysis of the weight of different product mixes, and therefore saving the company untold dollars.”

Two key factors in the evolution of the practical use of Machine Learning are increasing computational power and the increasing availability of data. As such, big data projects are the places where it often has the most impact.

Many companies have big data projects underway, and those that apply Machine Learning, combined with human understanding, are able to radically drive down the time it takes to prepare data and get to the direct applications for their business – we’re talking from several weeks to just a few hours.

With all of the promise, Hellerstein also highlights challenges posed from Machine Learning techniques.

“One of the biggest challenges with machine learning techniques is that many sophisticated techniques are not explainable: the decisions made by algorithms cannot be explained naturally to humans. For software engineers, techniques that are not explainable are inherently more difficult to debug. So for users, it can mean that some machine-learning-based software is unpredictable and there is potential for disruption. It is therefore critically important that machine-learning-based software include carefully-designed user interfaces that allow human intelligence to observe, steer and override machine learning algorithms when they get off track,” he warns.

Big impact

“However most of the successes of machine learning have been in technical domains like data analysis, where we have a shortage of human resources. So the biggest impact of machine learning in the near term may be to increase employment by lowering the technical barrier-to-entry for these tasks, enabling more people to work effectively with data without having to become deep experts in computer science.”

As for the future of Machine Learning, Hellerstein said he expects to see it impacting more on the internal processes and tasks of business such as improving data bottlenecks; improving organisational communications by email prioritisation and bot-assisted group chat and improving auditing and compliance.

“By uniting all data sources into a digital core, the IT infrastructure is prime for automating data analytics and maintenance. Machine learning can find more efficient, cost-effective scenarios to operate complex systems based on quality data without risking disruption and noncompliance. With this we are on the path to a self-learning Enterprise System, there is a clear demand for greater integration across the enterprise, to an Intelligent enterprise system that assists in the core activities of a company and takes decisions for repetitive end-to-end tasks.

“In terms of data analysis and extracting valuable information from business data - leveraging huge stores of data was previously deemed too complex or too messy to mine for value. With Machine Learning, it will be as easy to extract useful information from pictures and recordings in the future as it is from clean tables of well structured data today.

“This will radically change the types of services companies can offer. Today you can search through a CRM system filled with extremely valuable, but often painfully collected, information. In the future you'll be able to simply search recorded audio from customer support calls or databases of images collected during insurance claim filings. Once you shift to a world where direct search on these complex data sets is possible - things get very interesting indeed,” he added.

Machine learning is clearing pathways to businesses growth, process optimisation, as well as daily employee empowerment. By automating redundant and low-value activities, companies are addressing changes in real time and delivering the best-possible outcomes.  Extending this further they are moving to a deeper emphasis on integrated intelligent systems, leveraging collaborative workspace tools and therefore enabling greater efficiency.

And as far as moving into the future, Hellerstein went on to conclude that as the process continues to evolve, businesses will innovate cutting-edge applications and use cases that could drive increased efficiency, intelligence, agility, and customer-centricity. However, he suggested that those that move their IT architecture to the cloud stand a better chance to get ahead of the competition and create a wave of disruption that sets the stage for market leadership.

By allowing company-wide access to the right data anytime and anywhere, employees can better follow processes, truly understand customer needs, and respond to market dynamics. More importantly, the entire workforce – regardless of role and organisation – would be encouraged to collaborate, to leverage and realise the full power of machine learning to build a stronger bottom line.

Joe Hellerstein, Co-founder and CSO, Trifacta
Image Credit: John Williams RUS / Shutterstock

Joe is Trifacta’s Chief Strategy Officer, Co-founder and Jim Gray Chair of Computer Science at UC Berkeley. His career in research and industry has focused on data-centric systems and the way they drive computing. In 2010, Fortune Magazine included him in their list of 50 smartest people in technology, and MIT Technology Review magazine included his Bloom language for cloud computing on their TR10 list of the 10 technologies “most likely to change our world”.