Intelligent CIO Europe Issue 15 - Page 57

////////////////////////////////////////////////////////////////////////// FEATURE: MACHINE LEARNING Using supervised Machine Learning to reach a desired solution A s well as utilising ML for cybersecurity purposes, it is also widely used across a number of industries and can help to decipher between documents and data within the workplace. We spoke with Sascha Eder, COO, NewtonX, who describes how supervised Machine Learning technology works and Machine Learning’s role within society. NewtonX, an AI-Powered Knowledge Access Platform, connects clients with experts using a unique set of proprietary automation tools. Where traditional expert networks rely on a finite number of pre-onboarded experts, NewtonX leverages a proprietary search algorithm to identify and onboard the best possible experts in the world for each specific request – whether or not those experts are already part of the NewtonX network. Supervised Machine Learning refers to a certain type of algorithm which is able to learn through a supervised process. The algorithm is fed labelled training data in successive rounds to progressively reach a state where it can recognise this data independently. It’s called supervised learning because during training, the human ‘supervisor’ knows the output that they want the machine to reach and corrects the algorithm as it tries different paths to reach the desired output. The inspiration behind the supervised ML technology The inspiration derived from the fact we were involved in an industry that consisted of indexing data (knowledge, skills, experience), but accessing this data was still mostly manual. We realised we could automate aspects of the process and leverage Machine Learning to make our search engine increasingly precise. Our first step in pursuing this was building the NewtonX knowledge graph which is a data model we designed to be able to map the knowledge of each of our NewtonX experts. This data architecture uses nodes and vectors to represent knowledge rather than a traditional database or tagging system to improve precision. That allowed us to structure data in a way where each area of expertise is defined by a set of nodes and vectors which then define complex knowledge. Creating the technology Myself and my team started it very manually which often, in most successful ML algorithms, is the underlying process – you manually create the supervised set of data that you then manually define. We looked at roughly 5,000 CVs and defined all the different areas we thought were relevant, including job titles, industries and areas of expertise. We initially focused on areas of technology, for example, general IT, cloud computing and mobile IT. We defined all these categories and sub-categories and every single profile we studied, we categorised based on all the different variables. We put them into relationships and built a database around defining these profiles and then started using a broader database of 50-100,000 CVs that we tried to map to the previously defined profiles we’d created. This is based on where we built the algorithm that was automatically defining which category the profile would fall into. We would then have a feedback loop where we’d go back in and see what level of precision was used and whether the profile was categorised correctly. The main industries utilising the offering The main industry we’re working with currently is the consulting industry – management and strategy consultants – to help them receive a lot of information to their product line. We work with private equity, asset management and hedge funds which need information to make investment decisions. We also work with more traditional market research firms that are gathering reports, running surveys or doing in-depth interviews with their clients. So, in summary, companies where a large amount of data handling is required. A successful client deploying the technology We’re currently working with a large investment bank that has a massive employee base, with thousands of very experienced people and a lot of finance specialists. The bank was struggling to harness the power of these different experiences and connect employees to each other in real-time to share their expertise. We are currently working with the company to rollout our technology to its own employee base. The client is now able to access real- time expertise within its employee base. Machine Learning enabling technological development One benefit of ML is the automation of functions that previously only humans were capable of. People can now focus on higher value tasks that are more difficult to automate, in order to provide higher value for clients, which in turn increases company value. In DevOps, for instance, ML enables the automation of tech cases, optimisation of load balances and improvement of IT infrastructure. ML’s role within society moving forward One of the more popular cases where Machine Learning is impacting society is autonomous driving. A common complex problem the automotive industry attempts to solve is extremely unpredictable environments and decisions often need to be made in micro seconds. Another example is the finance industry – by using ML, you can gather much more information about people that would qualify them for a certain credit type and so forth. More generally, nearly all industries are being disrupted by ML, meaning we’ll be able to make better decisions, predictions and have improved transparency and applications for society. Despite the benefits of ML, it’s important not to disregard the potential for a high level of abuse. Concern surrounding the amount of data which needs to be leveraged and required to run ML operations is something people might not want to reveal. There is an undeniable impact on society, whether it be good or bad, in terms of there being a lot of automation, decision- making and new industries emerging that weren’t around 10 years ago, like autonomous driving. n INTELLIGENTCIO 57