Intelligent CIO Africa Issue 43 | Page 47

FEATURE: MACHINE LEARNING FEATURE: of more consistent decision making and streamlined operations. Garnering enterprise momentum With Machine Learning seeing an increase in enterprise wide adoption in the MEA, pundits say CIOs and C-level executives need to be flexible when deploying new technology. Adam Pantanowitz, Co-Founder and Chief of Innovation at AURA, said there is no doubt that there is greater accessibility to and democratisation of Machine Learning, an upsurge and excitement around the technology. Pantanowitz added that the big cloud players have embraced it and made their tooling more accessible. “The open source community are very active and are providing accessible toolkits. In addition, the scientific community have made some great advancements which allows one to leverage knowledge gained in another domain by another researcher and their resources and move it to your problem domain.” He said that computation has certainly advanced, enabling more cost-effective access to the requisite processing power (GPUs and CPUs) to enable these applications. Pat Gelsinger, CEO (VMware), said Machine Learning is the most important new workload to emerge in the IT enterprise space in the last 10 years. Today, said Gelsinger, Machine Learning is seen as being the way of the future by many enterprises as it is deemed an effective and automated means to provide companies with a competitive advantage, driving innovation and ultimately increase profitability. “The main sectors utilising Machine Learning are healthcare, travel and hospitality, retail, finance and manufacturing. By using Machine Learning, a business can prevent fraud, provide dynamic pricing, effect predictive maintenance, leverage alerts and diagnostics as well as use real-time information to drive business decision making. This may be achieved by intelligent operations management, from applications to infrastructure, all by using dynamic threshold and capacity planning,” he said. Secret sauce With vendors often claiming to have some Machine Learning ‘secret sauce’ in their wares that will revolutionise an enterprise’s business how should CIOs and their IT teams go about selecting the right tools and systems? Dr Nicola J. Millard, Principal Innovation Partner, BT, reiterated that no amount of secret sauce will revolutionise a business unless its data is in order and of high quality. According to Millard, Machine Learning does not work by magic as it depends on data and if that data is unstable, inconsistent and spread among multiple legacy systems, it becomes more difficult and far more expensive to do. “Aside from data, enterprises also need to step back and ask what problem they want Machine Learning to solve and whether it will actually solve it. For example, if they want to deploy a chatbot to improve customer experience, does it actually improve it or just add another level of frustration for customers if the ‘bot’ hits a dead end and abandons them if it can’t understand what they want,” she said. Enterprises need to understand how Machine Learning will integrate with legacy systems, processes like in the contactcentre, how much training it will require and what the business case is given that return on investment (ROI) may take some time. Tony Bartlett, Director, Data Centre Compute at Dell Technologies South Africa, agreed with Millard adding that: “Firstly as with most systems, enterprises are advised to look to vendors who can provide an end-to-end solution. Machine Learning systems are part and parcel of a broader business strategy, which includes Digital Transformation, cybersecurity, Edge Computing, automation and data analysis to name a few.” Bartlett said enterprises should consider whether the vendor they rely on for Machine Learning solutions has the technology depth, breadth and specialisation to meet their requirements whether it be at the Edge, core or cloud. “As the volumes of data increases and systems are tasked with processing more data, in real time, IT systems will need to depend on high performance computing to keep up with the demands and technologies such as GPU’s, FPGA’s, high speed memory and storage are essential in deriving outcomes in real-time,” he said. Data quality With data and AI playing a crucial role in any Machine Learning deployment, experts are urging CIOs and business line executives to ensure data quality. Ramprakash Ramamoorthy, Product Manager, ManageEngine Labs, said your model is as good as the data you train on and spurious correlations, biases, data imbalance present in your data can adversely affect the quality of Machine Learning model’s prediction. “Organisations should track datasets and make sure they comply with local regulations and are free from inherent biases. Sometimes labelled data availability would be a challenge. For example, there are practically no commercial grade labelled datasets for service desk sentiment analysis, but you have a wide range of datasets available for e-commerce product reviews, hotel reviews and movie reviews,” he said. Ramamoorthy pointed out that transfer learning techniques can help bootstrap your small service desk dataset with the learning from the larger consumer datasets and whatever process used to generate and label datasets, will have to run fairness checks before deployment. INTELLIGENTCIO 47