Enhancing efficiency via machine learning

Mumbai: Gurgaon-based employability assessment company Aspiring Minds is perhaps best known for giving the industry a very dim view of the quality of engineers in India. According to its 20 January report, “more than 80% of engineers in India continue to be unemployable”.

Aspiring Minds, however, does much more than just track the employability of engineers in the country. In the words of its co-founder and chief technology officer (CTO) Varun Aggarwal, Aspiring Minds is “interested in the big picture”. “We ask questions like, ‘How do we identify what jobs people in the job market will be successful at?’; ‘How do we make this assessment, or automatically assess programming skills?’; ‘Or, for that matter, gauge how well a candidate speaks English?’”

But such questions typically warrant predictive models, for which Aspiring Minds uses machine learning technology that allows companies to build such models using structured and unstructured data.

A machine learning system does not need programming and can teach itself from mountains of data. It is a subset of artificial intelligence, which itself is a collection of technologies and concepts. Deep learning, which uses neural networks that are loosely modelled on the human brain, can be said to be a variant or even a subset of machine learning. deep learning, thus, takes machine learning closer to artificial intelligence (AI).

Aspiring Minds, on its part, uses algorithms powered by Machine Learning that draw on structured data to address complex issues. One of its products called SVAR uses advanced speech recognition and Machine Learning algorithms to accurately gauge the quality of speech in various accents against a neutral accent. SVAR, according to Aggarwal, “has helped companies improve recruitment efficiency by over 35% and reduce voice evaluation costs by 55%”.

To train the Machine Learning algorithm, Aggarwal’s team had to initially collect about 1,000 speech samples that included “good, bad and average pronunciations”. The company then got humans to rate the samples for fluency in English. “We then ran the machine learning algorithm on this data so that it could learn from it. Please remember that speech is unstructured data. So you have to extract some numbers from it to make it structured data, before you run the machine learning algorithm on it,” Aggarwal explained.

Aspiring Minds has another tool it calls Automata, which “is a programming assessment solution that uses machine learning for grading programs”, he pointed out.

Other than Aspiring Minds, many companies globally and in India, including some start-ups, are using Machine Learning tools to infuse intelligence in their business by using predictive models. Popular machine learning applications include Google’s self-driving car, online recommendations from e-commerce companies such as Amazon.com and Flipkart.com, dating site Tinder.com and streaming video site Netflix.com.

Closer home, a site like inclov.com—a matchmaking app that focuses on people with disabilities and health disorders—uses “artificial intelligence” to intelligently map its users, according to Shankar Srinivasan, the co-founder of Inclov.

Credit scoring and offers are based on machine learning applications and so are new pricing models, email filtering and pattern and image recognition.

Microsoft’s Face API, a case in point, can detect up to 64 human faces in an image. Facebook’s DeepFace uses technology designed by an Israeli start-up called face.com, a company that Facebook acquired in 2013. The software, developed by the Facebook AI (artificial intelligence) research group, uses a deep learning neural network—software that simulates how real neurons work. Google Inc. has developed a similar deep neural net architecture and learning method that also uses a facial alignment system based on explicit 3D modelling of faces.

On 9 May, Facebook spoke about FBLearner Flow, a piece of software that manages machine learning models for employees throughout the social networking company.

Indian IT services providers, too, have developed machine learning tools. Infosys Ltd has a machine learning platform, Mana, for companies in a bid to “drive automation and innovation”.

Tata Consultancy Services (TCS) Ltd’s retail solution, the Optumera Digital Merchandizing Suite, leverages the power of Big Data analytics, analysing heterogeneous data sources with advanced machine learning algorithms. In turn, retailers can localize and optimize store space, curate a shopper-centric omni channel assortment, respond in real time to competitor pricing strategies, and automate planogram compliance through image processing.

Tech Mahindra Ltd’s TACTiX (TechM Actionable Intelligence Extended) is an AI-powered platform with NLP (neuro linguistic programming) and machine learning capabilities to address operational optimization and digital implementation scenarios.

Mphasis Ltd has its ‘InfraGraf’, developed by Mphasis Next Labs. The automation platform, based on machine learning, pattern recognition and graph-theory-based algorithms is an infrastructure automation platform, and optimizes enterprise technology infrastructure investments and aids in strategic decisions-making.

Indian start-ups such as Arya.ai offer deep learning algorithms in language processing, computer vision, speech processing and reasoning for developers to build intelligent AI systems. Drishti, from AIndra Systems Pvt. Ltd, uses technology based on AI to provide hand-held devices with in-built cameras such as smartphones, tablets and laptops, and the ability to detect and identify people. Bengaluru-based Snapshopr Inc, a visual intelligence platform designed for retailers, helps online retailers improve their retention and conversion numbers using their image search platform. Users simply have to take a photo of what they like, following which they can instantly purchase it.

Bigger technology companies such as Microsoft Corp., International Business Machines Corp. (IBM), Dell Inc., Oracle Corp., Amazon Web Services Inc. and VMWare Inc. now want users to see the cloud as intelligent or smart, and one that can power advanced services such as machine learning, or using statistical models to analyse data, and the Internet of Things (IoT).

For instance, by using machine learning and other cognitive computing technologies, IBM scientists can generate solar and wind forecasts that are up to 30% more accurate than ones created using conventional approaches, whether minutes or days in advance.

Microsoft, on its part, has made its Azure Machine Learning offering available to developers. Microsoft has also been using machine learning in its products such as Bing, Office, Windows, Xbox, and Skype Translator, and now also to analyse security data from the Microsoft cloud with the help of its machine learning system that is capable of processing 10 terabytes of data every day, something that humans cannot do at this pace.

Meanwhile, machine learning is already helping doctors make better diagnoses. Advanced cyber-defence systems use machine learning that mimic the human immune system to learn autonomously, adapt to changes in corporate technology or its users, and spot nefarious activities. Researchers at the Technical University of Munich have even designed some machine learning algorithms that will answer the question: Which character is more likely to die next in the Game of Thrones serial?

And going forward, Facebook is building “AI algorithms that can help build AI algorithms”, according to a 6 May report in Wired magazine.

However, there are challenges to be overcome. In an 11 March interview to Mint, Una-May O’Reilly, principal research scientist at Anyscale Learning For All (ALFA) group at the Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory), pointed out that one major set of challenges of machine learning is “the software transformation of raw data into influential explanatory variables or sophisticated response variables to enable effective predictive modelling”.

According to O’Reilly, it takes a lot of work to close the gap, for instance, between an educational expert saying “a student is likely to drop out because she is procrastinating” and the definition and extraction of an operational variable that identifies how early a student started the problem set. Is an early or late start good evidence of “procrastination”?

The same question arises about relational trends: Is a patient in the lower decile acutely ill? Is a student’s current grade 10% below last month’s? She concluded, “Facilitating the natural and efficient translation of human conceptual descriptions into Machine Learning-ready data is both challenging and motivating because this is at the crux of human-data interaction.”