Machine Learning basics for Business Applications and Career Growth

 “Technology disrupts many jobs but it creates new jobs with more efficiency and consistency to become the lifeline for economic growth of any country”.

Machine Learning (ML) is the most exciting field where many experts of multi-disciplines, collectively evolve required algorithms and carry out exhaustive numerical processing, analysis and interpretation. ML is not straight software programming but the very iterative process of training a machine to respond to emerging situations. It is like teaching a small child to store various options in his/her brain and when required, respond very judiciously. A similar concept is being followed by leading manufacturers of cars for faults diagnostic and analysis, medical labs for testing and diagnostic of the disease. A traditional automation has an input to a computer resident program which processes input and gives output for further human analysis and interpretation. In the case of ML, there is input and expected output both acting as inputs to the computer where software and algorithm are resident, to process inputs and generate a program which does the final task. ML is a very laborious task to cater for all types of situations/constraints and make the computer action autonomously and which is very reliable.

The evolution of Big Data supported by enormous computing power and reach of cloud computing has given tremendous momentum to business decision making. Both the quantum of data available and processing speed continue to rise exponentially. Consequently, the ability of machines to learn and support intelligent decision making has increased manifolds. Although, ML may identify previously unidentified problems to be solved, yet the machine is not autonomously creative and decision maker. The computing machine will neither spontaneously develop new hypotheses from data under process nor it can determine a new way to respond to the emerging stimuli. The output of ML is entirely dependent on the data and algorithm and any change in data will change the result. At present, AI is based on ML, and ML is essentially different from statistics. It must be recognized that ML relies on algorithms to analyze huge datasets and it can't provide the type of AI that the movies present like reconfigurable Robots fighting in autonomous mode. Even the best algorithms can't think, feel, present any form of self-awareness like humans, ML can just perform predictive analysis much faster than any human can. Hence ML can help humans to work more efficiently but cannot replace humans.

Scope for positive use ML AI and ML are already employed in various weapon systems, surveillance systems and many other applications to defeat the adversary. In this article, the scope is to make you understand MI concepts and its application in a positive sense, which help the society and not has any negative effect. Some people advocate that AI and ML are job snatchers and resist their application. They are only partially aware of ML whereas ML also offers  many new job opportunities  and productivity   enhancement of both man and machine

Definitions.

•       Machine. It is a mechanically, electrically, or electronically operated device with large memory and very high processing speed to perform an assigned task.
•       Learning. It is a process of gaining knowledge or skill by studying, practising, instructions being given or experiencing something by doing.
•         Machine Learning. It is an integration of machine and knowledge where decision making logic in the form of a comprehensive algorithm is placed in the machine as embedded software.

How do machines learn? Machines learn by studying data to detect patterns or by applying known decision-making rules to:

• Categorize customers, vendors, materials, financial potential and shopping habits, lifestyle, likes and dislikes of people.

• Predict likely outcomes based on identified patterns, business practices and statistical rules. It can also predict waiting/processing time at a facility and do optimal rescheduling.

• Identify unknown patterns and relationships among peoples to forecast emerging trends /risks.

• Detect anomalous or unexpected behaviour of people and overcome uncertainty.

Algorithm. The real heart of ML is embedded software (algorithm) developed and proven by a multi-disciplinary team of experts. Algo­rithms are set of formulas or equations or matrices or expressions indicating various constraints and range of values with acceptable tolerance or error. These mathematical expressions are numerically solved by very high-speed processors, data analysed and the outcome is made to go through a number of iterations until it is close to the expected result. Different algorithms learn in different ways. As new data regarding observed responses or changes to the environment are provided to the computing machine and progressively the algorithm’s performance improves. This process increases “Machine Intelligence” over time.

Validation  Criteria. ML is a synergistic exercise between a person and machine. In real life applications, ML requires the human application of the scientific method and human communication skills. The process of selecting, auditing and fine-tuning an algorithm to deliver reliable results is very critical to MP application. However, we should not confuse ML as a black box activity. Human knowledge and machine must draw synergy from each other. ML by itself can predict emerging trends as per algorithm but cannot give the final conclusion. For validation of algorithms many factors must be considered as indicated below:   

•     We must know what are we trying to predict. ML project should start with a clear statement of the problem and the hypothesis to be investigated.
•          Be sure if resulting correlations are predictive or  Causal or there any inherent biases.
•          Be clear about exceptions conditions to be addressed and what the acceptable limits are.
•    Selection of the appropriate  algorithm is crucial for problem-solving and  data under investigation
•        Identify what data features should be included in formulating the algorithm.
•      Data scientists and subject matter experts must work together to figure out the sources of data and the key features of the machine. Data visualization can play a key role in helping to highlight and test features that can be fed into ML algorithms.
•          Can the data be pre-processed, cleansed, transformed for ease of processing and better results?
•           How should the algorithm’s parameters be fine-tuned for optimal performance?

Applications of ML.Today the corporate world is better equipped with market intelligence like customer’s buying capacity, frequency, preferences and lifestyle. This comes from fast analysis of shopping data and social media as to why customers of a particular country, community, gender, age and financial status buy their products, use their services, or engage their expertise.  Earlier, it has been done through company’s business experts and in-house IT resources. With today data deluge in social media, data is really big in terms of its velocity, variety and Volume ( V3) and its fast analysis needs assistance from computing machines. With enormous data storage and processing power available, computing machine can quickly detect purchase patterns and preferred channels partners to support the sale of products. Imagine you are driving on the freeway and GPS + Intelligent machine can use historical and real-time data to deter­mine that you are fond of a particular brand of coffee, send you a message, that your favourite coffee shop is around the corner.  You will be very much excited and thankful for free and personalised info reaching you just in time. This happens by integrating GPS and Big Data being handled on the 24x7 basis by powerful computing machines. ML is particularly suited to problems where -

·         Applicable rules/constraints cannot be easily codified by simple logical rules,

·         Accuracy is more important than interpretation or interpretability,

·         The data is problematic for traditional analytic techniques.

•          Health Care Services. ML can help to discover what genes are involved in specific diseases. ML Based on the patient’s genetic information, demographic information ML can also determine which treatment will be most effective for an indi­vidual patient. It is a well established that every human being has more than 20,000 genes. A high degree of varia­tion within each of those 20,000+ genes adds to the complexity to data set for analysis. It is well recognised that genes in isolation may not predict health outcomes or disease expression. Biochemical, environmental and other factors must also be considered, thereby requiring integrated data from multiple, sources. Earlier scientists were using super-computers for analysing data related to genetic disorders for various diseases on persons and their relative going back say 2-3 generations. Many times past records were not available for drawing any co-relation. Faster and deeper analysis of vast data using ML has opened a new horizon for genetic analysis and interpretation.
•          Driving assistance.Unconsciously we all use dynamic programming technique of Operations Research (OR) while navigating our car through busy routes to reach from point A to D. As GPS keep  giving  us near real-time  traffic status  about road conditions, traffic jam on a bridge, roadblock due sone vehicle accident, road  under  repair  and so on ,we  first  slowdown  and either wait or switch to alternate  route to reach our destination on time. We do this on the fly and intuitively make quick decision to select our route and carry on.  With the availability of new technologies, ML can identify the best routes from point A to D, consider weather conditions, travel time and predict the best route based on current location and emerging road conditions. Car driving assistance is the most common, simple and useful application of ML.
•         Driverless Car. After Google had launched its driver-less cars in mid-2017, many such cars /buses have come on the road in a few countries.ML can drive a car without requiring input from a driver.  This is a great boon for senior citizens or those who can not drive a car effectively. Driving is a complicated but well-bounded problem. There are, in fact, a limited number of actions a vehicle may take: start, stop, go forward, go backward, turn, speed up and slow down. However, the decision to take any of action is influenced by numerous factors including but not limited to road conditions, weather conditions, presence and behaviour of other vehicles, two-legged persons and their four-legged friends, and the rules of the road – just to name a few. While a human driver instinc­tually assesses all these inputs on the fly, capturing discrete rules for every possible combination is impossible.
•      Deep Learning (DL). A modern, advanced machine learning technique that makes use of extremely sophis­ticated neural networks is Called Deep Learning because the models generated are significantly more complex or deep than traditional neural networks. It follows the concept of Reinforcement Learning(RL) which is equivalent to teaching someone to play a game. The rules and objectives of the game are clearly defined. However, the outcome of any single game depends on the judgment of the player who must adjust his/her approach in response to the skills and actions of the opponent. This is a responsive or dynamic game where the outcome of every situation acts as an input to the next situation while the game is being played by two opposing persons or two teams. DLengages many ML systems and has further improved our ability to understand analyse and interpret an image, sound and video. This has been made possible by various advances in ML research and availability of very large data and very fast computing machines.
•  ``` Cognitive Computing. CC relates to systems and processes which attempt to understand and emulate human behaviour. It also provides an intuitive interface between a person and machine. Cognitive computing employs many techniques including NPL, advanced ML algorithms (including DL) and Natural Language Generation (NGL). CC makes machines (software systems) more accessible and intu­itive. As a result, CC is the key to increasing adoption of automated systems and analytic solutions. This will pave the way to a computing environment where a person and machine seamlessly work together.
•          Natural Language Processing (NLP). NLP has the ability to translate a language into a form that a machine or algorithm can interpret. NLG allows the machine to communicate results in “plain English” (or any other language it can support). Some NLP tools simply perform translation, mapping the words in a command to a system dictionary.  NPL is widely used by service providers like Google to allow users to select their convenient language for communication on the web.
•      Merchandisers. ML can be used for real-time online pricing. In this case, the machine determines optimal price points while humans validate gross thresh­olds. A feedback loop allows the algorithm to learn from observed sales results. The model also takes buyer input on missed opportuni­ties or errors. This seemingly simple shift required a huge change in how merchandisers and buyers were measured and grade. In Amazon’s automated distribution centres, humans do the packing while robotic systems collect required supplies and validate the right stuff gets in the right box. The interaction between human and robot. Underscoring the point that deliberate design of not just the algorithm, but the ongoing engagement between man and machine, is critical to success.
•          Insurance. The jobs/tasks so far being done by the middlemen (insurance brokers or insurance agents) can be done now more efficiently by using Big Data, AI and ML. Such new analytic tools will automate determining the total amount a person is qualified for insurance and for what period. It will also compute premium to be paid and terms of payment. Hence the decision-making process for issuing an insurance policy or clearing various claims/bills will be speeded up. This will result in reducing manpower but improve customer satisfaction.
•          Journalists. Many routine jobs of journalists, press reporters and editors can now be automated using AI and ML tools. Initially, machine capability will be used in sports reporting, road conditions, traffic jams, weather reporting, which rely heavily on data and numbers. As more people will use Smartphone and wearable devices, they will get timely, accurate and consistent information. This will cut on jobs of press reporters moving around in villages and towns for collecting information.
•          Financial industry. In most of the banks, tellers have already been replaced by ATMs and very soon even loan officers and account managers could be easily replaced by automated systems. Software algorithms can quickly analyze financial data and prepare accounts statement or balance sheet without the need for accountants. Each individual can easily prepare his/her Income Tax statement and submit online without paying a fee to the Charter Accountant (CA). Even Income Tax departments of government are now using Big Data, AI and ML to check tax returns of corporate as well as those of individuals and identify potential fraud cases.
•          Search and Aid Attorneys. During the initial phase of any lawsuit, lawyers and Para-legal staff have to sift through thousands of documents, depending on the complexity of the case. Now with help of appropriate technology like Big Data, ML and tools for syntactic analysis will speed up sifting through large documents. The introduction of software like “Discovery” capable of analysing large volumes of legal documents have been expected to reduce the number of legal clerks and paralegals, who act as human search engines. It is likely that with the ML system legally trained to review past documents, precedents and case history, can draft a legal brief for the senior advocate who is handling the case. A statistical model using ML could predict the outcomes of major legal cases and make many highly paid legal experts redundant.

•  Digital Manufacturing .Manufacturers are applying ML to identify potential equipment failures just in time before they happen. This requires fitting equipment with sensors and embedding analytic sensing systems. It will impact customer service, maintenance and warranty policies and procedures. ML will enable the digital exchange of product-related information between design and manufacturing groups. ML will considerably reduce design effort and production time. It will also provide efficient documentation for version control and easy traceability of defects.

•        Fraud  Detection.  All banking systems are already using AI to detect any fraud in using your credit card. You get an email/SMS message from your credit card bank asking whether you made a particular purchase using your credit card. The bank is alerting you that someone else could be making a purchase using your card,. The Al embedded within the credit card bank’s code detected an unfamiliar spending using your credit card and alerted you to it.   

•         Animal protection: The Ocean might seem large enough to allow animals and ships to cohabitate without a problem. Unfortunately, many animals get hit by ships each year. A machine learning algorithm could allow ships to avoid animals by learning the sounds and characteristics of both the animal and the ship.
•          Predicting wait times: Most people don't like waiting when they have no idea of how long the wait will be. ML allows an application to determine waiting times based on staffing levels, staffing load, the complexity of the problems the staff is trying to solve, availability of resources, and so on.

Issues and concerns. 

•  Lack of standards.  As you all know that AI and ML have both been around long enough to create specifications, but you currently won't find any standards for either of these.
•   Insecurity of Jobs.  Many fear that AI and ML both are job snatcher and must stay away.  This is not really true as both AI and ML offer new job opportunities which of course need new skills. 
•   Lack of trust in the outcome. The basis ML is math where algorithms determine how to process big data and interpret its outcome. You discover that algorithms process input data in specific ways and create predictable outputs based on the data patterns. What isn't predictable is the data itself. The reason you need AI and machine learning is to decipher the data in such a manner to be able to see the patterns in it and make sense of them.

Adopting  ML. Rapid developments in technology and increased use of AI, Robotics, Digital manufacturing, ML as these have enhanced productivity and quality. Therefore, you must learn new technologies faster, be proficient and become highly productive. To stay current and deliver results, ML algorithms must be continuously refreshed and refined based on data that reflect current circumstances. This is true whether evaluating the impact of new customer segments or retention or rebalancing the power supply network to account for unexpected spikes in energy demands and avoid blackouts.  Computer Professionals who implement algorithms use programming language that works best for the task.  ML  relies on programming languages like  Python and R, and to some extent on  Matlab, Java and C++.

If you apply the ML model successfully, you can account for customer behaviour, lifestyle, buying capacity, shopping habits and frequency of shopping. ML can pre-process a lot of data and predict the outcome for subject experts/professionals to utilise that outcome to speed up their decision making. With help of ML you could be encouraging consumers to buy more, utilize different commercial channels, user-friendly navigation and driving assistance. For successful implementation of ML, it is not a merely to keep all interconnected devices on but also to keep updated/upgraded various algorithms, hardware, embedded software, application software and data interfaces.  

For   more information, refer to my book- “Career Challenges during Global Uncertainty”, available on www.amazon.com
Sarbjit Singh, PhD, Former E]xec Director Apeejay College of Engineering, Gurgaon, Haryana, India