Learning, Machine Learning

Very few people go to Goa and sit inside a glass building to learn about machine learning. But recently I did it along with a small cohort of a very diverse set of people.

 The hype around machine learning (ML), neural networks, and artificial intelligence is at its peak. Hardly a conversation goes by that does not mention these words many times. Even people who know little or close to nothing about it use it freely. It’s little like French wine you may or may not know anything about it, but if you can say Cabernet Sauvignon or Merlot, most people will not ask you the difference. Words are enough, even if you don’t know what it means during hype most people assume you do. The more you pepper your sentences with words like ML, AI the higher you resonate.

 Everyone may not know the difference between simple machine learning and neural learning. And to ask the difference is to be branded a neophyte. Which is anathema in a Google dominated world where you can know everything. But the hype is not new, Ray Kurzweil wrote about ‘Singularity is Near’ in 2005 more than 14 years ago, that is two generation ago. He said that machine intelligence will exceed human intelligence and biological evolution will getting a boost by technology. This concept of singularity has been talked about even earlier. Singularity as a term was used John Neumann in the 60’s. The history of predicting machines becoming intelligent has been actively fertilized by science fiction writers. Martin Ford, a futurist and author of ‘Rise of the Robots’ is the new cheer leader.

 If you talk to your android phone to get directions or search information you have interacted with a machine that is learning and responding to you in real time. The voice learning algorithms searching giant databases of Google is a classical ML tool that gets better in understanding your unique accent with usage.

Back to the ML workshop organized by CL Educate. Most people know CL Educate as a test prep company, but it is morphing into a skilling company. A crucial pivot indicating, at a critical intersection of two mega trends. Namely the collapse of higher education and the kinds of skills industry needs. CL is at cusp of a global trend, and it may drive, determine or change this trend much like Infosys did with the global offshoring and outsourcing trend. The degree of success will depend upon its leadership ability to build scale-able business model.

 The search for business model around ML also attracted three venture capitalists to the ML-workshop –Sanjeev Agarwal (Helion Ventures ), Gopal Srinivasan ( TVS Ventures) and Suman Nandy (CX partners). Not only venture capitalists but two women entrepreneurs Lalitha Srinivasan (TVS Electronics) Sujatha Kshirsagar (Dristikon). Munira Lokhandwala a mathematician and a quant whiz. Two IIT Kanpur Professors- B V Phani and Professor Rajat Garg. Entrepreneurs like NV Subramaniam, Nilesh  . Consulting guru Kamlesh Vyas and a Super cop Asim Arun from Uttar Pradesh. There was enough brainpower in the room to rival a Google data center.

The workshop was led by the inimitable Sujit Bhattacharya. Chief innovation Officer of CL Educate, who followed a simple formula. Explain the basic theory and make everyone practice writing the algorithm to see its impact on a data set. There was no introduction to the subject as everyone was expected to watch Andrew NG videos before coming to the workshop. Andrew has been teaching ML at Stanford University and is considered the top ML expert in the world. His video introducing ML can be found here .

Everybody, extrapolates from their experience with ML also everybody wanted to apply it to their own world and see if it could give them an edge. Asim Arun wanted to use it for detecting or identifying terrorist activity. Even identify vulnerable spots that they would target to help him plan the security in a state. Identify linkages between different terrorist groups using big data.

 Python is the language to talk to a machine learn and make it learn and sort patters from data. Python is much simpler than other computer languages as the syntax in it are much lesser and it is readable even by English literature graduate, with some effort. The way to learn is to read a line and see where the key directions are being given. It does require you to memorize its own grammar, and an understanding of statistical concepts is mandatory. Finally, learning to program a machine to learn is to make it see patterns in a data sets. Hence, everything needs to be converted into a dataset. An image to be seen by a machine has to be down into pixels, and each pixel becomes a box which is measured. The machine reads an image by combining all the pixels, the program has to help in interpreting the combination of all the pixels. This is how a machine reads handwriting or a face. The program helps it interpret the similarity by saying if the pixels are right in so many boxes than it’s correct or not. This computing power that is available to it

There are couple of things to know besides the dynamics of the machine learning. One, while Google and Facebook have an advantage of having data and are using machine learning every company can use its proprietary data to build ML hypothesis. The ability of the machine lies in learning in two forms a supervised learning format or unsupervised format. If you have lots of data you can throw it at the machine and it will learn to detect patterns in the data. Supervised learning is more structured and can be done with limited data for instance if you want the machine to learn about detecting a photo of an horse. You can give it enough photos of the horse labeled as a horse and then test whether it has learnt to detect a horse or not with a smaller test data.

The important thing is that this processing power is available to everyone if you have the data the processing can be rented. Some of the ML voice processing engines like Amazon’s Alexa are open for individuals. You can install an Alexa device at your reception in the office and program it answer specific questions about your company or availability of the people in the office and do away with the receptionist. Your imagination and the data is the only limit for training an Alexa device.

To give you a sample of what you should be learning in your first course on ML

Sequence of Sessions-

?Intro to ML

?Intro to Python, Pandas

?Data Visualization - Matplotlib

?Supervised Regression Problem

?Supervised Classification Problem

?Support Vector Machines

?Decision trees, Random Forests

?Deep Learning – Artificial Neural Networks

?Voice Bot Building

Online Sessions

?Unsupervised - Recommender System

?Unsupervised - Association Rule Mining

?Advanced Techniques in ML

?Recommendation systems

?Projects

–Visualization

–End to End Regression

–End to End Classification

--------------ENDs