Deep learning, Meetup and a nice discussion

Our meetup at Princeton was full this week. On a snowy, super-cold Monday evening, more than 80 people showed up and stayed for almost two hours. What was the reason? It had to be the meeting topic - TensorFlow, one of the sexiest words among technology enthusiasts today. The meetup title was also juicy, "no reason Deep Learning has to be Scary". Sounds really attractive, isn't it? The words "Deep learning" may be simple to use, but the technology behind it is equally hard to describe. Most presentations quickly slant towards a scary blackhole of mathematical zigsaw puzzles. So, when we heard someone would tell us why "Deep Learning" doesn't have to be scary, we felt motivated right away. And that motivation was reflected in such a large audience turnout. People from a variety of industries came over and participated in a wonderfully lively discussion.

At the end of the session, it was time for networking. There were data science experts from pharmaceutical industries, some were from biotech, a few from finance and IT. One of them was a key executive of small financial services company. The discussion with him was very interesting. When I told him how an investment company uses machine learning to segment customers and fine tune campaigns to target specific customer groups, he asked if they used deep learning technologies. I said, “No, some other less weighty algorithm works just fine for them”. He said, “It means they are not using real machine learning tools. If you do not use deep learning, how can you claim to use latest machine learning tools?”. Then I realized, to many like him, machine learning means deep learning, nothing else. Like any other passionate techie, I felt a strong urge to correct his misunderstanding as if the future of the world depended on it. I spent next few minutes babbling about the evolution of machine learning and its various tools and techniques. As other folks chimed in, a multi-way conversation ensued. I doubt how much it changed the executive's thoughts on AI. But near the end, he asked me an interesting question, “I know these technologies are evolving rapidly. But I don’t have the time and energy to go to a formal training. Where can I get a reliable overview of AI-ML as it stands today? And it must be written from a business perspective, not in some cryptic techie language. Could you show an online resource that is authentic, complete, reliable?”. I got stumped. I did not have anything handy. I couldn't recollect the name of any online resource that could fulfill his criteria.

Later as I scoured through several sites one after another, I found most resources either partial (covering only some aspects of ML), or too tech-biased (the writer seemed more interested to show how deep his technical knowledge was), or kind of old (like those one-year old article describing two-year old materials). Then I found this one from McKinsey which I thought could come close to our expectation. It is structured for interactive exploration. The reader can pick what they are interested in and decide how deep they want to go. On each tool, it shows what it is, when to use it, how it works and what business case it addresses. All descriptions are short, just a few lines each. Starting with a timeline of AI evolution, it ends with latest advancements like reinforcement learning or recurrent neural networks. You can ask me, is it the best executive presentation available anywhere on AI-ML? Definitely not. But it is one of the most complete and executive-friendly ones out there. You could check it out and find by yourself how large is my bias or how far is it from our expected outcome. Here is the link:

An executive’s guide to AI

https://www.mckinsey.com/business-functions/mckinsey-analytics/our-insights/an-executives-guide-to-ai