Starting out with AI

I had the goal this year to get experience in cybersecurity, machine learning and site reliability engineering. My job at eShares has provided me with day to day opportunities to do the first and the last, so on my own time I decided to dive into machine learning and AI.

My background is mostly software and digital infrastructure operations but I also enjoy (and study on my own) writing, literature, poetry, philosophy and international politics. I like crossing disciplines and recycling what I learn and reusing it in new places of learning and so I thought I would bring all my interests to bear and see if they could be useful for this task.

The field is so vast and so foreign to me so, I started out like most millennials…I listed to a podcast. I found This Week In Machine Learning and AI which turned out to be good for me to understand the industry and trends but was not all the helpful for me to learn the basics.

Next step on my train ride was to Sentdex where I was able to dive into all sorts of concepts at all sorts of levels.

Whilst toodling around, I found this blog post on implementing a neural network from scratch which almost made sense to me and I stopped and thought about how I was going about learning about ML and AI and I wanted to try about learning ML and AI the way the machines learn about their tasks. Brute force.

I don’t know all the math behind the different ML algorithms that take in features and labels and spit out models but I understand that by throwing much data in the right format into these statistical models, they will shape themselves until the data makes sense to them.

Imma try that. Neural nets are modeled after our brains, so maybe learning like they do may be the quickest way to get a wide and deep understanding of a concept.

I can get a wide understanding from just podcasts and some articles and blog posts coming from the best in the business. And I can get a deep understanding by picking something that will be useful at work and learning all about it. Then implementing something with it and maintaining it and operationalizing it in a business context.

I think both can be useful, but I am fascinated with how to be a generalist and also good and/or great at something. I’m not under any deadlines and I’m not in school and I also like to meta-learn and I’m ok with failing. So I’m going to put out a plan, try it out and report back on its effectiveness.

My plan has a few steps.

  1. Start at a shallow and wide level and get a small sense of all there is in the field.
  2. Inundate myself with as many case studies of as many types of ML and AI as I can get my hands on until I have an understanding on how to approach most ML and AI problems.
  3. Find maybe five of the major types of ML and AI problems to solve and solve each of them in turn using a framework with as little work from me as possible. Then operationalize the models in some way so that I can consume them in the future.
  4. Do these same problems, but does them more manually. Use less of the frameworks’ helpers until I really understand what they are doing and how these frameworks are solving pain points.
  5. Observe until a problem arises at work that is ripe for an ML solution and implement a solution wholesale. Do the whole kit and kaboodle and learn business context, problems of scaling, learning the best ways to iterate on models.
  6. Do step 5 a few times if possible until I feel very comfortable with ML in practical settings.
  7. By this point I think that I’ll have enough understanding to have arrived at a mile wide and a half-mile deep. Then we’ll see if I want to dive even deeper and become a true expert in any piece of the field.

I have a feeling that my plan will be effective because it will utilize cross-disciplinary learning to recycle learning about one piece of the field to intuit things about the other. Doing that over and over again, digging deeper and deeper until I’m a mile wide and a half-mile deep.

I’ll publish along the way to share the resources that were the most relevant to my progress in learning and maybe I’ll discover what is the most important feature about these resources to help me predict if a new resource is going to be useful in furthering this learning.

I hope this becomes a blueprint for those after me on how to get into ML and AI.