How I'm planning to get started on Deep Learning?

This post is to describe how I find the book Meta Learning: How To Learn Deep Learning And Thrive In The Digital World is giving me a hope and road map to learn Deep Learning. Not only learning DL and also where to go from there viz Kaggle.

Who am I?

I’m part of Data landscape for more than a decade now. I started as ETL developer, then evolved into Data Engineer. When I was crawling the ETL world, the Data Scientists were already declared as the sexiest job, and Big Data was a novelty. The 1st time I heard about the power of the data and analytics was from the famous Forbes article on Target. I was awestruck. Then later on I started working on Big Data with AWS, as a Data Engineer.

It’s always a dream to get into the Machines Learning and AI. There are countless articles in the web, all of them stating maths is inevitable. The math we are talking about is not the friendly Algebra or Geometry, but the hostile Calculus, Probability and statistics.

First time hearing about Fast.ai

It was a TED talk by Rachel Thomas, shame the video got only 35k views. She talks about the power of AI and what the fast.ai can do. For the reasons unknown I didn’t went deeper into fast.ai. But, later on I tried to learn AI through padhai, it turned out to be a failure. I couldn’t completed the math assignment and told myself AI is not for me. Then, the work became the priority and thought about learning AI took the back seat, or got down from the vehicle.

Other venture with Data

Instead of AI/ML, I tried my hands with Data Visualization. Its good. I have few dataViz to denote blog post over the years, CWC23 result, Nifty50 charting and a harmless clock. D3 is good. JavaScript is powerful storyteller. I had a plan to continue with Data Visualization and data story telling as a viable future. See, I’m dealing with data from several different systems; cleaning & processing it is the day job. DataViz is a value add to my day job.

Things were going good until the arrival of ChatGPT. With its arrival LLM became the cool kid in the alley. Every Tom, Dick & Harry started chasing the LLMs. The employers too insisted us to get certified or trained on Generative AI. I’m of the opinion Generative AI is not the right entry point into AI/ML. Especially, prompt engineering looked like a snake oil. And so I’m confused whether to take up Generative AI or not.

“Deep learning worked” - Sam Altman

Sam Altman, the man who disrupted via ChatGPT, published a blog post - the Intelligence Age,

How did we get to the doorstep of the next leap in prosperity? In three words: deep learning worked.

For me who was already skeptical Generative AI, his words on DL was reassuring. Then, on reading more about it, I came to know DL is in play behind the working of GenAI. Thanks to articles on how CNN worksand scientists made DL possible. This made me to look for ways to learn DL.

Fast.ai again

Thanks to evil algorithms that keep us hooked, it suggested me to watch a talk by Jeremy Howard on Fast HTML. Yes, the same guy who’s behind the fast.ai. I went back to the site after years, what caught my attention is the section - is this course for me?

Just high school math is sufficient

I was relieved to see high school math is sufficient. On further reading I came across a book recommendation - Meta Learning. I tried reading the book in Kindle and realized this book involves several notes + external references. Then started reading the book at 9.30 pm and able to complete in by midnight. The book was so clear and concise, I read it by midnight, and took notes as well.

What is the takeaway from the book?

Below are the paraphrased version,

1. We all have limited time, replace time spent on other activity with learning DL.

2. Learning can be Theoretical & Practical. Fast.ai offers them.

3. Consistency is key. Build a habit to learn regularly.

4. Getting into flow state is important. Deep Work is necessary.

5. Practise. Practise. Practise.

6. Incremental changes on long run result in big changes.

7. Give back to community by blog posts, interact with community, Learn in Public.

   On Deep Learning,

8. Kaggle is the place to harness our learning.

   1. Join challenge at the earliest and tweak the solution as required.

9. Suggested additional courses like - Learning from Data.

10. Setting up GPU at home vs using AWS (or GCP’s resources)

Todo

1. Allocate a focus time to get started and be consistent.
2. Practise alongside the theory.
3. Blog regularly on the progress.