Career pivot: applied ML 🤖

Traun Leyden

• 3 min read
Career pivot: applied ML 🤖
Generated by DALL-E: "The sky with floating computer brains"

I've recently decided to pivot my career into applied ML, more specifically applied Deep Learning.  I'm planning to document my journey, so if you're taking the same journey or interested in doing so, check my twitter feed for regular updates.

Rather than trying to juggle this with another job, I'm going to work on this full-time.  In some sense it feels like a big risk to burn through savings, but I'm hoping it will pay off in the long run.  

According to Reid Hoffman in his The Startup of You book, taking 6-12 months off to level up your "soft asset" skills and expertise in a new area can be a huge career boost! 🚀

My background

Over the last several years I've worked as a software engineer at high growth startups like Couchbase and Databricks.

While at Databricks I was working on the ML Feature Store product, a central part of the company's overall enterprise MLOps strategy. Databricks was an absolutely amazing place to work.  I grew tremendously as an engineer and had the opportunity to work closely with some of the brightest and nicest people in the industry.  

The hypothesis for the pivot

I've always been fascinated and inspired by the NASA autonomous Mars rover robots; it's hard to imagine a more interesting engineering challenge.  I think we're about to enter a new age of autonomy across many industries, so this feels like a good time to dive in head first.

Take the self-driving cars for example.  Despite all of the setbacks and failed startups, we have gone from a Google X moonshot to having completely autonomous Waymo and Cruise vehicles now roaming the streets of San Francisco.  What will this trajectory look like over the next 10 years?

Despite my enthusiasm for AI, I think that the current AI and AGI tech is still in its infancy and we still haven't even gotten close to replicating human-level intelligence in machines.  I think the most promising recent development in AI is the NeuroAI initiative from Yann LeCun, Yoshua Bengio, Jeff Hawkins et al.  Their main point being that we should update the Turing test to the "Embodied Turning Test", and focus much more effort studying our only existence proof of intelligence: the mammalian brain.  

My plan is to try to level up in the current iteration of AI technology and be ready to jump to the forefront of the next disruptive wave of innovation if and when it comes along.

Overall plan

I'm particularly interested in computer vision, so will be focusing on that.  

I'm not planning to get a masters degree from a university, instead opting to take a much more self-guided approach:

  1. Hire a mentor from MentorCruise
  2. Work on pure learning projects with toy problems and public data sets
  3. Contribute to relevant open source projects
  4. Backfill gaps in fundamentals via books and exercises

After leveling up, I'll be looking to join (or even help bootstrap) a startup to put my new skills to work.

Learning resources 🧠

Here's my current list of learning resources:

  1. Andrej Karpathy's Neural Networks: Zero to Hero video lecture series.  Andrej is an excellent teacher and covers just enough theory to understand what's happening under the hood.
  2. Jeremy Howard's FastAI course and thriving community.  This takes a very practical approach and tends to steer towards solving real problems, taking a "progressive disclosure" approach to diving into the theory, thus lowering the barrier of entry.
  3. Deeplearning.ai Coursera course by Andrew Ng.  This is widely considered to be the the "gold standard" for getting into the theory and math behind deep learning.
  4. Udacity - they offer several courses focusing on deep learning, and they pair you with a mentor and offer collaboration opportunities with other students.
  5. The Mathematics of Machine Learning by Tividar Danka.  This is similar in spirit to Deep Learning by Ian Goodfellow, Yoshua Benjio, et al, but Danka's style makes the dense material far more accessible.

Open source + other projects

Here's a few projects I have on my radar and I'm planning to contribute to:

  1. Autoware - an open source self-driving project with corporate sponsorship and a thriving community of contributors.
  2. FruitPunchAI - a University sponsored project that partners with other organizations to use AI for the good of society, for example to use drone and satellite imagery to spot roads and bridges that are in need of repair.  Anyone can participate in their mentor-led projects.
  3. FreeMoCap - AI motion capture ("mocap") is one of the more successful applications of computer vision that has a ton of practical applications - character animation, workout coaching, etc.  This project is quickly gaining momentum to become a state-of-the-art open source AI motion capture tool.

Calling all AI hackers

If you have any tips of your own, or just want to hack on a project together, don't hesitate to reach out on twitter!  (DMs are open)

Stay tuned!