Machine Learning Accelerator: Investing in Career Growth for Solutions Engineers at Facebook

— Written by Vanya Chary, Vitaliy Zasadnyy, Jason Ogden

As Solutions Engineers (SE) our mission is to help businesses succeed through technology and partner solutions. To deliver on this, it is important for SEs to be equipped with all the tools necessary for building future-proof solutions.

The Solutions Engineering Org at Facebook proactively invests in the long term career growth of their engineers. One such initiative that was born out of a desire to level up Solutions Engineers’ skills in ML is the Machine Learning Accelerator. ML Accelerator is for SEs looking to advance their ML understanding in their current role — without any prior experience. The program is 4 weeks long:

• 2 weeks rigorous training and labs (~4 hours a day)
• 2 weeks to work on a project with regular check-ins (~2–4 hours a day)
• Demo day to showcase your work

What do SEs gain from the ML Accelerator?

1. Master ML Basics: Experience 30 hours of curated content and hands-on labs, diving into the nuts and bolts of ML and demystifying key concepts.
2. Learn Transferable Skills: Gain experience with Python ML tools like PyTorch, doing supervised/unsupervised learning with regressions, neural nets, and decision trees.
3. Leverage Facebook Infrastructure: Learn how to move fast using Facebook infrastructure for Natural Language Processing, Computer Vision, and personalization with guided tutorials.
4. Gain Pragmatic Knowledge: Choose feasible ML projects and avoid wasting time by analyzing where to invest: tuning parameters, preparing data, and model architectures.
5. Work with Expert Advisors: Take advantage of experienced people to help answer questions and advise you in your project, and if invited, become an advisor yourself.
6. Demo Your Original Project: As part of the program, build original models, with the support of your peers and advisors, and share your work on Demo Day.

At the beginning of 2020, we kicked off Cohort 0 of the program — dogfooding the content ourselves. At the time, we had a solid plan, but we needed to make sure it worked. A year later, we offer a full-fledged learning experience, regularly available to all of our employees on the Facebook Learning Management System. In our first year of operation, we ran 6 cohorts, training almost 100 engineers who have delivered 40+ ML projects, 54% of which either spawned a new Solutions Engineering Build project or were used to improve ongoing projects. One of the projects was even presented to Chris Cox (CPO) and Mike Schroepfer (CTO) at a company-wide prototype forum!

How did we achieve that?

1. Building a Support System

From day 0 we designed a program to be collaborative. Even during the mandatory WFH, we supported people to watch content and go though the labs together. This enabled more experienced participants to help others and learn from answering questions along the way.

2. Balanced Content Strategy

When we designed content, we had an assumption that to ramp-up people fast we have to provide as many up-to-date hands-on tasks as possible.

For instance, we’ve created labs in Pytorch that focus on implementation of theoretical concepts with more practical Jupyter notebooks. It is important that participants can apply their learnings in a widely used open source language.

3. Tight Feedback Loop

As Machine Learning is a fast moving area, we keep improving and updating content based on the multilevel feedback collected from the program participants and the latest developments in the industry. For instance, we recently added an entire module on fairness in ML models.

ML Accelerator goes beyond passive learning — it taps into our builders’ spirit, letting students apply their knowledge to custom projects. It was designed for Solutions Engineers, but the mission of the program resonates beyond our org and addresses a growing need for engineers to be familiar with Machine Learning.