Been a manager for couple of years now for various SDE teams. Got an opportunity to manage a team which is a mix of ML SDE and scientists. Great opportunity but very nervous and confused if I should take it up? I have done a few data science courses on coursera but that's pretty much it, I don't have any hands on research or engineering experience with ML. Not comfortable with reading research papers yet as there is some learning curve for me to understand all the jargons and math. Will it be hard? Should I just stick to non ML teams? Anybody in my position or experience with these decisions can help me out pls? #machinelearning #manager
Way too broad a scope to say - you might actually be better placed than someone who has ML/DS experience or it might be absolutely necessary to have strong experience. What org you in? How is ML being used? Managing scientists is a red flag to me however, suggesting the latter case.*
Lot of deep learning based models used for text and image data. I saw that folks discuss latest advancements from conferences which got me intimidated
Then you need to really understand ML workflows E2E. Going to be a steep learning curve - not to say you can't do it but you're gonna have to spend alot of time ramping up. ML infra / ops / tooling is getting pretty deep and very cross disciplinary. Papers get easier, you'll learn to filter relevant info.
Are you familiar with the product and the algorithms that are being used within the team?
Close to impossible to do it well with zero experience.... But with just a little experience it's doable
Not sure what your looking for but this scenario could work. Have the beast L6/L7 IC (if you have one) tank the heavy tech stuff while you handle politics for him.
Agreed. That's a good strategy. There is a principal scientist in the team, no PE yet. Maybe I can count on L6 SDE to help me out.
As an ML scientist, I'd recommend refraining from jumping into something like this. I had my fair share of managers without knowledge of science, research or ML, and it ended up in disaster every single time
Yes, I have the same experience. But I don't think it's impossible if OP is very smart and has scientific thinking
Thanks your inputs make sense. The nature of the end2end workflows in ML projects is something that I am not familiar with. Plus the technical details to rampup on. I was assuming I could transition smoothly into this manager type role in this new team but now it does seems hard.
The key difference between regular SWE teams and ML SWE teams is scoping of work and projects. The usual sprint approach and regular cadence of pushing code is not the usual way ML teams work. A lot of time is spent in understanding and cleaning data, and then a lot of time in training and tuning models, and the worst part is when something goes wrong, it’s not a quick bug fix you send out but you have to go all the way back to the data, training and eval to figure out what went wrong. In almost every ML project I’ve been in, we missed deadlines due to unforeseen circumstances and projects easily crept into the next quarter. As long as you are aware of these challenges, it should be fine. But be willing to be super flexible and understanding if the hard work of the team doesn’t translate to immediate results
Thanks this helps a lot. It looks like it's going to be a bumpy ride for me.
Well it will certainly present its challenges but as long as you go in with an open mind and expect that a lot of your experience as a SWE manager needs to be “unlearnt” so you can learn the new stuff you should be fine.
Why can’t they promote a manager from within? I feel bad for the prospective reports.
Hard to get respect from your team if you don’t have the domain knowledge or the expertise. You will mostly end up becoming a people manager.
Partially true. The higher up you are, the more generalists leaders are.
Thinking short buddy