Machine Learning

1 readers

1 users here now

Community Rules:

Be nice. No offensive behavior, insults or attacks: we encourage a diverse community in which members feel safe and have a voice.
Make your post clear and comprehensive: posts that lack insight or effort will be removed. (ex: questions which are easily googled)
Beginner or career related questions go elsewhere. This community is focused in discussion of research and new projects that advance the state-of-the-art.
Limit self-promotion. Comments and posts should be first and foremost about topics of interest to ML observers and practitioners. Limited self-promotion is tolerated, but the sub is not here as merely a source for free advertisement. Such posts will be removed at the discretion of the mods.

founded 11 months ago

MODERATORS

communick@academy.garden

[D] how to explain why RL is difficult to someone who knows nothing about it? (alien.top)

submitted 9 months ago by savvyms@alien.top to c/machinelearning@academy.garden

17 comments fedilink hide all child comments

How to explain why RL is difficult to someone who knows nothing about it?

I’ve been working on an RL project at work. The person who assigned it to me is a computer scientist who is not an expert on RL, but understands it’s a difficult problem. (My boss is on equal footing with the person who assigned the project to me. My boss is not a computer scientist and doesn’t know anything about RL.) This guys boss is a business manager who doesn’t know anything about RL and knows very little about ML. The business manager wants a report on how the project is going from me and I’m getting the sense that he doesn’t really understand why this is taking so long.

For context, I’ve been working on this project for about 4 months for 15 hours per week. In that time, I’ve built an entire code base for the problem from scratch and programmed up several models. I have one that mostly works at the moment, but I need to make some changes to the reward functions to get it performing well consistently. I’m the only one working on this project, so I’ve done all of this myself. I also had only done vanilla RL prior to this, so I’ve had to learn a ton about deep RL to make this work. Luckily I know someone who’s an expert in deep RL (outside work) and has been able to give me pointers. I’m feeling like I’ve made a ton of progress and am nearing the home stretch in terms of having a completely polished model. However I’m getting the sense that this guy is not super thrilled with me. This guy doesn’t have any official authority over me, so this is mainly about trying to explain how much work RL is in addition to mg normal slides about the project and where I’m at.

you are viewing a single comment's thread
view the rest of the comments

[–] matty961@alien.top 1 points 9 months ago (1 children)

Unfortunately it's hard to sell your project's difficulty to someone who doesn't know as much about ML.

I'm not sure if you've done this, but IMO the best way to give non-technical people confidence in a long-running project is to break down your work into milestones each with a concrete deliverable, and then estimate how long you think each milestone will take. That way, the business manager can determine:

When will the project be done?
Is the project on-track to finish on time?
What will be delivered and at what dates?

Which is probably all that they care about. If the person who assigned you the project and your boss think your timeline is reasonable, your project is on time and you're delivering what you're supposed to, then no one can really complain.

I've always been of the opinion that writing timelines for ML projects is a lot harder than non-ML work, since often for ML you are trying to hit some "good enough" bar, but you don't know exactly what it will take to get there.

[–] savvyms@alien.top 1 points 9 months ago

I have done that for several other projects (usually they ask at the outset, tells you something about how this one is being run lol). Neither the project lead or the business manager has given me a time limit or target date for this project, but that might be the best way to recalibrate their expectations.