Great post from a16z on long-tailed AI problems: https://a16z.com/2020/08/12/taming-the-tail-adventures-in-improving-ai-economics/
My favorite quotes:
1 - Long-tailed distributions are also extremely common in machine learning, reflecting the state of the real world and typical data collection practices.
2 - The long tail can contribute to high variable costs beyond infrastructure. Even worse, AI businesses working on long-tailed problems can actually show diseconomies of scale – meaning the economics get worse over time relative to competitors.
3 - ML is not a religion, but science, engineering, and a little art.
4 - If you are working on a long-tail problem, it’s critical to determine the degree of consistency across customers, regions, segments, and other user cohorts. If the overlap is high, it’s likely you can serve most of your users with a global model (or ensemble).
5 - In "componentizing", each model addresses a global slice of data – rather than a particular customer, for instance – and that the sub-problems are relatively bounded and easy to reason about.
6 - Over time, models that share similar functionality can be joined together with a common “trunk” to reduce complexity. The goal is to make the trunk models as “thick” as possible (i.e. doing most of the work) while making the task-specific “branch” models as “thin” as possible – without sacrificing accuracy
"An engineering team that built their own ML Platform from the ground up, flawed as it may be, will attribute more value to it than if they just bought something out-of-the-box from a vendor. They give it a fancy name, write blog posts about it, and everyone gets promoted."
pip-sync requirements.txt requirements-dev.txt
I have registered and paid for the upcoming spring-21 Berkeley offering of the course but I haven’t received any communication other than a Stripe receipt. And a few week later I got another email asking again if I want to register for the course. Did anyone get any communication after the payment from the FSDL team? I want to make sure that I am registered.
I also see discrepancy in the schedule mentioned on the course page (supposed to start on week of Jan 19) and in the email (supposed to start on week of March 1). I had reached out to the FSDL team around a week back on a support email mentioned in the earlier emails but haven’t got any reply so any pointers would be appreciated. Thank you!
Hey guys, I have little experience working in the area of deep learning before getting fully into it. Well, I love the idea of Data Science to be end - to - end who can actually deliver values too. Will it be fine taking this course at first and getting to know about the complete workflow of Deep learning end - to -end (or) work on deep learning for some time and jump back to this course?
I am a little confused about how should I take this course, any suggestions? Thank ya!