Shreya Shankar
PhD student in databases focusing on MLOps @Berkeley_EECS @UCBEPIC. Prev. 1st ML engineer @viaduct_ai, research @googlebrain, BS & MS @Stanford CS. She/they.
View on 𝕏Threads
My thoughts on baselines, a concept that is *extremely* relevant in industry ML but does not exactly translate from academic ML: 1/9
Seeing lots of people on my feed hate on an “intense work ethic.” I’m assuming “intense” means “hard.” I think it’s harmful to hate on this — as if AI/ML practitioners and research...
I thought about this a lot when releasing ML APIs that include classification model outputs in the response. Example: individual probabilities may be hard to trust, but a calibrati...
The ML research ecosystem can be amazing. A few hours ago, I wondered: do pruned neural networks converge to high accuracies faster than the original networks? I'm sure I can find...
I have been feeling tired lately when thinking about the differences between MLOps and DevOps. There are so many “gotchas” to keep track of in production ML systems, but I don't th...
this is also applicable for those working in industry! some other things i've found to be helpful as a practitioner to keep up with ML literature: * pick subfields intentionally -...
Thrilled to see a NeurIPS 2020 paper that eeks out a bit more performance on benchmark image datasets by...removing mislabeled examples from the train set! "Identifying Mislabeled...
This isn't just for consumer products! Few people building dev tools or PLs talk publicly about understanding programmers' emotions. Programmers aren't just "logical" -- they also...
Not sure I fully buy “our goal as machine learning engineers should be to raise, rather than beat, human level performance” — I think this is a temporary solution to get immediate...
as a practitioner who 1) threw a vanilla transformer at a large time series dataset 2) got “good enough” results 3) didn’t take the time to deeply understand the architecture or f...
Unit testing for ML pipelines is challenging given changing data, features, models, etc. Changing I/O make it hard to have fixed unit tests. To hackily get around this, I liberal...
Going to try another way of explaining why I think ML product dev is broken, this time with a clear software analogy: (1/6) https://t.co/4jxQ9Ny0yg