Good survey paper that includes ethical considerations ("neutral" demographic data is biased) and the importance of interpretability for ML deployments.
"Many pain points we have described in this work were already experienced by communities in these fields, and the ML community should turn to them for solutions and inspiration.”
https://dl.acm.org/doi/10.1145/3533378
Mastodon Source 🐘
Some of these challenges are shared with normal software engineering practices, but others are unique to ML.
Mastodon Source 🐘
This closing observation highlights the distance between theory and practice:
"As an observation that follows from the process of collecting papers to review in this survey, we note the relative shortage of deployment experience reports in the academic literature.”