Co-design of dataset & analytical model

Typically, the analytical dataset workflow - design, acquire, prepare" precedes developing the analytical model. Increasingly it is found that co-designing (develop simultaneously) the two results in higher performance (X2000) and improves accuracy. Hence, not data-centric or model-centric to use Ng’s terms, but a balances view on both.

[1] Olteanu, Dan The Relational Data Borg is Learning, PVLDB 13(12): 3503 - 3516, 2020 VLDB 2020 keynote.

How does this relate to S+Data-centric AI?
Michael L Brodie, Harvard

1 Like