3 ms·
Good data is hard. Anyone conducting research in the physical sciences knows this firsthand. It takes painstaking effort to conduct carefully controlled experim
by glitchc 6y ago
Good data is hard. Anyone conducting research in the physical sciences knows this firsthand. It takes painstaking effort to conduct carefully controlled experiments and collect a batch of good data that could then be used for analysis.
The promise of ML has always been to churn out good results from not-so-good data. If I now need to sanitize my data carefully, what's the advantage?
- claytonjy 6y agoWhat makes you think that was the promise? I would say the promise has been to replace code with data, and to build things with data that would be practically impossible to build with code. "Garbage in, garbage out" has been the mantra for a long time. Sure, there are tools and techniques to deal with not-so-good data, but those are add-ons, not a core part of the value proposition of ML.