Ask HN: What's a promising area to work on?

Re:Ask HN: What's a promising area to work on? by closeparen on Wednesday October 23, 2019 @12:36PM (#21336316)

As someone using ML in production at scale, it's not the sophistication of the models themselves, but all the "boring" operational stuff around them.

- Keeping them fed with clean, high quality, low latency feature data.

- Understanding the impact of an ML intervention on the overall system or business process, in aggregate and for all the relevant subpopulations.

- Understanding why scores may be drifting. These are very challenging alerts to investigate.

- Resource efficiency at very high QPS or in mobile contexts.

[ Parent | on HN ]

Re:Ask HN: What's a promising area to work on? by alexhutcheson on Wednesday October 23, 2019 @10:01PM (#21340997)

+ Running constant live experiments on actual traffic to make sure it continues to do what you think it does.

[ Parent | on HN ]