My impressions from the SW Architecture 2017 conference.
http://conference70.wixsite.com/sw-architecture-2017
Generally speaking the conference was interesting and had some interesting insights, but overall not overwhelmeangly important. I’ve put together some of these thougths, while not all of them are important, I believe it is a good idea to capture them.
- Separate ML Inference from data storage for services
- Privacy issues with Big Data - plan ahead
- Time drift - be aware that you will need to retrain on new inputs
- enrich data labeling based on previous inputs
- Air-gapped systems are challanging for ML
- Multi staged processing is usually very beneficial
- Use different time resolution, use compression
- Prefetch data for analysis to increase speed
- Consider self-healing when designing your system
- Voting systems can be used to decide between regular and alternate flow, reduce risks
- if you can easly explain it, you can tag it
- use a classifier to identify NN weak spots (e.g. when it is raining)
- Customers want control of their data
- Most customers prefer on-premis
- Alpha and Beta participation with your customer