Xavier Amatriain, VP of Engineering, Quora
summary
Presents ten concrete lessons from building Quora’s ML stack: prefer implicit signals beat explicit (more data, but combine with explicit to avoid clickbait), mix supervised and unsupervised features and embeddings, use ensembles to combine models, and treat model outputs as features. Describes feed ranking as a weighted sum of predicted actions, a debuggability UI that traces feature contributions, and a case where a Spark job on 15 machines took 6 hours but a C++ single‑machine rewrite runs in 10 minutes. ensemble is the master algorithm
AI-written summary · about
chapters (18)
- 0:00Start
- 2:28What we care about
- 4:36ML Applications @ Quora
- 5:31Models
- 7:02Implicit vs. Explicit
- 11:59Training a model
- 13:31Example 2 - Quora's feed
- 15:09Supervised/Unsupervised Learning
- 20:25Ensembles & Feature Engineering
- 21:03The Master Algorithm?
- 22:10Outputs will be inputs
- 23:10ML vs Software
- 26:26Feature Engineering Example - Quora Answer Ranking Quora
- 29:45Machine Learning Infrastructure: Experimentation & Production Quora
- 31:31Model debuggability
- 34:05Distributing ML
- 37:02Data Scientists and ML Engineers
- 38:01The data-driven ML innovation funnel
related talks
25mXavier Amatriain, VP of Engineering, Quora @ MLconf SEA
04mXavier Amatriain interview
18mBuilding a Machine Learning Platform at
38mFrom text to knowledge via ML algorithms - the Quora answer
24mJosh Wills, Head of Data Engineering, Slack
50mDemocratizing AI - Back-fitting end-to-end machine learning
20mWhat can AI do for me?: evaluating machine learning interpretations in cooperative play
29mXavier Amatriain, Director of Algorithms Engineering, Netflix