Lightning Talks
summary
Seven rapid projects present concrete models, algorithms and field tests: a Wayfair ranking algorithm uses a threshold scan to fix ranks in one pass, proves a 1/2-approximation with correlated attention (1−1/e if independent) and yields 5–30% lift in historical-click simulations; persuasion with non‑expected‑utility agents gives a convex program and at most |Ω| signals (binary case reduces to an LP, needs six signals in one queueing example); belief‑based transfer terms enable new full‑implementation schemes; anonymous‑network targeting uses menus and group‑IC contracts; public signals map to k‑cores and admit a double‑interval structure; a field experiment (≈180k jobs) shows adjective scales deflate top scores from ≈80% to <36% top ratings and predict rehiring; and a sample‑efficient automated mechanism design method learns correlated distributions with polynomial samples and approaches surplus extraction in simulations.
AI-written summary · about
chapters (4)
- 0:00Start
- 11:46Bayesian persuasion: standard framework
- 18:49Special case: binary persuasion
- 53:07Optimality of Double Intervals
related talks
24mHighlights from Current Programs part 2:
46mFrom correlation to causation: measuring ad effectiveness at scale
50mAnalytics – Balancing ‘Advanced’ and ‘Actionable’
32mMLDS 2019: Pradeep Gulipalli of Tiger Analytics at Machine Learning Developers Summit 2019
43mToward Human-Centered Machine Learning
15mBeyond Efficiency
22mReal-world applications of machine learning: Beyond the hype
47mDebate "The force behind decision-making – data, wisdom, and the crowds"