Dynamically Aggregating Diverse Information
summary
Claims an exact, closed‑form characterization of optimal dynamic information acquisition when K Gaussian attributes are observed via Brownian motion signals and attention (a unit budget) is continuously allocated; under mild prior restrictions (e.g. perpetual substitutes or diagonal dominance) the optimal policy is a deterministic, history‑independent strategy that samples one source initially then switches at computable times to constant mixtures proportional to payoff weights. Methods use posterior‑variance minimization, a static-to-dynamic reduction and Blackwell comparisons; applications cover binary choice generalization, attention‑diversion effects with correlation, and a two‑news‑source equilibrium showing more polarization and lower precision.
AI-written summary · about
chapters (8)
- 0:00Welcome
- 1:07Workshop overview
- 6:34Aggregating information model
- 14:00Optimal attention strategy
- 20:47Solution properties
- 31:24Proof intuition
- 44:30Model applications
- 52:20Conclusion and model limitations
related talks
30mBuying Data Over Time: Approximately Optimal Strategies for Dynamic Data-Driven Decisions
01h 11mA Theory of Multi-objective Machine Learning
59mWhen do dependencies in your data help?
23mDynamic inference in probabilistic graphical models
01h 01mStochastic Bandits: Foundations and Current Perspectives
35mLearning General Gaussian Mixtures With Efficient Score Matching
45mAGT and Machine Learning
31mInterpreting Machine Learning From the Perspective of Nonequilibrium Systems