- formatting
- images
- links
- math
- code
- blockquotes
- external-services
•
•
•
•
•
•
-
Online Estimation: Exponential Forgetting
Deriving exponentially decaying mean and covariance estimates in full, correcting the unrolled historical expansion, and distinguishing decay applied once per batch from decay applied once per observation.
-
Building Intelligent Agents for Connect-4: Opening Databases
How the 8- and 12-ply opening books encode millions of solved positions in compact fixed-width records, and how the original database pipeline is used by BitBully today.
-
Online Estimation: Weighted Observations and Mini-Batches
Carrying the weighted scatter derivation from the thesis through in full, to obtain exact single-observation and mini-batch updates, together with a four-point example which catches an incorrect batch formula.
-
Building Intelligent Agents for Connect-4: Transposition Tables
Why the same Connect-4 position keeps reappearing in the search tree, how Zobrist hashing exploits three properties of the exclusive OR, and why the more recent solver manages without it entirely.
-
Online Estimation: The Mean and the Covariance Matrix
Deriving a running mean and scatter matrix from the definitions, carrying every step of the algebra through, and distinguishing the unbiased covariance from the Gaussian maximum-likelihood estimate.