Articles, newest first
Preconditioning Changes the Problem Your Iterative Solver Sees
Large sparse linear systems are often limited less by arithmetic than by conditioning and spectral geometry. A preconditioner changes the system seen by the ...
Read article
Permutation Importance with Correlated Features: When the Ranking Lies
A near-duplicate sensor with no effect of its own outranks a feature that genuinely drives the outcome. Permutation importance is working exactly as designed...
Read article
Unlabelled Data Does Not Identify the Decision Boundary
Unlabelled observations can estimate where the data live, how dense different regions are and which points are close. They do not, by themselves, identify wh...
Read article
Post-Stratification: Weighting a Survey That Answered Unevenly
The survey says 7.36 and the population says 7.10. Nobody lied: older customers answered twice as often as younger ones and they score higher. Weighting to t...
Read article
When the Bootstrap Fails: Dependent Data, Small Samples, and Extremes
Resample the data, recompute the statistic, read off the spread. On an autocorrelated series of 200 points the interval that comes out is typically 2.4 times...
Read article
DPO Changes Preferences, Not Knowledge
Supervised fine-tuning teaches a model what a good answer looks like. Preference optimization teaches it which of two answers should be preferred. Those are ...
Read article
Interference in Experiments: When Treated Users Take What Control Users Would Have Bought
The ranking change raises purchase intent from 10 to 12 percent, and the A/B test reports a 20 percent lift in sales. Rolled out to everyone, it delivers 3 p...
Read article
When Results Become Rhetoric: Evidence, Authority and Commercial Incentives in Online Health Communication
Client results can be genuine without supporting the explanation used to sell a method. A wall of the twenty best transformations says more about how many cl...
Read article
Paired vs. Independent Samples: The Design Choice Behind the Test
The choice between paired and independent tests is not a software option. It is a statement about the study design and the dependence structure in the data.
Read article
Network Structure Changes Dynamics Before Any Model Is Fit
Network topology is part of the data-generating process. The same node-level population can spread, fragment, synchronize and transmit information differentl...
Read article