DigiNews

Tech Watch by Johan Denoyer

← Back to articles

The effective sample size

Quality: 8/10 Relevance: 9/10

Summary

The post explains effective sample size in the context of importance sampling and covariate shift, showing how reweighting data increases variance and reduces usable information. It defines n_eff via Kish's method, derives it through two approaches (variance of a weighted sum of normals and Hoeffding’s inequality), and discusses applications to replay buffers in off-policy reinforcement learning and Sequential Monte Carlo.

🚀 Service construit par Johan Denoyer