Best partition is scored by stationarity distance to yield a faster Markov chain
michaelchchoi · x · 2026-07-22
- The second part of the thread says the best partition \(\mathcal{O}\) is selected by scoring how far it is from stationarity using the Frobenius norm.
- The quotient chain \(G{\mathcal{O}}P\) is then claimed to be a faster Markov chain than the original \(P\).
- In other words, the method is not just clustering states; it is trying to produce a reduced chain with better mixing behavior.
Related event: Markov Chain Partitioning via Spectral Features and Weighted k-Means(2 posts)→
More from Research
- Four teams independently shipped the same “LLM wiki” pattern after Karpathy’s gist — garrytan · 2026-07-22
- LightInteraction speeds interactive video world models up 2.59× without retraining — 新智元 · 2026-07-22
- DA-Nav gives robots recovery-aware long-horizon navigation and 98.15% correction success — 新智元 · 2026-07-22
- Weights & Biases’ interactive confusion matrix gets praise and a marimo comparison — _ScottCondron · 2026-07-22
- CyberGym joke points to benchmark leakage in Hugging Face data — Nunki08 · 2026-07-22
- LTX2.3 PoleDance LoRA MK6 shows better results at step 1,000 of 12,000 — JahJedi · 2026-07-22