List your company
ResearchFormerly WizWand

Next-step prediction on Synthetic HMM Undercomplete case, N=4000

1.368Validation LossHMM Filter
1.228642.169323.114.05068Nov 13, 2025
Updated 6mo ago

Evaluation results

MethodLinks
HMM Filter2025.111.368
nanoGPT-s# parameters=73,920, training time=2.5 min, # iterations=2,000 iter.2025.111.458
nanoGPT-m# parameters=221,760, training time=7.5 min, # iterations=2,000 iter.2025.111.475
Belief Net# parameters=12,352, training time=20 min, # iterations=2,000 iter.2025.111.569
Spectral# parameters=524,416, # iterations=(PCA)2025.111.624
Baum-Welch# parameters=12,352, training time=51 min, # iterations=20 iter.2025.111.951
Random2025.114.852