Unsupervised Neural Hidden Markov Models
About
In this work, we present the first results for neuralizing an Unsupervised Hidden Markov Model. We evaluate our approach on tag in- duction. Our approach outperforms existing generative models and is competitive with the state-of-the-art though with a simpler model easily extended to include additional context.
Ke Tran, Yonatan Bisk, Ashish Vaswani, Daniel Marcu, Kevin Knight• 2016
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Forecasting | Exchange (test) | MSE100 | 63 | |
| Time Series Forecasting | Traffic | MAE1.5881 | 58 | |
| State estimation | 3-variable simulated dataset (No. 2) with frequent transitions v1 (test) | Accuracy68.5 | 18 | |
| Forecasting and state estimation | 10-variable simulated dataset (infrequent transitions) | Accuracy75.55 | 18 | |
| Unsupervised POS tagging | WSJ entire corpus (full) | M1 Score79.1 | 13 | |
| Forecasting | Exchange dataset | MAE4.4926 | 13 | |
| Forecasting | 3-variable simulated dataset with infrequent transitions (test) | MAE0.1277 | 9 | |
| Forecasting | Simulated dataset (3 variables) with infrequent transitions | MAE0.1277 | 9 | |
| Forecasting | Traffic (test) | MAE1.5881 | 9 | |
| Forecasting | SMachine | MAE0.0194 | 9 |
Showing 10 of 20 rows