My nuts and bolts solutions on some Deep Learning/Machine Learning/NLP projects
Dataset: 400K+ pairs, task: binary classification Link to download
Model Architecture: ![Model Architecture] (https://raw.githubusercontent.com/lampts/my_dl_solutions/master/QUORA.png)
Leaderboard
| Model | # Params | Test (Accuracy) |
|---|---|---|
| Bilateral Multi-Perspective (Wang, IBM) | N/A | 88.17 |
| Small cat RNN+CNN this model (Laam, Sentifi) | 1.4M | 84.89 |
Notes: we need the apple-to-apple comparison, so I just put the result based on same test data from Wang.
My solution: F1 on validation is 94.92