Chao Xing 2015-09-21
From cslt Wiki
Last Week Solution: None Plan to do: 1. Finish Bi-Resource Training. 2. Help Lantian with his paper. 3. Test to different benchmark, such as word similarity, and text classification. 4. Run for a small vocabulary size, such as 5000. Problem: 1. DNN not as good as Linear Transfer Learning. 2. Linear Transfer learning same channel worse than different channel.
Solution: 1. Tuning DNN. 2. Experiment for different resource. Plan to do: 1. Help Lantian finish his papers. 2. Run for a small vocabulary size, such as 5000.