Difference between revisions of "Chao Xing 2016-04-18"

From cslt Wiki
Jump to: navigation, search
(以“Last Week: ---- 1. Review DSSM work group's work during last four years. 2. Code done DSSM model including : DNN-DSSM, CNN-DSSM. 3. Test for DNN-DSSM performance s...”为内容创建页面)
 
Line 15: Line 15:
  
 
2. Code compare with :  
 
2. Code compare with :  
**1. DNN-DSSM target on cosine distance (same to their papers. Just consider positive sample).
+
:*1. DNN-DSSM target on cosine distance (same to their papers. Just consider positive sample).
  
**2. DNN-DSSM target on negative sample max margin.
+
:*2. DNN-DSSM target on negative sample max margin.
  
**3. DNN-DSSM target on triplet max margin (consider both positive sample & negative sample).
+
:*3. DNN-DSSM target on triplet max margin (consider both positive sample & negative sample).
  
**4. Compare both DNN, CNN, RNN model target on different cost function above.
+
:*4. Compare both DNN, CNN, RNN model target on different cost function above.
  
 
3. Test for 9 model above.
 
3. Test for 9 model above.

Revision as of 00:46, 18 April 2016

Last Week:


1. Review DSSM work group's work during last four years.

2. Code done DSSM model including : DNN-DSSM, CNN-DSSM.

3. Test for DNN-DSSM performance same to dongxu's experiments set.

  • Get the results for test dataset : 60%, and train dataset : 77%

4. Code for mixture data version DNN-DSSM.

This Week:


1. Test for mixture-DNN-DSSM.

2. Code compare with :

  • 1. DNN-DSSM target on cosine distance (same to their papers. Just consider positive sample).
  • 2. DNN-DSSM target on negative sample max margin.
  • 3. DNN-DSSM target on triplet max margin (consider both positive sample & negative sample).
  • 4. Compare both DNN, CNN, RNN model target on different cost function above.

3. Test for 9 model above.