Xiangyu Zeng 2015-10-05

last week:

1.accomplished the sequence training of adam-max, and got some results which shows that with adjust-lr adam-max can get a better performance.

2.accomplished the code of multitask with speech rate, but still some problems haven't been solved, such as, it diverges after improving weight of speech rate.

this week:

1.supplement some experiments on adam-max sequence training, add noisy data into clean data and see if adam-max can adjust itself. 2.implement the multitask with speech rate.