My DNN-nework is the same as the paper says,I only use three people as train data.
I testing my model when I train it. the train loss is becoming smaller and the valid accuracy on valid data is becoming bigger in the first 300steps as fllows.It means this model is correct.
train loss is 1.0305396
valid acc is 0.46666667
train loss is 0.9465262
valid acc is 0.6
train loss is 0.7817914
valid acc is 0.73333335
train loss is 0.68828535
valid acc is 0.6666667
train loss is 0.62642306
valid acc is 0.8666667
train loss is 0.53032285
valid acc is 0.93333334
train loss is 0.51416075
valid acc is 0.93333334