深度学习loss下降，精度不提高，怎么解？

Question

深度学习
RNN+FC
loss下降，精度不提高，怎么解？
model.fit(x_train,y_train,batch_size=64,epochs=10)
Epoch 1/10
50000/50000 [==============================] - 13s 251us/step - loss: 0.4816 - acc: 0.0011
Epoch 2/10
50000/50000 [==============================] - 11s 222us/step - loss: 0.3131 - acc: 0.0016
Epoch 3/10
50000/50000 [==============================] - 11s 221us/step - loss: 0.1625 - acc: 0.0025
Epoch 4/10
50000/50000 [==============================] - 11s 211us/step - loss: 0.0714 - acc: 0.0029
Epoch 5/10
50000/50000 [==============================] - 10s 208us/step - loss: 0.0303 - acc: 0.0031
Epoch 6/10
50000/50000 [==============================] - 11s 219us/step - loss: 0.0127 - acc: 0.0031
Epoch 7/10
50000/50000 [==============================] - 11s 213us/step - loss: 0.0059 - acc: 0.0031
Epoch 8/10
50000/50000 [==============================] - 11s 211us/step - loss: 0.0031 - acc: 0.0031
Epoch 9/10
50000/50000 [==============================] - 11s 213us/step - loss: 0.0018 - acc: 0.0031
Epoch 10/10
50000/50000 [==============================] - 11s 215us/step - loss: 0.0013 - acc: 0.0031

Answer 1

遇到同样的问题，我把学习率调小之后可以了，但是你的acc是不是太低了一些，可能还有别的原因