mkaze

+ Keras CIFAR10_CNN example: training with Adam optimizer with its default parameters
+ Reaches 75% validation accuracy after 5 epochs, and 79% after 18 epochs
-------------------------------------------------------------------------------------

Not using data augmentation.
Train on 50000 samples, validate on 10000 samples
Epoch 1/100
50000/50000 [==============================] - 91s 2ms/step - loss: 1.5120 - acc: 0.4473 - val_loss: 1.1130 - val_acc: 0.6052
Epoch 2/100
50000/50000 [==============================] - 91s 2ms/step - loss: 1.1032 - acc: 0.6068 - val_loss: 0.9453 - val_acc: 0.6673

mkaze

Keras model performance on bAbI QA1 task:

Training
Train on 950 samples, validate on 50 samples
Epoch 1/40
950/950 [==============================] - 5s 5ms/step - loss: 2.7767 - acc: 0.1726 - val_loss: 2.0589 - val_acc: 0.2000
Epoch 2/40
950/950 [==============================] - 3s 3ms/step - loss: 2.0278 - acc: 0.1495 - val_loss: 1.8242 - val_acc: 0.2200
Epoch 3/40
950/950 [==============================] - 3s 3ms/step - loss: 1.9147 - acc: 0.1695 - val_loss: 1.8467 - val_acc: 0.0800

mkaze

/* Scan Example

   This example code is in the Public Domain (or CC0 licensed, at your option.)

   Unless required by applicable law or agreed to in writing, this
   software is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
   CONDITIONS OF ANY KIND, either express or implied.
*/

/*