Keras: Decrease in training losses (increase in accuracy), while increase in confidence increases (decrease in accuracy)

Question

I am working on a very meager dataset to predict 6 classes. I tried to work with a large number of models and architectures, but the problem remained the same.

When I start training, acc for training will begin to increase slowly, and the loss will decrease where validation will do the exact opposite.

I have really tried to handle retraining, and I just can't believe that is what studies this issue.

What i tried

Transfer training to VGG16:

• exclude the top layer and add a dense layer with 256 units and 6 units of the output softmax layer.
• finetune top block CNN
• finetune top 3-4 blocks CNN

To handle retraining, I use heavy magnification in Keras and falling out after 256 dense layers with p = 0.5.

Creating your own CNN with VGG16-ish architecture:

• including batch normalization where possible
• Regulation L2 on each dense CNN + layer
• Disconnect from somewhere between 0.5-0.8 after each CNN layer + tight + pool
• On-the-fly data growth at Keras

Understanding that maybe I have too many free parameters:

• network reduction to contain only 2 CNN blocks + dense + output.
• engaged in processing in the same manner as described above.

Without exception, all training looks as follows: Training and validation + accuracy

The last architecture mentioned is as follows:

    reg = 0.0001

    model = Sequential()

    model.add(Conv2D(8, (3, 3), input_shape=input_shape, padding='same',
            kernel_regularizer=regularizers.l2(reg)))
    model.add(BatchNormalization())
    model.add(Activation('relu'))
    model.add(Dropout(0.7))
    model.add(MaxPooling2D(pool_size=(2, 2)))
    model.add(Dropout(0.5))

    model.add(Conv2D(16, (3, 3), input_shape=input_shape, padding='same',
            kernel_regularizer=regularizers.l2(reg)))
    model.add(BatchNormalization())
    model.add(Activation('relu'))
    model.add(Dropout(0.7))
    model.add(MaxPooling2D(pool_size=(2, 2)))
    model.add(Dropout(0.5))

    model.add(Flatten())
    model.add(Dense(16, kernel_regularizer=regularizers.l2(reg)))
    model.add(BatchNormalization())
    model.add(Activation('relu'))
    model.add(Dropout(0.5))

    model.add(Dense(6))
    model.add(Activation('softmax'))

    model.compile(loss='categorical_crossentropy', optimizer='SGD',metrics=['accuracy'])

Keras flow_from_directory:

    train_datagen = ImageDataGenerator(rotation_range=10,
                                width_shift_range=0.05,
                                height_shift_range=0.05,
                                shear_range=0.05,
                                zoom_range=0.05,
                                rescale=1/255.,
                                fill_mode='nearest',
                                channel_shift_range=0.2*255)
    train_generator = train_datagen.flow_from_directory(
                train_data_dir,
                target_size=(img_width, img_height),
                batch_size=batch_size,
                shuffle = True,
                class_mode='categorical')

    validation_datagen = ImageDataGenerator(rescale=1/255.)
    validation_generator = validation_datagen.flow_from_directory(
                                            validation_data_dir,
                                            target_size=(img_width, img_height),
                                            batch_size=1,
                                            shuffle = True,
                                            class_mode='categorical')

Answer 1

, ( , ):

, 30 . ~ 30 ( ). Keras Callbacks , ModelCheckpoint, (Ctrl + C) . ModelCheckpoint :

#save best True saves only if the metric improves
chk = ModelCheckpoint("myModel.h5", monitor='val_loss', save_best_only=False) 
callbacks_list = [chk]
#pass callback on fit
history = model.fit(X, Y, ... , callbacks=callbacks_list)

(Edit:) , , , - EarlyStopping, "" , . , callbacks, .

, ( , ), , , ; ( ).

, , , , - , . . , , .

, , , . , ? CNN , , .

, , , . keras, . , , , , . , .

Answer 2

IMHO, DL. Keras , ( , ), , , .

. ModelCheckpoint EarlyStopping .

P.S. , , - , ?