I use a CNN model and I am working with the MNIST dataset. However, the model's performance is unexpectedly low - I am getting an accuracy of just 0.0664.
Below is the code for my model:
model = Sequential()
model.add(Conv2D(32, kernel_size=(5, 5), strides=(1, 1), padding="same", kernel_initializer='glorot_uniform', input_shape=(64, 64, 1), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2), padding="same"))
model.add(Flatten())
model.add(Dense(1024, activation='relu'))
model.add(Dense(num_classes, activation='softmax'))
model.compile(optimizer = Adam(learning_rate = 0.01), loss = 'categorical_crossentropy', metrics = ['accuracy'])
model1 = model.fit(x_train, y_train,batch_size=128, epochs=20)
How I can improve the performance?
A potential improvement could come from reducing the size of your Dense layer. Try minimizing the number of units in your Dense layer to 64 instead of 1024. Update your model with:
model.add(Dense(64, activation='relu'))