I have an autoencoder trained using Keras and Tensorflow in the following configuration. autoencoder = models.Sequential([ layers.Input(shape=(input_dim,)), layers.Dense(10, activation='relu'), layers.Dense(6, activation='relu'), # Bottleneck layers.Dense(10, activation='relu'), layers.Dense(input_dim, activation='relu') # Output layer ]) # Compile the model autoencoder.compile(optimizer='adam', loss=rmse_loss) # Train the model history = autoencoder.fit(X_train_, X_train_, epochs=50, batch_size=128, shuffle=True, validation_split=0.25, verbose=1) Now, I have a bit of a challenge. Instead of giving it an input on the encoder side, I want to give it an input on the decoder side to observe its corresponding input from the encoder. Is something like this even possible, considering autoencoders are not invertible functions? Continue reading...
