The autoencoder keeps learning about the function form for the feature to capture the relationship between input and output. An example of how the computer is visualizing the image after 1,000 iterations is shown in the following figure:

After 1,000 iterations, the computer can distinguish between a major part of the object and environment. As we run the algorithm further to fine-tune the weights, the computer keeps learning more features about the object itself, as shown in the following figure:

The preceding graph shows that the model is still learning, but the learning rate has become smaller over the iterations as it starts learning fine features about objects, as shown in the following image. There are instances when the model starts ascending instead of descending, due to batch gradient descent: