Autoencoders
An autoencoder is a type of DL which can be used for unsupervised learning. It is similar to other dimensionality reduction techniques such as Principal Component Analysis (PCA) which we studied earlier. However, PCA projects data from higher dimensions to lower dimensions using linear transformation, but autoencoders use non-linear transformations.
In an autoencoder, there are two parts to its structure:
- Encoder: This part compresses the input into a fewer number of elements or bits. The input is compressed to the maximum point, which is known as latent space or bottleneck. These compressed bits are known as encoded bits.
- Decoder: The decoder tries to reconstruct the input based on the encoded bits. If the decoder can reproduce the exact input from the encoded bits, then we can say that there was a perfect encoding. However, it is an ideal case scenario and does not always happen. The reconstruction error provides a way to measure the reconstruction effort of the decoder and judge...
