Autoencoders are non-linear, encoder + decoder.

Autoencoders are non-linear, encoder + decoder.

• A. Autoencoders
• B. PCA
• C. K-means
• D. t-SNE

Answer: (A)

Autoencoders learn by compressing data into a latent representation with an encoder and then reconstructing the original input with a decoder. The networks used for both parts typically include non-linear activation functions, which lets the model capture complex patterns beyond what a linear transformation could learn. This encoder–decoder structure is what distinguishes autoencoders from the other methods: PCA uses a linear projection with no learned inverse mapping, K-means is about clustering without reconstruction, and t-SNE builds a low-dimensional embedding focused on preserving local neighborhoods and doesn’t provide a straightforward inverse that reconstructs the input. While a strictly linear autoencoder is possible, the common, effective form relies on non-linearities, making the statement true in typical use.

Autoencoders learn by compressing data into a latent representation with an encoder and then reconstructing the original input with a decoder. The networks used for both parts typically include non-linear activation functions, which lets the model capture complex patterns beyond what a linear transformation could learn. This encoder–decoder structure is what distinguishes autoencoders from the other methods: PCA uses a linear projection with no learned inverse mapping, K-means is about clustering without reconstruction, and t-SNE builds a low-dimensional embedding focused on preserving local neighborhoods and doesn’t provide a straightforward inverse that reconstructs the input. While a strictly linear autoencoder is possible, the common, effective form relies on non-linearities, making the statement true in typical use.