9/24/2021: Autoencoders for sparse data compression, Julio Godoy, University of Concepción

Speaker/presenter: Julio Godoy

i) Presentation description:

Autoencoders are a specific type of feedforward neural network, which aim at reconstructing the input after encoding it into a lower-dimensional latent representation. The autoencoder is composed by an encoder, which transforms the input into a lower dimensional representation, and a decoder which tries to reorganize the original input from this representation. Autoencoders are mostly used for dimensionality reduction, image segmentation and coloration, and for anomaly detection tasks. They can also be used for compression, but there might be better alternatives for specific data types. For dimensionality reduction, they have been shown to perform better than PCA, at the cost of higher complexity.

Reference: 

• Vincent, Pascal and Larochelle, Hugo and Lajoie, Isabelle and Bengio, Yoshua and Manzagol, Pierre-Antoine and Bottou, León. Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion, Journal of Machine Learning Research, vol 11(12), 2010.