As a data scientist, Larry Finer has a strong interest in research that requires statistical and analytical skills to complete. Among his projects has been the development of an autoencoder based on a convolutional neural network, with an aim of creating an art recommendation system.
The autoencoder is part of a deep learning subset within the AI machine learning category. This neural network takes an input and puts it through an encoding process with three layers to obtain a compressed representation, then a decoding process with three layers. Employing an unsupervised machine learning algorithm, the autoencoder employs backpropagation in setting target values that are equal to inputs.
A key benefit of the autoencoder’s approach is that it learns a compressed representation of the input so as to later reconstruct it, which results in dimensionality reduction. It can also help to alleviate issues of sparse data and to denoise input. The latter is useful when an input is partially corrupted and a corresponding uncorrupted input needs to be reconstructed.
Many digital device users may be familiar with the autoencoder as a deep learning algorithm that takes various inputs and generates recommendations, such those found in Netflix, Amazon, or Spotify.
In a related project, Larry Finer estimated a linear regression model that was able to price artworks based on characteristics of the artist and the artwork.
