“Tile2Vec: Unsupervised Representation Learning for Spatially Distributed Data”, Neal Jean, Sherrie Wang, Anshul Samar, George Azzari, David Lobell, Stefano Ermon2018-05-08 (CNN; backlinks)⁠:

Geospatial analysis lacks methods like the word vector representations and pre-trained networks that boost performance across a wide range of natural language and computer vision tasks.

To fill this gap, we introduce Tile2Vec, an unsupervised representation learning algorithm that extends the distributional hypothesis from natural language—words appearing in similar contexts tend to have similar meanings—to spatially distributed data [using CNN embeddings].

We demonstrate empirically that Tile2Vec learns semantically meaningful representations on 3 datasets.

Our learned representations improve performance in downstream classification tasks and, similar to word vectors, visual analogies can be obtained via simple arithmetic in the latent space.

Backlinks: