Ryan Adams and I just uploaded to the arXiv our paper “High-Dimensional Probability Estimation with Deep Density Models”. In this work, we introduce the deep density model (DDM), a new approach for density estimation.
In the spirit of Ryan’s most recent post, I will discuss a fundamental snippet from numerical linear algebra that facilitates computation for the same price of not facilitating it. In our everyday lives, we often come across theoretical expressions that involve matrix inverses stapled to vectors, such as with . When we proceed to code this up, it is very tempting to first compute . Resist doing this! There are several points for why there is no point to actually find an explicit, tangible inverse.
Before diving into more technical posts, I want to briefly touch on some basic questions, and the big picture behind unsupervised learning. I also want to do a bit of handwaving on sparsity—a topic that has gotten a lot of attention recently. Let’s say we are given observations . These points are assumed to contain some underlying structure, which we seek to capture in order to perform tasks such as classification or compression. We can apply our algorithms on the data in their raw form—which carries unidentified redundancy—and hope for the best. However, a more sensible approach would be to first