Updated readme

readme.md

@@ -3,9 +3,7 @@ Automatic Model Construction with Gaussian Processes
 
 <img src="https://raw.githubusercontent.com/duvenaud/phd-thesis/master/figures/topology/mobius.png" width="200"> <img src="https://raw.githubusercontent.com/duvenaud/phd-thesis/master/figures/additive/3d-kernel/3d_add_kernel_321.png" width="200"> <img src="https://raw.githubusercontent.com/duvenaud/phd-thesis/master/figures/deep-limits/map_connected/latent_coord_map_layer_39.png" width="200">
 
-
-
-I'm about 10 days from submitting my thesis.  In the spirit of radical openness, I've made the entire repo public.  Think I'm missing something?  Want me to cite you?  <a href="mailto: [email protected]">Let me know!</a>  Any feedback would be much appreciated.
+Defended on June 26th, 2014.
 
 Individual chapters:
 
@@ -16,6 +14,7 @@ Individual chapters:
 5. [Deep Gaussian Processes](deeplimits.pdf)
 6. [Additive Gaussian Processes](additive.pdf)
 7. [Warped Mixture Models](warped.pdf)
+7. [Discussion](discussion.pdf)
 
 Or get the whole thing in [one big PDF](thesis.pdf)
 
@@ -24,23 +23,22 @@ Or get the whole thing in [one big PDF](thesis.pdf)
 Abstract:
 ----------
 
-This thesis shows how to automatically construct and describe a large class of models useful for forecasting and finding structure in domains such as time series, geological formations, and physical dynamics.
+This thesis develops a method for automatically constructing, visualizing and describing a large class of models, useful for forecasting and finding structure in domains such as time series, geological formations, and physical dynamics.
 These models, based on Gaussian processes, can capture many types of statistical structure, such as periodicity, changepoints, additivity, and symmetries.
-Such structure can be encoded through a *kernel*, which has historically been chosen by hand by experts.
-We show how to automate this task, creating a system which explores a large space of models and reports the structures discovered.
+Such structure can be encoded through *kernels*, which have historically been hand-chosen by experts.
+We show how to automate this task, creating a system that explores an open-ended space of models and reports the structures discovered.
 
-The introductory chapters show how to express many types of structure through kernels, and how combining together different kernels combines their properties.
-Among several examples, we show how composite kernels can produce priors over topological manifolds such as cylinders, toruses, and Mobius strips, as well as their higher-dimensional analogues.
+To automatically construct Gaussian process models, we search over sums and products of kernels, maximizing the approximate marginal likelihood.
+We show how any model in this class can be automatically decomposed into qualitatively different parts, and how each component can be visualized and described through text.
+We combine these results into a procedure that, given a dataset, automatically constructs a model along with a detailed report containing plots and generated text that illustrate the structure discovered in the data.
 
-To automatically search over an open-ended space of models, we define a simple grammar over kernels, a search criterion (marginal likelihood), and a breadth-first search procedure.
-Combining these, we present a procedure which takes in a dataset and outputs an automatically-constructed model, along with a detailed report with graphs and automatically generated text illustrating the qualitatively different, and sometimes novel, types of structure discovered in that dataset.
-This system automates parts of the model-building and analysis currently performed by expert statisticians.
+The introductory chapters contain a tutorial showing how to express many types of structure through kernels, and how adding and multiplying different kernels combines their properties.
+Examples also show how symmetric kernels can produce priors over topological manifolds such as cylinders, toruses, and Mobius strips, as well as their higher-dimensional generalizations.
 
 This thesis also explores several extensions to Gaussian process models.
-First, building on earlier work relating Gaussian processes and neural nets, we explore the natural extensions of these models to *deep kernels* and *deep Gaussian processes*.
-Second, we examine the model class consisting of the sum of functions of all possible combinations of input variables.
-We show a close connection between this model class and the recently-developed regularization method of *dropout*.
-Third, we combine Gaussian processes with the Dirichlet process to produce the *warped mixture model* -- a Bayesian clustering model with nonparametric cluster shapes, and a corresponding latent space in which each cluster has an interpretable parametric form.
+First, building on existing work that relates Gaussian processes and neural nets, we analyze natural extensions of these models to *deep kernels* and *deep Gaussian processes*.
+Second, we examine *additive Gaussian processes*, showing their relation to the regularization method of *dropout*.
+Third, we combine Gaussian processes with the Dirichlet process to produce the *warped mixture model*: a Bayesian clustering model having nonparametric cluster shapes, and a corresponding latent space in which each cluster has an interpretable parametric form.
 
 
 Source Code for Experiments