diff --git a/content/03.overview.md b/content/03.overview.md index 59fba52..348d207 100644 --- a/content/03.overview.md +++ b/content/03.overview.md @@ -130,4 +130,6 @@ Context-aware models can be organized along a spectrum of assumptions about the * **Smooth models**: $\theta_i = f(c_i)$, with $f$ assumed to be continuous or low-complexity. * **Latent models**: $\theta_i \sim P(\theta \mid c_i)$, with $f$ learned implicitly. +![A spectrum of context awareness in modeling, showing global, grouped, smooth, and latent models](images/spectrum_context.png){#fig:spectrum-context width="70%"} + Each formulation encodes different beliefs about parameter variation. The next section formalizes these principles and examines general strategies for adaptivity in statistical modeling. For a discussion of how subpopulation shifts influence generalization, see [@arXiv:2106.04486]. diff --git a/content/images/spectrum_context.html b/content/images/spectrum_context.html new file mode 100644 index 0000000..b155dda --- /dev/null +++ b/content/images/spectrum_context.html @@ -0,0 +1,601 @@ + + + + + + Spectrum of Context-Awareness + + + +
+
+
+
Global
+
θi = θ
+
+ +
+
Single shared
parameter for all
+
+ +
+
Grouped
+
θi = θc
+
+ +
+
Discrete clusters,
finite partitions
+
+ +
+
Smooth
+
θi = f(ci)
+
+ +
+
Continuous mapping,
regularized structure
+
+ +
+
Latent
+
θi ~ P(θ | ci)
+
+ +
+
Learned implicit
representations
+
+
+
+ + + + \ No newline at end of file diff --git a/content/images/spectrum_context.png b/content/images/spectrum_context.png new file mode 100644 index 0000000..396b203 Binary files /dev/null and b/content/images/spectrum_context.png differ