Given a set of sample points: discover their defining function
+given $\{(x_1,y_1), \ldots, (x_K,y_K)\}$, find $f(x) = y$
+Fit data to a predefined equation / family of equations.
+$RELU$: Discontinuity Function
+One big mega-function to fit to the data
+In general: Pre-specify layering structure as a NN workflow.
+Not looking at how to train a NN!
+Why are NNs dangerous?
+Why does it make the choices it does?
+Today's discussion:
+Observation 1: Describing the entire model concisely is hard
+Observation 2: Describing the model on a single input is easier
+Contrast with dynamic slicing
+Given a target point, figure out which of the point's features are most responsible for the classification.
+“Why Should I Trust You?” Explaining the Predictions of Any Classifier
+Ribero, Sing, Guestrin
+Arguments to $f$ are input features. For example:
+ +Note: The source model doesn't have to be an NN.
+Focusing on similar inputs, learn an explainable model $g$.
+ +Define similarity by a distance function:
+ $$\pi_x : R^d \times R^d \rightarrow [0,1]$$ +Overall Goal: Find a $g$ that minimizes $\mathcal L(f, g, \pi_x) + \Omega(g)$
+Boolean- (not Real-)valued features.
+Pick a $g$ (resp., $\{w_i\}$) that minimizes this!
+Each feature is a word: '1' if the word is present, '0' if not.
($x = x'$)
Complexity Function: (at most $K$ features) + $\Omega(g) = \begin{cases} 0 & \textbf{if }|\{w_i > 0\}| > K \\ \infty & \textbf{otherwise}\end{cases}$ +
+ +Simplified model: Find the $K$ features most responsible for differentiating between the target class.
+Each feature is a "superpixel" (a contiguous region of similarly colored pixels): $x_i$ is the color of the pixel; $x_i'$ is 1 if the superpixel is identical to the original.
+ +Complexity Function: (at most $K$ superpixels) + "Lasso" (as in the lasso tool) pixels together into a contiguous region of no more than $K$ superpixels. +
+ +Simplified model: Find the $K$ superpixels most responsible for differentiating between the target class.
+If the model has the features identified by $g$, it is probably of the target class.
+ACTIVIS: Visual Exploration of Industry-Scale +Deep Neural Network Models
+Kahng, Andrews, Kalro, Chau
+Solution: Let users see the nodes themselves.
+