explanations-can-be-manipulated-and-geometry-is-to-blame.md

 # Explanations can be manipulated and geometry is to blame
 **Keys:**
-<img src="uploads/expl_manipulate_fig1.png" width="300" style="float: right;">
 * Adversarial attacks on explanation maps: 
   * they can be changed to an arbitrary target map by applying visually hardly perceptible input pertubation
   * The pertubation does not change the output of the network for that input
-  * $`\rightarrow`$ explanation maps are not robustly interpretable
-
-
-
+  * $`\rightarrow`$ explanation maps are not robustly interpretable  
+    <img src="uploads/expl_manipulate_fig1.png" width="300">
 * This phenomenon is related to geometry of the networks output manifold
 * We can derive a bound on the degree of possible manipulation. The bound is proportional to two differential geometric quantities:
   * principle curvatures