diff --git a/15_Adaptive_Filtering/Adaptive_Filtering.ipynb b/15_Adaptive_Filtering/Adaptive_Filtering.ipynb
index d1d7705..de801df 100644
--- a/15_Adaptive_Filtering/Adaptive_Filtering.ipynb
+++ b/15_Adaptive_Filtering/Adaptive_Filtering.ipynb
@@ -1,7 +1,7 @@
 {
  "metadata": {
   "name": "",
-  "signature": "sha256:86fc954574672293ddc089433553cec36487eac3099563fcbb3b3a65fc38906d"
+  "signature": "sha256:4194c0e0a4f144a3a622c419fc2d60daaf82f71101e32e53b614d64e571f872d"
  },
  "nbformat": 3,
  "nbformat_minor": 0,
@@ -1428,7 +1428,9 @@
       "\n",
       "As you might imagine this is a broad topic, and there are many ways of designing and implementing MM estimators. But consider a simple approach for the target we have been tracking in this chapter. One idea would be to simultaneously run a constant velocity and a constant acceleration filter, and to switch between their outputs when we detect a manuever by inspecting the residuals. Even this choice gives us many options. Consider the dynamics of a turning object. For example, an automobile turns on a wheelbase - the front wheels turn, and the car pivots around the rear wheels. This is a nonlinear process, so for best results we would want to use some type of nonlinear filter (EKF, UKF, etc) to model the turns. On the other hand, a linear constant velocity filter would perform fine for the steady state portions of the travel. So our bank of filters might consist of a linear KF and an EKF filter for the turns. However, neither is particularly well suited for modelling behaviors such as accelerating and braking. So a highly performing MM estimator might contain a bank of many filters, each designed to perform best for a certain performance envelope of the tracked object.\n",
       "\n",
-      "I trust the idea of switching between filters is clear, but what mathematical foundations should we use to implement it? The problem that we face is trying to detect via noisy meaurements when a change in regime should result in a change in model. \n",
+      "Of course, you do not need to just base your filters on the order of the model. You can use different noise models, different adapters in each. For example, in the section above I showed many plots depicting the effects of changing parameters on the estimate of the velocity and position. Perhaps one setting works better for position, and a different setting for velocity. Put both in your bank of filters. You could then take the best estimates for the position from one filter, and the best estimate for the velocity from a different filter.\n",
+      "\n",
+      "I trust the idea of switching between filters to get the best performance is clear, but what mathematical foundations should we use to implement it? The problem that we face is trying to detect via noisy meaurements when a change in regime should result in a change in model. \n",
       "\n"
      ]
     },