diff --git a/unit04_model_sel/lab_neural_partial.ipynb b/unit04_model_sel/lab_neural_partial.ipynb
index 694d38c0..d95b28eb 100644
--- a/unit04_model_sel/lab_neural_partial.ipynb
+++ b/unit04_model_sel/lab_neural_partial.ipynb
@@ -192,11 +192,11 @@
     "\n",
     "One way we can improve the model accuracy is to used delayed version of the features.  Specifically, the model we used above mapped the features \n",
     "\n",
-    "    yhat[i,k] = \\sum_{j=0}^{p-1} X[i,j]*w[j,k] + b[k]\n",
+    " $yhat[i,k] = \\sum_{j=0}^{p-1} X[i,j]*w[j,k] + b[k]$\n",
     "    \n",
     "where `p` is the number of features and `w[j,k]` is a matrix of coefficients.  In this model,  `yhat[i,:]` at time `i` was only dependent on the inputs  `X[i,:]` at time `i`.  In signal processing, this is called a *memoryless* model.  However, in many physical systems, such as those that arise in neuroscience, there is a delay between the inputs `X[i,:]` and the outputs `y[i]`.  For such cases, we can use a model of the form,\n",
     "\n",
-    "    yhat[i+d,k] = \\sum_{k=0}^d  \\sum_{j=0}^{p-1} \\sum_{m=0}^d X[i+m,j]*W[j,m,k] + b[k]\n",
+    " $yhat[i+d,k] = \\sum_{k=0}^d  \\sum_{j=0}^{p-1} \\sum_{m=0}^d X[i+m,j]*W[j,m,k] + b[k]$\n",
     "    \n",
     "where `W` is a 3-dim array of coefficients where:\n",
     "\n",