diff --git a/NN/nnapplygrads.m b/NN/nnapplygrads.m
index 781163b..649258d 100644
--- a/NN/nnapplygrads.m
+++ b/NN/nnapplygrads.m
@@ -10,7 +10,12 @@
             dW = nn.dW{i};
         end
         
-        dW = nn.learningRate * dW;
+        % to apply different learning rates to each layer
+        if isempty(nn.learningRatePerLayer)
+            dW = nn.learningRate * dW;
+        else
+            dW = nn.learningRatePerLayer(i) * dW;
+        end
         
         if(nn.momentum>0)
             nn.vW{i} = nn.momentum*nn.vW{i} + dW;
diff --git a/NN/nnsetup.m b/NN/nnsetup.m
index b8ec742..7a1162d 100644
--- a/NN/nnsetup.m
+++ b/NN/nnsetup.m
@@ -8,6 +8,7 @@
     
     nn.activation_function              = 'tanh_opt';   %  Activation functions of hidden layers: 'sigm' (sigmoid) or 'tanh_opt' (optimal tanh).
     nn.learningRate                     = 2;            %  learning rate Note: typically needs to be lower when using 'sigm' activation function and non-normalized inputs.
+    nn.learningRatePerLayer             = [];           %  learning rate per layer - for transfer learning pre-training and fine-tuning different parts of the network (should be of length nn.n - 1)
     nn.momentum                         = 0.5;          %  Momentum
     nn.scaling_learningRate             = 1;            %  Scaling factor for the learning rate (each epoch)
     nn.weightPenaltyL2                  = 0;            %  L2 regularization
diff --git a/NN/nntrain.m b/NN/nntrain.m
index af844a6..1e2d1e8 100644
--- a/NN/nntrain.m
+++ b/NN/nntrain.m
@@ -72,6 +72,17 @@
         
     disp(['epoch ' num2str(i) '/' num2str(opts.numepochs) '. Took ' num2str(t) ' seconds' '. Mini-batch mean squared error on training set is ' num2str(mean(L((n-numbatches):(n-1)))) str_perf]);
     nn.learningRate = nn.learningRate * nn.scaling_learningRate;
+    if ~isempty(nn.learningRatePerLayer)
+        nn.learningRatePerLayer = nn.learningRatePerLayer * nn.scaling_learningRate;
+    end
+    
+    if isfield(opts,'tol')
+       if opts.validation == 1 && loss.val.e(end)<opts.tol
+           break;
+       elseif loss.train.e(end)<opts.tol
+           break;
+       end
+    end
 end
 end