diff --git a/mobile_app/train_mobile.py b/mobile_app/train_mobile.py
index 59dc628fe34252e6ee872dbac28d3cd70a3fef10..487ed79f07e57c8dc32483963446747ef366c80e 100644
--- a/mobile_app/train_mobile.py
+++ b/mobile_app/train_mobile.py
@@ -54,8 +54,8 @@ if __name__ == "__main__":
     
     mobile_model = NetTensorFlowWrapper(args.sparse_checkpoint, None)
         
-    for param in mobile_model.sparse_model.parameters():
-        param.requires_grad = False
+#     for param in mobile_model.sparse_model.parameters():
+#         param.requires_grad = False
 
     mobile_model.to(device)
     
@@ -107,7 +107,7 @@ if __name__ == "__main__":
     criterion = torch.nn.BCEWithLogitsLoss()
 
     if args.train:
-        prediction_optimizer = torch.optim.Adam(mobile_model.parameters(),
+        prediction_optimizer = torch.optim.Adam(mobile_model.predictive_model.parameters(),
                                                 lr=args.lr)
 
         for epoch in range(args.epochs):
diff --git a/sparse_coding_torch/mobile_model.py b/sparse_coding_torch/mobile_model.py
index 9b6e18d8b9a0ecba986364ce562795d11dc3611d..09476b1d2e72847dd053f6a3e120df66c2656f69 100644
--- a/sparse_coding_torch/mobile_model.py
+++ b/sparse_coding_torch/mobile_model.py
@@ -411,7 +411,7 @@ class NetTensorFlowWrapper(nn.Module):
                        rectifier=True,
                        lam=0.05,
                        max_activation_iter=200,
-                       activation_lr=1e-1)
+                       activation_lr=1e-2)
 
         self.predictive_model = SmallDataClassifierConv3d()