diff --git a/metalearner.py b/metalearner.py
index d906833..55ff675 100644
--- a/metalearner.py
+++ b/metalearner.py
@@ -20,7 +20,7 @@ def __init__(self, input_size, hidden_size, n_learner_params):
         self.n_learner_params = n_learner_params
         self.WF = nn.Parameter(torch.Tensor(input_size + 2, hidden_size))
         self.WI = nn.Parameter(torch.Tensor(input_size + 2, hidden_size))
-        self.cI = nn.Parameter(torch.Tensor(n_learner_params, 1))
+        self.cI = nn.Parameter(torch.Tensor(n_learner_params, 1), requires_grad=False) # Freeze the parameters to avoide getting optimized by gradient descent (Adam)
         self.bI = nn.Parameter(torch.Tensor(1, hidden_size))
         self.bF = nn.Parameter(torch.Tensor(1, hidden_size))