diff --git a/src/TensorFlowNET.Core/Gradients/Tape.ComputeGradient.cs b/src/TensorFlowNET.Core/Gradients/Tape.ComputeGradient.cs
index 0d0ecbe25..73c9e501e 100644
--- a/src/TensorFlowNET.Core/Gradients/Tape.ComputeGradient.cs
+++ b/src/TensorFlowNET.Core/Gradients/Tape.ComputeGradient.cs
@@ -90,7 +90,7 @@ public Tensor[] ComputeGradient(Tensor[] target_tensor_ids,
 
                     in_gradients = trace.backward_function(out_gradients.ToArray(), unneeded_gradients.ToArray());
 
-                    if (in_gradients.Count() != trace.input_tensor_id.Count())
+                    if (in_gradients.Length != trace.input_tensor_id.Length && in_gradients.Length + unneeded_gradients.Count != trace.input_tensor_id.Length)
                         throw new RuntimeError($"Recorded operation '{trace.op_type}' returned too few gradients. Expected {trace.input_tensor_id.Length} but received {in_gradients.Count()}");
                     if (!_persistent)
                     {
@@ -103,9 +103,11 @@ public Tensor[] ComputeGradient(Tensor[] target_tensor_ids,
                     in_gradients = new Tensor[trace.input_tensor_id.Length];
                 }
 
-                for (int i = 0; i < in_gradients.Length; ++i)
+                bool skip_unneeded_id = trace.input_tensor_id.Length > in_gradients.Length;
+                for (int i = 0, k = 0; i < in_gradients.Length && k < trace.input_tensor_id.Count(); ++i, ++k)
                 {
-                    var id = trace.input_tensor_id[i];
+                    if (skip_unneeded_id && unneeded_gradients.Contains(k)) ++k;
+                    var id = trace.input_tensor_id[k];
                     if (in_gradients[i] != null)
                     {
                         var unaggregated_grads = gradients[id];
diff --git a/src/TensorFlowNET.Core/Gradients/math_grad.cs b/src/TensorFlowNET.Core/Gradients/math_grad.cs
index d9bc9b228..22d3c641b 100644
--- a/src/TensorFlowNET.Core/Gradients/math_grad.cs
+++ b/src/TensorFlowNET.Core/Gradients/math_grad.cs
@@ -639,6 +639,20 @@ public static Tensor[] _SqrtGrad(Operation op, Tensor[] grads)
             });
         }
 
+        [RegisterGradient("Rsqrt")]
+        public static Tensor[] _RsqrtGrad(Operation op, Tensor[] grads)
+        {
+            var grad = grads[0];
+            var y = op.outputs[0];
+
+            return tf_with(ops.control_dependencies(grads), delegate
+            {
+                y = math_ops.conj(y);
+                var factor = constant_op.constant(-0.5f, dtype: y.dtype);
+                return new Tensor[] { grad * (factor * math_ops.square(y) * y) };
+            });
+        }
+
         [RegisterGradient("Asin")]
         public static Tensor[] _ASinGrad(Operation op, Tensor[] grads)
         {
diff --git a/src/TensorFlowNET.Core/Keras/Layers/ILayersApi.cs b/src/TensorFlowNET.Core/Keras/Layers/ILayersApi.cs
index 525bfd354..f1860da1b 100644
--- a/src/TensorFlowNET.Core/Keras/Layers/ILayersApi.cs
+++ b/src/TensorFlowNET.Core/Keras/Layers/ILayersApi.cs
@@ -55,6 +55,20 @@ public ILayer Conv2D(int filters,
             IRegularizer bias_regularizer = null,
             IRegularizer activity_regularizer = null);
 
+        public ILayer Conv2DTranspose(int filters,
+            Shape kernel_size = null,
+            Shape strides = null,
+            string output_padding = "valid",
+            string data_format = null,
+            Shape dilation_rate = null,
+            string activation = null,
+            bool use_bias = true,
+            string kernel_initializer = null,
+            string bias_initializer = null,
+            string kernel_regularizer = null,
+            string bias_regularizer = null,
+            string activity_regularizer = null);
+
         public ILayer Conv2D(int filters,
             Shape kernel_size = null,
             Shape strides = null,