Backwards - ReLU & Dense Lasyer

Made the backwards for dense layers and ReLU, but I think I soon need to refactor some code to make it all steamlined.

include/modules/neural_networks/activation_functions/Activation_ReLU.h

@@ -11,12 +11,29 @@ namespace neural_networks{
 	template <typename T>
 	struct Activation_ReLU{
 
-		//utils::Matrix<T> inputs;
+		utils::Matrix<T> _inputs;
 		utils::Matrix<T> outputs;
 
+		utils::Matrix<T> dinputs;
+
         void forward(const utils::Matrix<T>& inputs){
+        	_inputs = inputs;
 		    outputs = numerics::matclip_low(inputs, T{0});
         }
+        void backward(const utils::Matrix<T>& dvalues){
+        	// Since we need to modify the original variable,
+        	// let's make a copy of the values first
+        	dinputs = dvalues;
+
+        	// Zero gradients where input values were negative
+        	for (uint64_t i = 0; i < dinputs.rows(); ++i){
+        		for (uint64_t j = 0; j < dinputs.cols(); ++j){
+        			if (_inputs(i,j) <= T{0}){
+        				dinputs(i,j) = T{0};
+        			}
+        		}
+        	}
+        }
 	};