MatthieuHernandez · MatthieuHernandez · Aug 3, 2025 · Aug 3, 2025 · Aug 12, 2025 · Aug 18, 2025
diff --git a/.clang-tidy b/.clang-tidy
@@ -1,5 +1,6 @@
 Checks: >
     *,
+    -altera-unroll-loops,
     -boost-use-ranges,
     -cppcoreguidelines-non-private-member-variables-in-classes,
     -fuchsia-default-arguments-calls,
@@ -19,7 +20,6 @@ Checks: >
     -cert*
 WarningsAsErrors: >
     *,
-    -altera-unroll-loops,
     -altera-id-dependent-backward-branch,
     -bugprone-easily-swappable-parameters,
     -fuchsia-overloaded-operator,

diff --git a/examples/.clang-tidy b/examples/.clang-tidy
@@ -1,5 +1,6 @@
 Checks: >
     *,
+    -altera-unroll-loops,
     -boost-use-ranges,
     -cppcoreguidelines-non-private-member-variables-in-classes,
     -fuchsia-default-arguments-calls,
@@ -23,7 +24,6 @@ Checks: >
     -readability-magic-numbers
 WarningsAsErrors: >
     *,
-    -altera-unroll-loops,
     -altera-id-dependent-backward-branch,
     -bugprone-easily-swappable-parameters,
     -fuchsia-overloaded-operator,

diff --git a/include/snn/neural_network/NeuralNetwork.hpp b/include/snn/neural_network/NeuralNetwork.hpp
@@ -74,12 +74,14 @@ class NeuralNetwork : public StatisticAnalysis
 
         [[nodiscard]] auto isValid() const -> errorType;
 
+        void resetLearningVariables();
+
         void trainOnce(const std::vector<float>& inputs, const std::vector<float>& desired,
                        const std::vector<float>& weighting = {}, bool temporalReset = true);
 
         auto operator==(const NeuralNetwork& neuralNetwork) const -> bool;
-        auto operator!=(const NeuralNetwork& neuralNetwork) const -> bool;
 };
+static_assert(LearningObject<NeuralNetwork>);
 
 template <class Archive>
 void NeuralNetwork::serialize(Archive& archive, [[maybe_unused]] const uint32_t version)

diff --git a/include/snn/neural_network/StatisticAnalysis.hpp b/include/snn/neural_network/StatisticAnalysis.hpp
@@ -76,8 +76,7 @@ class StatisticAnalysis
         [[nodiscard]] auto getMeanAbsoluteErrorMin() const -> float;
         [[nodiscard]] auto getRootMeanSquaredErrorMin() const -> float;
 
-        auto operator==(const StatisticAnalysis& other) const -> bool;
-        auto operator!=(const StatisticAnalysis& other) const -> bool;
+        auto operator==(const StatisticAnalysis& other) const -> bool = default;
 };
 
 template <class Archive>

diff --git a/include/snn/neural_network/StraightforwardNeuralNetwork.hpp b/include/snn/neural_network/StraightforwardNeuralNetwork.hpp
@@ -93,7 +93,6 @@ class StraightforwardNeuralNetwork final : public internal::NeuralNetwork
         }
 
         auto operator==(const StraightforwardNeuralNetwork& neuralNetwork) const -> bool;
-        auto operator!=(const StraightforwardNeuralNetwork& neuralNetwork) const -> bool;
 };
 
 template <logLevel T>

diff --git a/include/snn/neural_network/binary_classification.hpp b/include/snn/neural_network/binary_classification.hpp
@@ -21,6 +21,7 @@ struct binaryClassification
             archive& this->trueNegative;
             archive& this->falsePositive;
             archive& this->falseNegative;
+            archive& this->totalError;
         }
 };
 }  // namespace snn::internal
diff --git a/include/snn/neural_network/layer/BaseLayer.hpp b/include/snn/neural_network/layer/BaseLayer.hpp
@@ -47,9 +47,10 @@ class BaseLayer
 
         [[nodiscard]] virtual auto isValid() const -> errorType = 0;
 
+        virtual void resetLearningVariables(int batchSize) = 0;
+
         [[nodiscard]] virtual auto summary() const -> std::string = 0;
 
         virtual auto operator==(const BaseLayer& layer) const -> bool = 0;
-        virtual auto operator!=(const BaseLayer& layer) const -> bool = 0;
 };
 }  // namespace snn::internal
diff --git a/include/snn/neural_network/layer/Convolution1D.hpp b/include/snn/neural_network/layer/Convolution1D.hpp
@@ -29,7 +29,6 @@ class Convolution1D final : public Convolution
         [[nodiscard]] auto summary() const -> std::string final;
 
         auto operator==(const BaseLayer& layer) const -> bool final;
-        auto operator!=(const BaseLayer& layer) const -> bool final;
 };
 
 template <class Archive>

diff --git a/include/snn/neural_network/layer/Convolution2D.hpp b/include/snn/neural_network/layer/Convolution2D.hpp
@@ -29,7 +29,6 @@ class Convolution2D final : public Convolution
         [[nodiscard]] auto summary() const -> std::string final;
 
         auto operator==(const BaseLayer& layer) const -> bool final;
-        auto operator!=(const BaseLayer& layer) const -> bool final;
 };
 
 template <class Archive>

diff --git a/include/snn/neural_network/layer/FilterLayer.hpp b/include/snn/neural_network/layer/FilterLayer.hpp
@@ -41,7 +41,6 @@ class FilterLayer : public Layer<SimpleNeuron>
         [[nodiscard]] auto isValid() const -> errorType override;
 
         auto operator==(const BaseLayer& layer) const -> bool override;
-        auto operator!=(const BaseLayer& layer) const -> bool override;
 };
 
 template <class Archive>

diff --git a/include/snn/neural_network/layer/Layer.hpp b/include/snn/neural_network/layer/Layer.hpp
@@ -68,8 +68,9 @@ class Layer : public BaseLayer
 
         [[nodiscard]] auto isValid() const -> errorType override;
 
+        void resetLearningVariables(int batchSize) final;
+
         auto operator==(const BaseLayer& layer) const -> bool override;
-        auto operator!=(const BaseLayer& layer) const -> bool override;
 };
 
 template <BaseNeuron N>

diff --git a/include/snn/neural_network/layer/Layer.tpp b/include/snn/neural_network/layer/Layer.tpp
@@ -150,6 +150,15 @@ auto Layer<N>::getNumberOfParameters() const -> int
     return sum;
 }
 
+template <BaseNeuron N>
+void Layer<N>::resetLearningVariables(int batchSize)
+{
+    for (auto& neuron : this->neurons)
+    {
+        neuron.resetLearningVariables(batchSize);
+    }
+}
+
 template <BaseNeuron N>
 auto Layer<N>::operator==(const BaseLayer& layer) const -> bool
 {
@@ -175,10 +184,4 @@ auto Layer<N>::operator==(const BaseLayer& layer) const -> bool
         return false;
     }
 }
-
-template <BaseNeuron N>
-auto Layer<N>::operator!=(const BaseLayer& layer) const -> bool
-{
-    return !(*this == layer);
-}
 }  // namespace snn::internal
diff --git a/include/snn/neural_network/layer/LayerFactory.hpp b/include/snn/neural_network/layer/LayerFactory.hpp
@@ -18,7 +18,7 @@ extern auto Input(TInt... sizeOfInput) -> LayerModel
                      .numberOfNeurons = 0,
                      .numberOfOutputs = 0,
                      .neuron = {.numberOfInputs = 0,
-                                .batchSize = 0,
+                                .numberOfUses = 0,
                                 .numberOfWeights = 0,
                                 .bias = 0,
                                 .activationFunction = activation::identity},
@@ -40,7 +40,7 @@ auto FullyConnected(int numberOfNeurons, activation activation = activation::sig
                      .numberOfNeurons = numberOfNeurons,
                      .numberOfOutputs = -1,
                      .neuron = {.numberOfInputs = -1,
-                                .batchSize = -1,
+                                .numberOfUses = -1,
                                 .numberOfWeights = -1,
                                 .bias = 1.0F,
                                 .activationFunction = activation},
@@ -61,7 +61,7 @@ auto Recurrence(int numberOfNeurons, activation activation = activation::tanh, T
                      .numberOfNeurons = numberOfNeurons,
                      .numberOfOutputs = -1,
                      .neuron = {.numberOfInputs = -1,
-                                .batchSize = -1,
+                                .numberOfUses = -1,
                                 .numberOfWeights = -1,
                                 .bias = 1.0F,
                                 .activationFunction = activation},
@@ -84,7 +84,7 @@ auto GruLayer(int numberOfNeurons, TOptimizer... optimizers) -> LayerModel
                      .neuron =
                          {
                              .numberOfInputs = -1,
-                             .batchSize = -1,
+                             .numberOfUses = -1,
                              .numberOfWeights = -1,
                              .bias = 1.0F,
                              .activationFunction = activation::tanh,
@@ -106,7 +106,7 @@ auto MaxPooling(int kernelSize) -> LayerModel
                      .numberOfNeurons = 0,
                      .numberOfOutputs = -1,
                      .neuron = {.numberOfInputs = 0,
-                                .batchSize = 0,
+                                .numberOfUses = 0,
                                 .numberOfWeights = 0,
                                 .bias = 0.0F,
                                 .activationFunction = activation::identity},
@@ -128,7 +128,7 @@ auto LocallyConnected(int numberOfLocallyConnected, int kernelSize, activation a
                      .numberOfNeurons = -1,
                      .numberOfOutputs = -1,
                      .neuron = {.numberOfInputs = -1,
-                                .batchSize = -1,
+                                .numberOfUses = -1,
                                 .numberOfWeights = -1,
                                 .bias = almostZero,
                                 .activationFunction = activation},
@@ -149,14 +149,11 @@ auto Convolution(int numberOfConvolution, int kernelSize, activation activation
                      .numberOfInputs = -1,
                      .numberOfNeurons = 1,
                      .numberOfOutputs = -1,
-                     .neuron =
-                         {
-                             .numberOfInputs = -1,
-                             .batchSize = -1,
-                             .numberOfWeights = -1,
-                             .bias = bias,
-                             .activationFunction = activation,
-                         },
+                     .neuron = {.numberOfInputs = -1,
+                                .numberOfUses = -1,
+                                .numberOfWeights = -1,
+                                .bias = bias,
+                                .activationFunction = activation},
                      .numberOfFilters = numberOfConvolution,
                      .numberOfKernels = -1,
                      .numberOfKernelsPerFilter = -1,

diff --git a/include/snn/neural_network/layer/LocallyConnected1D.hpp b/include/snn/neural_network/layer/LocallyConnected1D.hpp
@@ -34,7 +34,6 @@ class LocallyConnected1D final : public FilterLayer
         [[nodiscard]] auto summary() const -> std::string final;
 
         auto operator==(const BaseLayer& layer) const -> bool final;
-        auto operator!=(const BaseLayer& layer) const -> bool final;
 };
 
 template <class Archive>

diff --git a/include/snn/neural_network/layer/LocallyConnected2D.hpp b/include/snn/neural_network/layer/LocallyConnected2D.hpp
@@ -34,7 +34,6 @@ class LocallyConnected2D final : public FilterLayer
         [[nodiscard]] auto summary() const -> std::string final;
 
         auto operator==(const BaseLayer& layer) const -> bool final;
-        auto operator!=(const BaseLayer& layer) const -> bool final;
 };
 
 template <class Archive>

diff --git a/include/snn/neural_network/layer/MaxPooling1D.hpp b/include/snn/neural_network/layer/MaxPooling1D.hpp
@@ -37,7 +37,6 @@ class MaxPooling1D final : public FilterLayer
         [[nodiscard]] auto summary() const -> std::string final;
 
         auto operator==(const BaseLayer& layer) const -> bool final;
-        auto operator!=(const BaseLayer& layer) const -> bool final;
 };
 
 template <class Archive>

diff --git a/include/snn/neural_network/layer/MaxPooling2D.hpp b/include/snn/neural_network/layer/MaxPooling2D.hpp
@@ -37,7 +37,6 @@ class MaxPooling2D final : public FilterLayer
         [[nodiscard]] auto summary() const -> std::string final;
 
         auto operator==(const BaseLayer& layer) const -> bool final;
-        auto operator!=(const BaseLayer& layer) const -> bool final;
 };
 
 template <class Archive>

diff --git a/include/snn/neural_network/layer/SimpleLayer.hpp b/include/snn/neural_network/layer/SimpleLayer.hpp
@@ -35,7 +35,6 @@ class SimpleLayer : public Layer<N>
         [[nodiscard]] auto isValid() const -> errorType final;
 
         auto operator==(const BaseLayer& layer) const -> bool override;
-        auto operator!=(const BaseLayer& layer) const -> bool override;
 };
 
 template <BaseNeuron N>

diff --git a/include/snn/neural_network/layer/SimpleLayer.tpp b/include/snn/neural_network/layer/SimpleLayer.tpp
@@ -89,10 +89,4 @@ auto SimpleLayer<N>::operator==(const BaseLayer& layer) const -> bool
 {
     return Layer<N>::operator==(layer);
 }
-
-template <BaseNeuron N>
-auto SimpleLayer<N>::operator!=(const BaseLayer& layer) const -> bool
-{
-    return !(*this == layer);
-}
 }  // namespace snn::internal
diff --git a/include/snn/neural_network/layer/neuron/BaseNeuron.hpp b/include/snn/neural_network/layer/neuron/BaseNeuron.hpp
@@ -2,6 +2,7 @@
 #include <vector>
 
 #include "../../optimizer/NeuralNetworkOptimizer.hpp"
+#include "LearningObject.hpp"
 
 namespace snn::internal
 {
@@ -32,10 +33,9 @@ concept HasCommonConstMethods = requires(const N neuron) {
     { neuron.getNumberOfInputs() } -> std::same_as<int>;
 
     { neuron.operator==(neuron) } -> std::same_as<bool>;
-    { neuron.operator!=(neuron) } -> std::same_as<bool>;
 };
 
 template <class N>
-concept BaseNeuron =
-    HasCommonMethods<N> && HasCommonConstMethods<N> && (HasNonTemporalOuputMethod<N> || HasTemporalOuputMethod<N>);
+concept BaseNeuron = HasCommonMethods<N> && HasCommonConstMethods<N> &&
+                     (HasNonTemporalOuputMethod<N> || HasTemporalOuputMethod<N>) && LearningObject<N>;
 }  // namespace snn::internal
diff --git a/include/snn/neural_network/layer/neuron/Circular.hpp b/include/snn/neural_network/layer/neuron/Circular.hpp
@@ -1,7 +1,4 @@
 #pragma once
-#include <boost/serialization/access.hpp>
-#include <boost/serialization/base_object.hpp>
-#include <cstdint>
 #include <vector>
 
 namespace snn::internal
@@ -11,10 +8,6 @@ class Circular final
 {
     private:
         friend class Circular<std::vector<float>>;
-        friend class boost::serialization::access;
-        template <class Archive>
-        void serialize(Archive& archive, uint32_t version);
-
         std::vector<T> queue;
         size_t indexPush = 0;
         size_t indexGet = 0;
@@ -29,6 +22,7 @@ class Circular final
         ~Circular() = default;
 
         void initialize(size_t queueSize, size_t dataSize = 1);  // Should be call after the ctor.
+        void reset();                                            // Do the same as initialize.
 
         [[nodiscard]] auto getBack() -> const T*;
         [[nodiscard]] auto getSum() const -> T;
@@ -39,21 +33,17 @@ class Circular final
         auto operator<=>(const Circular<T>& other) const = default;
 };
 
-template <typename T>
-template <class Archive>
-void Circular<T>::serialize(Archive& archive, [[maybe_unused]] const uint32_t version)
-{
-    archive & queue;
-    archive & indexGet;
-    archive & indexPush;
-    archive & divider;
-}
+template <>
+void Circular<float>::initialize(size_t queueSize, size_t dataSize);
+
+template <>
+void Circular<std::vector<float>>::initialize(size_t queueSize, size_t dataSize);
 
 template <>
-void Circular<float>::initialize(size_t size, size_t dataSize);
+void Circular<float>::reset();
 
 template <>
-void Circular<std::vector<float>>::initialize(size_t size, size_t dataSize);
+void Circular<std::vector<float>>::reset();
 
 template <>
 auto Circular<float>::getSum() const -> float;

diff --git a/include/snn/neural_network/layer/neuron/GatedRecurrentUnit.hpp b/include/snn/neural_network/layer/neuron/GatedRecurrentUnit.hpp
@@ -21,7 +21,6 @@ class GatedRecurrentUnit final
         int numberOfInputs{};
 
         float previousOutput = 0;
-        float recurrentError = 0;
         float updateGateOutput = 0;
         float outputGateOutput = 0;
 
@@ -51,19 +50,16 @@ class GatedRecurrentUnit final
         [[nodiscard]] auto getOptimizer() const -> NeuralNetworkOptimizer*;
         void setOptimizer(std::shared_ptr<NeuralNetworkOptimizer> newOptimizer);
 
+        void resetLearningVariables(int batchSize);
+
         auto operator==(const GatedRecurrentUnit& neuron) const -> bool;
-        auto operator!=(const GatedRecurrentUnit& neuron) const -> bool;
 };
+static_assert(BaseNeuron<GatedRecurrentUnit>);
 
 template <class Archive>
 void GatedRecurrentUnit::serialize(Archive& archive, [[maybe_unused]] const uint32_t version)
 {
-    archive& this->errors;
     archive& this->numberOfInputs;
-    archive& this->previousOutput;
-    archive& this->recurrentError;
-    archive& this->updateGateOutput;
-    archive& this->outputGateOutput;
     archive& this->resetGate;
     archive& this->updateGate;
     archive& this->outputGate;

diff --git a/include/snn/neural_network/layer/neuron/LearningObject.hpp b/include/snn/neural_network/layer/neuron/LearningObject.hpp
@@ -0,0 +1,14 @@
+#include <concepts>
+
+template <typename T>
+concept WithoutBatchSize = requires(T t) {
+    { t.resetLearningVariables() } -> std::same_as<void>;
+};
+
+template <typename T>
+concept WithBatchSize = requires(T t, int batchSize) {
+    { t.resetLearningVariables(batchSize) } -> std::same_as<void>;
+};
+
+template <typename T>
+concept LearningObject = WithoutBatchSize<T> || WithBatchSize<T>;
-Original file line number
+Diff line change
@@ Expand Up @@
             }
             auto operator==(const StraightforwardNeuralNetwork& neuralNetwork) const -> bool;
-            auto operator!=(const StraightforwardNeuralNetwork& neuralNetwork) const -> bool;
     };
     template <logLevel T>
@@ Expand Down @@