Learning backend cuda

Context

Support backend cuda for learning

Modified files

• SGD.hpp, make SGD optimizer support backend_cuda;
• Accuracy.hpp, Accuracy.cpp and pybind_Accuracy.hpp add Accuracy metric graph;

TODO

• Support backend_cuda for SGD
• Support Accuracy computation on backend_cuda

include/aidge/learning/metrics/Accuracy.hpp

+/********************************************************************************
+ * Copyright (c) 2024 CEA-List
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Eclipse Public License 2.0 which is available at
+ * http://www.eclipse.org/legal/epl-2.0.
+ *
+ * SPDX-License-Identifier: EPL-2.0
+ *
+ ********************************************************************************/
+
+#ifndef AIDGE_LEARNING_METRICS_ACCURACY_H_
+#define AIDGE_LEARNING_METRICS_ACCURACY_H_
+
+#include <cstddef>  // std::size_t
+#include <memory>
+
+#include "aidge/data/Tensor.hpp"
+
+namespace Aidge {
+namespace metrics {
+
+/**
+ * @brief Compute the Accuracy.
+ * This function returns the loss and set the ``grad()`` of the prediction
+ * input.
+ * @param prediction Tensor returned by the Aidge Graph, it is important that
+ * this tensor is not a copy as overwise the backward function will not have a
+ * gradient to start.
+ * @param target Tensor representing the ground truth, it must be one hot encoded.
+ * @param axis The classes axis.
+ */
+Tensor Accuracy(std::shared_ptr<Tensor>& prediction,
+                const std::shared_ptr<Tensor>& target,
+                std::int32_t axis);
+}  // namespace Metrics
+}  // namespace Aidge
+
+#endif /* AIDGE_LEARNING_METRICS_ACCURACY_H_ */