MNIST

Description

The MNIST database of handwritten digits, available from this page, has a training set of 60,000 examples, and a test set of 10,000 examples. It is a subset of a larger set available from NIST. The digits have been size-normalized and centered in a fixed-size image.

See more details

Neural network

This is the test with the neural network architecture used to obtain up to 98.32% accuracy on this dataset. To reach this accuracy you may need more attempts and more learning time.

TEST_F(MnistTest, convolutionalNeuralNetwork)
{
    StraightforwardNeuralNetwork neuralNetwork({
        Input(28, 28, 1),
        Convolution(1,5, activation::GELU, ErrorMultiplier(10.0f)),
        FullyConnected(70),
        FullyConnected(10)
        });
    neuralNetwork.train(*data, 1_ep || 15_s);
    auto accuracy = neuralNetwork.getGlobalClusteringRate();
    ASSERT_ACCURACY(accuracy, 0.93f);
}

See the code