autograph

Neural Networks

#[derive(Layer, Forward)]
#[autograph(forward(Variable4, Output=Variable2))]
struct LeNet5 {
    conv1: Conv2,
    relu1: Relu,
    pool1: MaxPool2,
    conv2: Conv2,
    relu2: Relu,
    pool2: MaxPool2,
    flatten: Flatten,
    dense1: Dense,
    relu3: Relu,
    dense2: Dense,
    relu4: Relu,
    dense3: Dense,
}

impl LeNet5 {
    fn new(device: Device, scalar_type: ScalarType) -> Result<Self> {
        let conv1 = Conv2::builder()
            .device(device.clone())
            .scalar_type(scalar_type)
            .inputs(1)
            .outputs(6)
            .filter([5, 5])
            .build()?;
        let relu1 = Relu;
        let pool1 = MaxPool2::builder().filter([2, 2]).build();
        let conv2 = Conv2::builder()
            .device(device.clone())
            .scalar_type(scalar_type)
            .inputs(6)
            .outputs(16)
            .filter([5, 5])
            .build()?;
        let relu2 = Relu;
        let pool2 = MaxPool2::builder().filter([2, 2]).build();
        let flatten = Flatten;
        let dense1 = Dense::builder()
            .device(device.clone())
            .scalar_type(scalar_type)
            .inputs(16 * 4 * 4)
            .outputs(128)
            .build()?;
        let relu3 = Relu;
        let dense2 = Dense::builder()
            .device(device.clone())
            .scalar_type(scalar_type)
            .inputs(128)
            .outputs(84)
            .build()?;
        let relu4 = Relu;
        let dense3 = Dense::builder()
            .device(device.clone())
            .scalar_type(scalar_type)
            .inputs(84)
            .outputs(10)
            .bias(true)
            .build()?;
        Ok(Self {
            conv1,
            relu1,
            pool1,
            conv2,
            relu2,
            pool2,
            flatten,
            dense1,
            relu3,
            dense2,
            relu4,
            dense3,
        })
    }
}

let mut model = LeNet5::new(device.clone(), ScalarType::F32)?;
model.init_parameter_grads()?;
let y = model.forward(x)?;
let loss = y.cross_entropy_loss(t)?;
loss.backward()?;
model.update(learning_rate, &optimizer)?;

See the Neural Network MNIST example.

Benchmarks

NVIDIA GeForce GTX 1060 with Max-Q Design

LeNet5(training, batch_size = 100)

LeNet5(inference, batch_size = 1,000)

See the Neural Network benchmark.

License

Dual-licensed to be compatible with the Rust project.

Licensed under the Apache License, Version 2.0 http://www.apache.org/licenses/LICENSE-2.0 or the MIT license http://opensource.org/licenses/MIT, at your option. This file may not be copied, modified, or distributed except according to those terms.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.