2025 USA-NA-AIO Round 1, Problem 2, Part 11

AI Olympiads 2025 USA-NA-AIO Round 1
1

Part 11 (10 points, coding task)

It is known by math that the combination of several linear layers can still be seen as an linear layer, so we can add some non-linear activation functions, such as ReLU, in between to get better effect.

Multi-Layer Perceptron (MLP), is such a neural network composed of multiple fully connected layers with non-linear activations, commonly used in deep learning.

Please define a class called My_MLP_Model that subclasses nn.Module and works in the following ways:

  • The architecture consists of two hidden layers and one output layer.

  • Each hidden layer consists of an affine transformation module and a ReLU activiation module.

  • Each affine transformation module shall be initialized with the build-in class nn.Linear.

  • Each ReLU activiatio nmodule shall be initialized with your self-defined class My_ReLU.

1 Like
2 
### WRITE YOUR SOLUTION HERE ###

class My_MLP_Model(nn.Module):
    def __init__(self, in_features, hidden_features1, hidden_features2, out_features):
        super().__init__()
        self.in_features = in_features
        self.hidden_features1 = hidden_features1
        self.hidden_features2 = hidden_features2
        self.out_features = out_features
        self.seq = nn.Sequential(
            nn.Linear(in_features, hidden_features1),
            My_ReLU(),
            nn.Linear(hidden_features1, hidden_features2),
            My_ReLU(),
            nn.Linear(hidden_features2, out_features)
        )

    def forward(self, x):
        x = self.seq(x)
        return x

""" END OF THIS PART """
1 Like
3

Does this work?

class My_MLP_Model(nn.Module):
  def __init__(self, in_features, hidden1_features, hidden2_features, out_features):
    super().__init__()
    self.fc1 = nn.Linear(in_features, hidden1_features)
    self.relu1 = My_ReLU()
    self.fc2 = nn.Linear(hidden1_features, hidden2_features)
    self.relu2 = My_Relu()
    self.fc3 = nn.Linear(hidden2_features, out_features)

  def forward(self, X):
    X = self.relu1(self.fc1(X))
    X = self.relu2(self.fc2(X))
    X = self.fc3(X)
    return X