Lightning AI Studios: Never set up a local environment again →

Log in or create a free Lightning.ai account to track your progress and access additional course materials  

Deep Learning Fundamentals

  • Pages

    • Final certification exam

    Deep Learning Fundamentals > Unit 7 > Unit 7.4
    Course Progress:

    Unit 7.4 Training Convolutional Neural Networks

    References

    Code

    What we covered in this video lecture

    In this series of coding videos, we implemented our first convolutional neural network from scratch. Then, we implemented the popular ResNet model, from scratch, in the form of ResNet-18. And lastly, we saw how we can load pretrained models from the torchvision hub, which is the recommended way for leveraging existing neural network architectures.

    Additional resources if you want to learn more

    Torchvision is an application library affiliated with PyTorch. If you want to learn more about what models it includes, I recommend checking out their models page. The individual models also contain links to their respective research papers if you want to learn more.

    Log in or create a free Lightning.ai account to access:

    • Quizzes
    • Completion badges
    • Progress tracking
    • Additional downloadable content
    • Additional AI education resources
    • Notifications when new units are released
    • Free cloud computing credits

    Quiz: 7.4 Training Convolutional Neural Networks

    What is the primary purpose of skip connections in a deep neural network?

    Incorrect. The runtime may even be a bit slower since it adds additional operations.

    Incorrect. It doesn’t change the interpretability.

    Correct.

    Incorrect. It does not add or remove parameters.

    Skip connections help to address which problem in deep neural networks?

    Incorrect. Skip connections do not directly address overfitting. Overfitting is typically managed through techniques like regularization, dropout, and early stopping.

    Correct. They help to maintain gradient flow during backpropagation, where gradients can become very small as they are propagated through multiple layers, making it difficult for the model to learn.

    Incorrect. Skip connections do not specifically target reducing computational cost, although they can enable more efficient training of deep networks by facilitating gradient flow.

    Incorrect. Skip connections do not directly address the issue of generalization, which is related to the model’s ability to perform well on new, unseen data.

    Skip connections are used to

    Correct. Skip connections bypass one or more layers in the network, allowing the gradient information to flow more efficiently during backpropagation.

    Incorrect. Skip connections do not reduce the number of layers; they bypass one or more layers.

    Incorrect. Skip connections do not directly increase the number of layers, but they enable the training of deeper architectures.

    Incorrect. Skip connections do not remove the need for activation functions in the network.

    Please answer all questions to proceed.
    Watch Video 1

    Unit 7.4

    Videos
    Follow along in a Lightning Studio

    DL Fundamentals 7: Computer Vision

    Sebastian
    Launch Studio →