- Instructor: Hoyeol Kim
- Email: hoyeolkim7@gmail.com
- Mastodon: @hoyeolkim@hcommons.social
- Twitter: @elibooklover
- Course Github Repository: https://github.com/elibooklover/Deep-Learning-for-Humanists-2023
This hands-on course will introduce exploratory data analysis (EDA), image preprocessing, and deep learning models for those who wish to explore deep learning for the digital humanities. This course will first introduce computer vision in deep learning, then exploratory data analysis using Google Colab. After that, participants will train GANs (Generative Adversarial Networks) and cGANs (conditional Generative Adversarial Networks) models to colorize black-and-white images. Throughout this course, participants will be able to create their own datasets for deep learning and run deep learning models with them.
Google Colab: a cloud-based notebook environment which is suitable for machine learning and data analysis. A Google account is required to use Colab. Participants are not required to purchase a Colab membership for this course since we will only be using the free features.
GitHub: Participants need to sign up for GitHub in order to access course materials. Participants do not have to get a Pro membership.
By the end of the workshop, participants will:
- Understand the fundamental concepts of deep learning.
- Be familiar with generative models and their deep learning algorithms.
- Be able to preprocess data for deep learning.
- Deploy deep learning models for computer vision tasks.
- Monday, June 12th (Day 1) - 2-5pm (CST)
- Introduction to machine learning and deep learning
- Generative models / Generative methods
- GAN
- GAN-based models
- Neural networks
- Activation functions
- Tuesday, June 13th (Day 2) - 2-5pm (CST)
- Tensor
- Normalization
- Loss functions
- Gradient descent
- Object detection (RetinaNet)
- Data augmentation
- Image processing
- EDA
- Wednesday, June 14th (Day 3) - 3-5pm (CST)
- L1 and L2 loss functions
- Experiments with GANs
- pix2pix
- Victorian400
- Transfer learning
- Multimodal deep learning
- Wrap-up