|
Instructor: |
Dr. David Doermann |
|
Office: |
113M Davis Hall |
|
Email: |
Prefers to be contact through Piazza |
|
Office Hours: |
|
Generative Adversarial Networks (GANs) are machine-learning frameworks where two neural networks compete; a generative network attempting to generate fake data that can fool a discriminative network trained a real data distribution. In this seminar, we will focus on applications of GANS, starting with their fundamental properties, and ultimately exploring applications from photorealistic imagery of faces, products and even 3D scene to applications for data augmentation and training of other neural networks.
You will learn best practices for reading and presenting the appropriate literature and are expected to program GANS and apply them to one of several interesting applications.
For this seminar, you MUST
- have taken machine learning and, if possible, advanced machine learning and deep learning, and
- be an excellent programmer with experience in using common machine learning frameworks and GPUs.
- If you do not meet these criteria, you must get written permission to take this course from the instructor during the first week of classes
Attendance: Required (80% of classes for Pass)
Enrollment:
- 1 Credit:
- Class Participation, Reading of all papers, Presenting 1-2 topics
- 3 Credits:
- Class Participation, Reading of all papers, Presenting 1-2 topics
- Written Survey Paper or GAN Related Project and Presentation
- If you would prefer to do a survey paper, you must obtain written permission to do so during the first week of class.
It is entirely your responsibility to follow the policies. Please ask the instructor if you have questions.
Your grade will be Pass/Fail based on a combination of
- Class Participation
- Reading of all papers,
- Presenting 1-2 topics
- Written Survey Paper or GAN Related Project and Presentation (3 credits Only)
If you have any disability which requires reasonable accommodations to enable you to participate in this course, please contact the Office of Accessibility Resources, 25 Capen Hall, 645-2608, and also the instructor of this course. The office will provide you with information and review appropriate arrangements for reasonable accommodations. This must be done in the first 2 weeks of class. http://www.student-affairs.buffalo.edu/ods/
All work must be your own
- Do not take the answers, words, ideas or research findings of other people as yours; cite and acknowledge properly, and develop your own ideas.
- No cheating
- According to departmental policy, any violation of academic integrity will result in a Failing Grade for the course, and termination of departmental financial scholarship.
- Tools will be used to check similarity. Similar submissions will result in Failing Grade for all involved parties.
- Use of a code from an online repository, e.g. Github, must include a proper and clearly visible attribution in your report.
- Deep Convolutional GANs (dcGANs) - https://arxiv.org/abs/1511.06434
- Conditional GANs (cGANs) - https://arxiv.org/abs/1411.1784
- StackGAN - https://arxiv.org/abs/1612.03242
- Discover Cross-Domain Relations with Generative Adversarial Networks (Disco GANS)- https://arxiv.org/abs/1703.05192
- Video Frame Prediction - https://arxiv.org/abs/1511.06380
- Super Resolution - https://arxiv.org/abs/1609.04802
- High Resolution Synthesis (semantic image -Photo) - https://arxiv.org/abs/1711.11585
- Image to Image Translation
- Pix2Pix - https://arxiv.org/abs/1611.07004
- CycleGANs https://arxiv.org/abs/1703.10593
- STARGAN - https://arxiv.org/abs/1711.09020
- ATTNGAN - https://arxiv.org/abs/1711.10485
- SSGAN (stenography) - https://arxiv.org/abs/1707.01613
- StyleBased GANS - https://arxiv.org/abs/1812.04948
- Text to Image Synthesis - https://arxiv.org/abs/1605.05396
- MirrorGAN: Learning Text-to-image Generation by Redescription - https://arxiv.org/abs/1903.05854
- Art
- Manipulation of Faces, Editing
- Cartography
- Medical Image Data Augmentation
- 3D Object Generation
- Attention Prediction
- Cartoons
- Face Front View Generation
- Photos to Emojis
- GAN Development - Keras and Programming GANs
- GAN Development - PYTorch and Programming GANs
- GANs for Data Augmentation - Synthetic data for training