Unit 2 - CNN | Transfer Learning/Fine-Tuning Draft Outline

[alperenunlu]

Hey everyone.

This is the proposed sections for the Transfer Learning/Fine-Tuning Chapter.

Transfer Learning/Fine-Tuning

• What is Transfer Learning?
• What is Fine-Tuning?
• Fine-Tuning with Torchvision [VGG16, ResNet50]
• Torchvision vs Timm
• Fine-Tuning with Timm [VGG16, ResNet50]
• Detailed Fine-Tuning tutorial on an image classification task. [Adapted on Different Models]

---

Open to suggestions for further improvement of the draft.

[ATaylorAerospace]

@alperenunlu wow thanks for stepping in to add the content for this chapter!

I had a few ideals for additional content to flesh the chapter out a bit..what do you think of the below items?

• Understanding Pre-Trained Models (Cover this before diving into Transfer Learning)

• The Role of Data in Transfer Learning (Explain how the quantity / quality of data affect Transfer Learning outcomes)

• Transfer Learning in Different Domains (Like object detection or segmentation domains)

• Fine-Tuning Approaches: Feature Extraction vs. Full Model Fine-Tuning (Cover these approaches prior to diving into the Torchvision and Timm Pytorch libs)

[sitamgithub-MSIT]

@alperenunlu wow thanks for stepping in to add the content for this chapter!

I had a few ideals for additional content to flesh the chapter out a bit..what do you think of the below items?

• Understanding Pre-Trained Models (Cover this before diving into Transfer Learning)
• The Role of Data in Transfer Learning (Explain how the quantity / quality of data affect Transfer Learning outcomes)
• Transfer Learning in Different Domains (Like object detection or segmentation domains)
• Fine-Tuning Approaches: Feature Extraction vs. Full Model Fine-Tuning (Cover these approaches prior to diving into the Torchvision and Timm Pytorch libs)

Those are good suggestions 👍🏻

[sitamgithub-MSIT]

@johko @merveenoyan any suggestions or changes on the above draft!?

[johko]

hey, sorry for the late reply.

The outlien looks good to me :+1: I especially like that you bring TIMM into it and compare it with Torchvision

I also like the additions from @ATaylorAerospace , but maybe handle the points

- The Role of Data in Transfer Learning (Explain how the quantity / quality of data affect Transfer Learning outcomes)
- Transfer Learning in Different Domains (Like object detection or segmentation domains)

as a bit lower priority for now. If you have time to cover it, cool. Else you can just shortly mention it without going into too much detail for now.

Any ideas on specific file structure, so what goes in an .mdx, what goes into notebooks?

[alperenunlu]

@ATaylorAerospace

• Understanding Pre-Trained Models

This is the topic of previous chapter you can check the TOC.

• Transfer Learning in Different Domains (Like object detection or segmentation domains)

This is a great section but also better suited Unit 6. They already have a section on YOLO RCNN etc. section on their draft. You can check their draft in here.

[ATaylorAerospace]

@alperenunlu Ok sounds good. I agree with you on transfer learning in Different Domains being better suited for Unit 6.

[alperenunlu]

Hey @johko thank you for your comment.

I especially like that you bring TIMM into it and compare it with Torchvision

Yeah Timm is somewhat hard for beginners but when you get used to it Timm becomes an incredible tool.

Any ideas on specific file structure, so what goes in an .mdx, what goes into notebooks?

I am not fully familiar with new file structure but the sections on TorchVision and Timm are gonna be notebooks. Other sections can be mdx files.