search

Bin-Cao / TrAdaboost

Boosting for transfer learning with single / multiple source(s) Regression / Classification
MIT License
34 stars 8 forks source link

What is the pre-training model that the program is based on #5

Open HangboZhu opened 10 months ago

HangboZhu commented 10 months ago

Dear Cao Bin: Glad to see your TrAdaboost project on github, I'm very interested in it! However, due to my shallow knowledge. I have a little bit of trouble understanding on which pre-trained model the regression model you constructed is based on for transfer learning.Can you use TrAdaBoost_R2 as an example and give me an idea of which pre-trained model he is finetuning on top of? I hope you take time out of your busy schedule to give me an answer, I would appreciate it! Best regards Hangbo Zhu

Bin-Cao commented 10 months ago

Thank you for referencing my open-source project. I would like to clarify that TrAdaboost is an instance transfer strategy proposed by Prof. Yang many years ago, which does not follow the pre-training and fine-tune strategy. The main focus of TrAdaboost is to identify the appropriate weights for the source domain training data in order to reduce the distribution distance between the source and target domains. If you are interested in the pre-training and fine-tune strategy, I recommend looking into parameter-based transfer learning.

HangboZhu commented 10 months ago

Thank you for your reply and good luck with your research!