I always see transfer learning applied to image recognition, but I would like an example of it being applied to predicting "numbers". Does anyone have an example?
It’s not really a technical term, I used it to give an intuitive understanding. What I meant with it is that when you’re working with a pre-trained model, it is already good at a certain task. With transfer learning you add an output layer that is randomly initialized so it is not yet good at producing good results. By running the pre-trained model and the freshly created output later together you let the output layer train to a level that performs to an acceptable level. During that time, the pre-trained model’s weight are not updated. And once the model performs well enough (this would depend on your model’s goal performance) it means the new output layer has caught up. Hope this clears it up! - Mısra
@AssemblyAI Good tutorial, I'll like to use this for my problem dataset to improve accuracy/detection. My challenge is I have about 60 Png images that are not particularly distinguishable, wondering if this method will work well for me. Any suggestions?
Hey Andrew, I don't have experience with keypoint detection. Have you searched the pre-trained model libraries to see if they have a useful pre-trained model? -Mısra
@@AssemblyAI of course. There are none that are easily adaptable. I'm using one from github, but can't get the environment set up so that it works. Python sucks even more than node.js
Very nicely explain the Transfer Learning. I want more videos in this line.....so keeping posting latest TL video
It'd be sick if you put links in the description of your shorts too, but glad to find them here. Mad nice, btw, subbed.
Awesome tutorial!
Thanks Python Engineer!
Your videos are so informative and knowledgeable . Can you make a video on hugging face models with API connections
nice information 🌷
Thanks!
I think this is the best tutorial on UA-cam! Just one question, can I load a custom dataset to use or it has to be found in the tfds? Thanks!
Thank you! You can use a custom dataset too of course. Just need to make sure it is the correct format . :)
we would like more transfer learning and fine tuning vids thanks
Love it
Great!
Thanks :) it's great
Thank you
Need some Explanation about Optimizers also.
She explains beautifully, what is her name?
Sandy
Does this also apply to text and numerical data? Can transfer learning be done on text and numerical data?
I always see transfer learning applied to image recognition, but I would like an example of it being applied to predicting "numbers". Does anyone have an example?
@AssemblyAI yeah, I too have been looking for the application of TL in a regression use case. Do you have any tutorials for this?
@@TerrenceMcGinnis-m6n mee too did you find anything please ?
@@TerrenceMcGinnis-m6n ?
Hi! At 8:23, you say "to catchup with the model". Could you explain what this means in further detail please? I've not heard that term used before!
It’s not really a technical term, I used it to give an intuitive understanding. What I meant with it is that when you’re working with a pre-trained model, it is already good at a certain task. With transfer learning you add an output layer that is randomly initialized so it is not yet good at producing good results. By running the pre-trained model and the freshly created output later together you let the output layer train to a level that performs to an acceptable level. During that time, the pre-trained model’s weight are not updated. And once the model performs well enough (this would depend on your model’s goal performance) it means the new output layer has caught up. Hope this clears it up! - Mısra
@@AssemblyAI Thank you!
Jeez, your hairs are awesome!
Thx
@AssemblyAI Good tutorial, I'll like to use this for my problem dataset to improve accuracy/detection. My challenge is I have about 60 Png images that are not particularly distinguishable, wondering if this method will work well for me. Any suggestions?
Sen nasıl bir yavrusun ya, ara ara karşıma çıkıyon bakmak zorunda kalıyorum.
Great tutorial, can explain the same but for pytorch? Thanks!. Sorry by my English
Awsome.. Can we have similar in pytorch?
We'll add that to our video planning. Thanks for the input! - Mısra
how would one do transfer learning with a keypoint detector? The few examples out there are obsolete and with conflicting dependencies.
Hey Andrew, I don't have experience with keypoint detection. Have you searched the pre-trained model libraries to see if they have a useful pre-trained model? -Mısra
@@AssemblyAI of course. There are none that are easily adaptable. I'm using one from github, but can't get the environment set up so that it works. Python sucks even more than node.js