The generalizability thing you’re talking about is related to the bias-variance trade-off. A good solution is Tikhonov regularization. (You may already know this, just wanted to add something).
I had to Google that one. I knew it by its other name, ridge regression. Yes, this indeed is one way to improve model generalizability (increase bias, reduce variance)
Thanks sir for useful information. Your way of teaching scientific complexities is enduring. I want to use ML in HEA for mechanical strengthened Alloys. Can you help me in this regard? I am a PhD student in Material Science and Engineering.
If you have a lot of data, or examples of where people have tried to experimentally or with simulations learn things about high entropy alloys, could you then find patterns in that data to then build predictive models to predict things like which phases will be present or what the properties will be.
@@032_devasyamishra4 Right now there are many, many machine learning studies on HEAs and the vast majority are pretty bad. They overfit to very specific data sets, don't follow machine learning best practices, and don't generalize well to answer any real problems that the field faces.
@@andrewb.7890 Thanks for clarifying, except youtube doesn't only show your videos to people with basic material science background. I won't expect someone completely new to this field to have prior Mat Sci knowledge but they too get recommended interesting videos like this.
I would suggest you to make videos in many parts . You have too much informations and data in one single video 😄. (This suggestion is for beginners who like to learn from you more)
thanks for the series
What else should I cover in this series?
Some cost to strength ratio analysis of some sort 😅
However, don’t know any suggestions on refining existing ML models…
The generalizability thing you’re talking about is related to the bias-variance trade-off. A good solution is Tikhonov regularization. (You may already know this, just wanted to add something).
I had to Google that one. I knew it by its other name, ridge regression. Yes, this indeed is one way to improve model generalizability (increase bias, reduce variance)
You should organize an HEA ‘moonshot’, get all these heads together. 👍
Thanks sir for useful information. Your way of teaching scientific complexities is enduring. I want to use ML in HEA for mechanical strengthened Alloys. Can you help me in this regard? I am a PhD student in Material Science and Engineering.
What are you trying to do?
@@TaylorSparks i want to incorporate B2 nanopercipitates in M/HEA for strength-ductility improvement
Hey taylor. I have been working on research on the same topic. Would you like to get connected so we can discuss about this sometime?
Let's chat! Shoot me an email
Can you please just explain me in gist about what is the use of ML in high entropy alloys
If you have a lot of data, or examples of where people have tried to experimentally or with simulations learn things about high entropy alloys, could you then find patterns in that data to then build predictive models to predict things like which phases will be present or what the properties will be.
@@TaylorSparks so exactly what is the conclusion?
@@032_devasyamishra4 Right now there are many, many machine learning studies on HEAs and the vast majority are pretty bad. They overfit to very specific data sets, don't follow machine learning best practices, and don't generalize well to answer any real problems that the field faces.
@@TaylorSparks Hey Taylor, Newbee here, where to get these datas to train the models?
@@arvindkishore7481 literature mostly.
It take less than 30s to define abbreviations that you'll use through out the whole video. What's FCC and what's BCC?
Face-centered cubic and body-centered cubic. These are very common crystal systems in materials, any Mat Sci 101 should be able to provide more info.
@@andrewb.7890 Thanks for clarifying, except youtube doesn't only show your videos to people with basic material science background. I won't expect someone completely new to this field to have prior Mat Sci knowledge but they too get recommended interesting videos like this.
I would suggest you to make videos in many parts . You have too much informations and data in one single video 😄. (This suggestion is for beginners who like to learn from you more)
Ya, my intro to MSE videos are broken up into small chunks.