Update: recordings are being posted at: https://www.youtube.com/playlist?list=PLS6ZvEWHZ3OP6-Z5qnzGKNKWA-l2Sry3l
Machine learning is changing the way microscopy operates on all levels – from analysis of imaging and hyperspectral data to microscope optimization to the way instruments scan, acquire spectra, and even design and execute experiments. The purpose of this school is to provide an introduction and hands-on skills that constitute the individual elements of this transition and take it to the next level as a community.
Lectures and hands-on Colab practice sessions will be scheduled on Tuesdays and Fridays 9 am EST via Zoom. The school will be free of charge. For registration, send e-mail to sergei2@utk.edu
-
This school will be primarily aimed at the practitioners of STEM, STEM-EELS, and 4D STEM. However, the methods, workflows, and notebooks discussed during the school can be applied to all other probe-based imaging and spectroscopy modes practically without change, including CITS in STM, F-d curves in AFM, IV in cAFM, and any other hyperspectral imaging modes (including micro-Raman, ToF-SIMS, etc).
-
The focus of the school will be the ML analysis and AE in microscopy. The lectures and hands-on exercise will start assuming that you are generally familiar with the general principles of STEM-EELS and 4D STEM, and understand physical meaning of the data.
-
Similarly, the school will emphasize the applications of ML to the microscopy data but will not go into the details of the deep network engineering. By now, there are multiple excellent references on it, as well as ChatGPT.
-
Many of you asked about the necessary level of the Python and ML knowledge. The purpose of the course is to give the basic hands-on skills in analyzing the imaging and hyperspectral data and configuring automated experiment using the pre-acquired data sets. The workflows will be implemented in the Google Colab Notebooks that will allow to reproduce the analysis in full. However, basic capability to read and modify Python code will be highly beneficial. Here: a. My favorite intro to Python is Packt (note that books are expensive – but subscriptions are fairly cheap): https://www.packtpub.com/product/the-python-workshop/9781839218859 b. If interested in analyzing own data sets, you should be able to convert your data from instrument format into the Numpy array.
-
Some housekeeping details: a. The course is open for all participants (students, industry, faculty, national laboratories) b. There is no fee c. The course will be given virtually twice a week, tentatively as 2-hour slots (lecture + hands-on demonstration) d. To the estimated length of 6-8 weeks (to be tuned) e. The course is expected to start in mid-June f. For registration, we will need your e-mail (as a primary way to contact, send Zoom invites, etc)
-
The tentative program (to be updated) is in this repository
-
As a distinctive feature of the course, you are welcome to send the questions “how can we use ML to [describe problem]”. While we cannot guarantee that we can answer them, we will aim to incorporate these scenarios into the presentations and/or discuss as imaginary scenarios and workflows throughout the course.