Day 14: Final Project Day and More Fun!

Original article was published by UCSF AI4ALL on Artificial Intelligence on Medium

Day 14: Guest Speaker Dr. Keiser and Our Final Project Day!

Today marks the final (regular) day of the UCSF AI4ALL program. We’ll miss the ebb and flow of each day, starting off with a morning ice breaker and segueing into our guest speaker. However, we are very excited for the student research presentations tomorrow; we can’t wait to see what our students have discovered through their research projects!

UCSF AI4ALL students and alumnae TA’s in their UCSF swag!

We began the morning in our UCSF AI4ALL swag! Each student received a T-shirt and water bottle commemorating the program. After taking a few magazine worthy Zoom shots, we returned for group conversation. Our final morning icebreaker was a return to an old UCSF AI4ALL classic: Desert Island. This time, we popped the question: What three movies are you bringing to a desert island? We “popcorned” through all the students and rediscovered some fan favorite movies, everything from Legally Blonde to The Artist.

Next, we had our last faculty guest speaker, Dr. Michael Keiser, PhD, talk to us about his work with Alzheimers and classifying amyloid beta deposits. He introduced his research by breaking down how doctors diagnose cannot diagnose the disease until after death, when they examine the build up of proteins in the brain. Typically, neuropathologists take lots of time looking and counting plaque burden. In the more computational aspect, he talked about his lab’s and others’ use of computer vision as a tool to better examine plaque. Through the use of computer vision, we can better identify small areas of plaque build up that humans often miss.

Dr. Keiser presenting a slide on how the saliency map shows which pixels in the image of the cord plaque were considered salient by their predictive model

However, the human eye matters; we need human annotators to interpret them. Then, he discussed how humans label different images in different ways. Scientists across the country working in different institutions do not use the same way to label images, or even stain plaque. In order to remedy this, he discussed the use of the individual vs consensus model, classifying based on a single individual’s observation or consensus by many. After breaking down a few figures and plots, Dr. Keiser revealed to us that the consensus model actually worked better. A huge thank you to Dr. Keiser for such an informative presentation! What a great end to our guest speakers this year!

AI and Proteomics group working on their presentation slides

After a quick break for lunch, students resumed their research projects in their penultimate group meeting. The AI and Proteomics research group worked on their support vector machine (SVM) algorithm implementations and developed confusion matrices for each of their results. The Gene Expression and AI group put a few final touches on their project as well. All of the groups worked on their research presentations slides and broke up sections by student.

UCSF AI4ALL students with Co-Directors Marina Sirota and Tomiko Oskotsky

We are very excited for the UCSF AI4ALL symposium tomorrow, where students will be able to share their projects and hear our keynote speaker, Dr. Marylyn Ritchie, PhD, talk about how AI has advanced genomics and medicine. With the program coming to a close, we would like to thank all of the project and lecture TA’s for their hard work and incredible commitment to our students throughout their research projects. We can’t wait to see what tomorrow holds!

Blog post by Isha Karim.