Amanda Gentzel

@akanksha_atrey Congratulations!!

🎓 Life update: I successfully defended my Ph.D. thesis today! 🎓

People who subscribe to Merriam-Webster Red™ will get exclusive access to the real definitions.

Okay, I read it so you don't have to. Here's a reaction thread to @openAI / @sama 's blog post from Friday "Planning for AGI and beyond": openai.com/blog/planning-…

Managed to crash RStudio by trying to calculate a contingency table. Truly, my R skill is to be feared.

@ToschEmma @norootcause Yup! I especially use that term for pecan tassies (which are functionally mini pecan pies)

Americans: what do you call these?

The Bronze Age Collapse A pivotal and mysterious moment in history, when much of the known world order imploded.

@annehelen Local board game groups. :)

What's an organization that isn't religious or through your kid's school that has allowed you to build connections, however loose, with people who aren't your own age?

@ToschEmma I'm still really bad at reaching out to people to talk about research, but I think I'm getting better. I feel like I'm learning and improving, while making enough money to live comfortably and enjoy my free time. Things definitely aren't perfect, but I'd generally say I'm happy.

@ToschEmma I would say I'm neutral to happy right now, chronic pain not withstanding. Dissertating/moving/starting a job during a pandemic still sucks, and it's hard not to just panic about the state of the world, but for me personally, things feel like they're going relatively well.

@ToschEmma Wrt to working remotely during a pandemic, I always have been a bundle of anxiety and imposter syndrome, and I don't see that changing. But I like my coworkers, and I talked with my manager about wanting to branch out to other types of projects, and it was well-received.

@ToschEmma It helps a lot that while I moved, I moved close to family. So I have some built in friends with my twin sister's friends (who I can game with weekly when the pain lets me) and can hang out with my parents. Plus I still have regular remote D&D with MA friends.

Just lost a few hours of yesterday's progress because my jupyter notebook on a remote server apparently doesn't auto-save and I got disconnected. In theory, it should be faster since I've already written it once. In practice, I expect to just be hate-coding the whole time...

I just got the security question option of "What is your favorite security question?" Unfortunate, it wasn't my favorite security question, so I didn't pick it.

Sometimes, after working in Python for months, I start to think that Python might not be that much worse than R for data exploration. And then I use R, and between count(), str(), ncol/nrow/dim, and ggplot, I find myself apologizing to R for ever doubting it.

@ToschEmma But Emma... There are numbers in the csv! Why would you want to use anything else?

@akanksha_atrey @umasscs This is such a lovely idea. Coming from undergrad, it can be very hard to know what PhD programs are looking for, so providing this sort of mentorship is great way to help!

The PhD Applicant Support Program at @umasscs is back! Important dates: 💻 Oct 28 — Info Webinar for all PhD applicants 👥 Oct 14 onwards — 1-1 support for underrepresented applicants in CS For info: paspumasscs.github.io For Q/A: pasp.umasscs@gmail.com #phdchat

@ToschEmma Mad teaching skillz 😎

@Sergei_Imaging @zacharylipton @yudapearl @tdietterich Agreed. When I use/hear the term 'distribution shift', my thinking around it is causal: I envision a situation where the distribution of a variable may change, but the underlying causal dynamics stay the same. The issue is not the term, but the understanding around it.

@zacharylipton @yudapearl @tdietterich Plus, out of distribution shift perfectly applies to many experimental problems. My microscope setting changed - affecting one of the SCM equations - but leaving the DAG and other SCM equations as they are.

The idea that "knowledge" is more than "distribution" has not caught on yet in DL circles and, consequently, we continue to see statistical vocabulary dominate the writings of DL authors. Expressions such as "distribution shift" and "out of distribution" will soon be replaced.

You can check out our paper and 5 minute talk here! icml.cc/virtual/2021/p…

Tonight at #ICML2021, I will be presenting my work with @purva_pruthi and David Jensen on "How and Why to Use Experimental Data to Evaluate Methods for Causal Inference". Our poster will be in spot A4 in Gather Town Room 7, tonight from 12am - 2am EDT.