Jess Sorrell
@JessSorrell
Followers
1,001
Following
676
Media
34
Statuses
241
Assistant Prof @ Johns Hopkins CS. Interested in theory of ML, responsible computing. All cat pictures are my own and do not represent the cats of my employer.
Joined March 2024
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
GOOD TIMES WITH BAMBAM
• 160201 Tweets
Córdoba
• 115869 Tweets
Cowboys
• 91925 Tweets
Ravens
• 82790 Tweets
Veracruz
• 82135 Tweets
連休最終日
• 70784 Tweets
Chiefs
• 68154 Tweets
Dallas
• 63705 Tweets
Panthers
• 61718 Tweets
49ers
• 56555 Tweets
Raiders
• 53507 Tweets
Jerry
• 44766 Tweets
Falcons
• 36355 Tweets
Rams
• 34886 Tweets
#SusanaGimenez
• 34011 Tweets
Mahomes
• 30046 Tweets
Andy Dalton
• 28321 Tweets
Jennings
• 25025 Tweets
Paredes
• 23544 Tweets
Purdy
• 21033 Tweets
Maha Paropkar Diwas
• 17256 Tweets
振替休日
• 15755 Tweets
Kelce
• 15319 Tweets
Niners
• 11794 Tweets
GARIME TOP DOS
• 11166 Tweets
Last Seen Profiles
I'm starting a new gig in July at
@JohnsHopkins
, as an Assistant Professor in the Department of Computer Science! I'm beyond chuffed to be signing on with JHU, and more grateful than I can articulate for the support I've received from friends, family, colleagues, and mentors.
71
22
663
I'm still dissecting my strongly negative knee-jerk reaction to this announcement, but there's something I clearly don't understand: When high school students do research, which is a thing I just learned in the comments apparently happens, who is advising them?
NeurIPS 2024 will have a track for papers from high schoolers.
78
90
596
33
5
141
Pro tip: be sure to remove all naturally occurring instances of the word "delve" from your NeurIPS submission to avoid desk rejection
5
1
86
The UCSD theory group is really something special. I was (and continue to be) incredibly lucky to learn from everyone there.
Congrats to
@MHop_Theory
and Rex Lei on defending!!
The UCSD theory group EOY celebration. We had a lot to celebrate: alum
@JessSorrell
joining JHU as an assistant prof,
@MHop_Theory
and Rex Lei graduating, lot of amazing work including Chris’s work selected as ICML Oral, a big cohort of students and postdocs joining in 24,
3
1
56
0
0
42
Buying hundreds of ping pong balls to turn my bathtub into a ball pit for the cats. We all celebrate NeurIPS deadline season in our own way.
3
0
34
Had an awesome time at the University of Copenhagen (thanks
@AmartyaSanyal
,
@RasmusPagh1
, and others)! Now off to Aarhus for the Aarhus Summer School on Learning Theory :)
3
1
30
Extremely excited to be talking tomorrow at the Denmark Learning Theory and Applications (DeLTA) seminar at the University of Copenhagen! Also green with envy over the parks and bicycle infrastructure I've seen here in just 2 hours, my god.
0
2
27
@thegautamkamath
I was walking through a campus back to my hotel rocking this look, and someone walking by, without additional context, said, "Good luck!" nodded encouragingly, and continued walking
0
0
23
This line up of talks looks awesome and I have it on good authority that the organizers are delightful people. I'm registered!
The 7th Eastern Great Lakes Theory of Computation Workshop is October 5-6 in Buffalo, NY! Ft talks including
@nsrg_shah
@praveshkkothari
@sushnt
Madhu Sudan, Jon Kleinberg, and the latest Turing winner, Avi Wigderson!
Free registration, website link in next tweet 👇
3
6
49
1
2
18
So good to be back in D.C. for day one of the Privacy and Public Policy Conference!
1
0
19
Academic Twitter, I return to you, with hat in hand, asking you to take me back.
3
1
17
Thanks, Jim Simons, for the profound material support of our research community, and my academic ambitions in particular. I look forward to reading your biography.
Truly sad to hear Jim Simons passed away. I saw him in his office just 6 months ago when visiting Flat Iron Institute, which he financed. The book about him "The Man Who Solved the Market" is my favorite biography and I highly recommend it, especially if you are a mathematician.
4
7
81
0
0
14
Trent Thomas for president
Everybody should be allowed one (1) Math Year™️ where you have no responsibilities and can spend all your time (re)learning any math you want at your own pace
15
222
2K
0
0
13
Honestly didn't realize that was an option, but I'd like to think so
0
0
12
@gabecubed
My advisor used to say, "Your reward for solving a challenging mathematical problem is the ability to be confused by a new mathematical problem." Whether that was encouraging or disheartening seemed to depend a lot on how frustrated with a recalcitrant problem the recipient was
1
0
9
"Impagliazzo has been a lifelong aficionado of tabletop role-playing games like Dungeons and Dragons "
Can confirm, Russell is both an amazing DM and advisor, and I refuse to believe these are unrelated skills
0
0
10
@guerzhoy
to provide sufficient resources (material support for junior researcher attendance and the scarcest resources of all: attention from senior researchers) to ensure the community has the chance to appreciate their work.
1
0
10
I will not make an April Fools' version of my job talk. I will not make an April Fools' version of my job talk. I will not make an April Fools' version of my job talk. I will not make an April Fools' version of my job talk. I will not make an April Fools' version of my job talk.
3
0
9
@ccanonne_
I didn't have the pleasure of meeting him, but his work on the connections between boosting, graph regularity, and dense models totally blew my mind and played a pretty big part in shifting my research trajectory
0
0
9
@letonyo
@ghostway_chess
It would also break many (all?) FHE schemes built from lattices, which generally need smaller noise to modulus ratios
2
0
9
Drop your advice for new professors in the replies, please and thank you!
8
0
8
This is my eclipse pic. There are many like it, but this one is mine.
0
0
7
Thoughts from
#NeurIPS
ethics reviews (I've done ethics review for only two years, so I have a small sample here):
- as an author and reviewer in ML venues, I now plan to start from the assumption that my work or the work of others DOES raise ethical considerations..
3
1
7
@krismicinski
Got Klein bottle earrings from him at... I wanna say the JMM in San Diego? Have you read Cuckoo's Egg by any chance? Totally wild story.
2
1
7
I kinda perversely want this to not only be correct, but also tight, where like GapSVP really just becomes hard at gamma = 4.5 for mysterious reasons (ideally involving the leech lattice)
2
0
5
Ok, everybody, it's been a long day in the scribble mines. Pack it up and get some rest, I wanna see you back here bright and early tomorrow for more of whatever this is
2
0
6
@henryquantum
@JohnsHopkins
Thanks! It's a picture of the George Peabody Library, which is affiliated with JHU via the Peabody Institute.
1
0
6
@thegautamkamath
I know multiple people on the market this year, myself included, who have motivated themselves with the mantra, "If T Swift can do Eras, I can do this," so I guess she's just an icon for the academic interviewee?
0
0
4
@TMoldwin
Creating things is different from doing novel research though, which generally requires a deeper understanding of what's been done before and, among the things that haven't, what technical obstacles are in the way. Seems unreasonable to expect anyone, high school student or not
2
0
5
@marcel_hussing
I'm rusty, but can take a crack!
The Learning w/ Errors cryptosystem is one of the top contenders for quantum-secure public key cryptography. We won't go into the specifics, but there are two properties of the cryptosystem that are relevant here:
1
2
5
@jasondeanlee
@aminkarbasi
Not generally. E.g., if the task if estimating some real-valued parameter of a distribution, different samples might yield different real values within a target tolerance, and so satisfy correctness, but not replicability
2
0
5
@guerzhoy
2. I was involved in organizing an affinity workshop last NeurIPS, and came away with the impression that the community is not suffering from a lack of junior researchers (in this case undergrads and MS) who've done work worth highlighting, but that we're already failing...
1
0
5
This morning we have lectures on sample compression from Amir Yehudayoff, with bonus abstract art!
1
0
5
What are your office AV setups for research meetings that involve whiteboarding, with some participants remote and some in office? I asked Reddit and all it told me was to not get an owl for a large conference room, which was fun to think about, but not relevant for my purposes.
1
0
4
Yesterday morning,
@shaibendavid5
kicked off the summer school motivating the study of learning theory with a cautionary tale about pigeon superstition (overfitting)
2
0
5
Luca's talk (presented by Pravesh and Salil) is starting soon!
TCS4All Inspiration talk written by Luca Trevisan starts at 12:30 pm Pacific.
Youtube live link:
1
7
25
0
0
5
@guerzhoy
is to have a lower bar for acceptance for such papers, then I think it's a strange choice to have a separate track rather than, e.g., a workshop, especially given that the conference is already struggling with noise in paper submissions and reviews, with no promising solutions
2
0
4
I fell off my posting schedule, but we're back with
@aminkarbasi
telling us about mind reading, one of the many applications of submodular optimization!
1
0
4
@shaibendavid5
Later on, Nikita Zhivotovskiy told us about the many faces of Exponential Weights, a true gem of an algorithm
1
0
4
Closing out day 2, we have Tamalika Mukherjee talking about the hidden costs of privacy choice for marginalized groups! Fun fact: Tamalika is a postdoc at Columbia IEOR on the academic job market this year :)
0
1
5
@aminkarbasi
Now we have
@shaibendavid5
ad-libbing a lecture on set theory, this is such a treat!
1
0
4
worth noting in the work, and then ask myself if there's enough evidence against the assumption that a section on ethical considerations is unnecessary
- if a paper uses/introduces a new dataset, it should undergo ethics review. Period...
1
0
3
0
0
4
@jasondeanlee
@aminkarbasi
And in the agnostic setting, where there could be many equally good models with very different behavior, replicability would guarantee that whp over internal randomness and sample, once you fix the internal randomness, you get the same model even as you vary the sample
1
0
4
@marcel_hussing
it's not amenable to attack by this algorithm.
I *think* it's the case the most public key variants of LWE use parameters that would be safe. The variants that show up in fully-homomorphic encryption, which need smaller error rates to work at all, might be broken by this though
1
1
3
@glen1macdonald
@LongFormMath
Oh I understood what the statement was saying. I just find it cool and surprising and will have to read the proof!
1
0
3
@guerzhoy
students from the hosting country would benefit, but virtual attendance does make me wonder what we expect students will get out of this other than a line on their CV. A significant benefit of participating in conferences as a junior researcher is the opportunity to get feedback
2
0
3
@marcel_hussing
Nice! Not DC, but I went over to Fairmount Park to get a peek at our own cherry blossoms this afternoon
0
0
3
1
0
2
@liminal_warmth
Oh man, I really dig Ze Frank, both True Facts and his vlogs about creative work
1
0
3
@guerzhoy
I'm not sure I object yet (my knee-jerk reaction is negative, but I think that's based on my cultural preferences and NeurIPS is and should be multicultural), but I can state a few things I don't like about the new track:
1. Why create a separate research track? If the intention
2
0
3
@TMoldwin
to be able to jump in, identify an interesting unsolved problem, and solve it without the mentorship of someone already doing research. Not impossible, but given that lots of HSers seem to be doing this, I assumed there was a mentorship system and was curious what it was
1
0
3
@SiaraRouzer
I bombed out of my math/physics undergrad program due to being a little punk kid. I worked for coffee shops and the postal service for six years before going back to finish my math degree. I failed math methods in physics and multiwavelength astro imaging, among others 😅
0
0
2
@guerzhoy
3. It looks like only 5 students will have the opportunity to attend in person. On the one hand this is good, because I don't think bringing more minors to NeurIPS is great from a legal perspective, and because assuming in-person attendance would pretty much guarantee only ...
1
0
3
@Indian_Bronson
This is not a ceiling fan, this is a Final Fantasy 7 random encounter in Midgar
0
0
3
@DimitrisPapail
@jasondeanlee
@aminkarbasi
The def of replicability can sort of capture both cases, it's just a matter of what you consider the output of the algorithm to be. Typically we call the model the output of the algorithm, and then replicability requires syntactic equivalence of models...
1
0
2
@ccanonne_
Oh sorry, that was just me hoping for a silly world where the leech lattice is integral to proving tightness of this result. I've found the leech lattice mysterious ever since a friend tried to explain monstrous moonshine to me
0
0
2
@gabecubed
Not slick and animated, but I would also like to contribute pictures of fractal dragons and little owl buddies
1
0
2
0
0
2
@KyleMorgenstein
Oh man, it's really good! It's hard to say what it's about without spoilers, but you are an alien setting out on your first journey into space and you slowly start to piece together the lore of another space-faring civilization that came before you
1
0
2
Heard from
@marcel_hussing
that all the cool kids are submitting to this workshop 👀
If you are wrapping up your
#NeurIPS2024
submission on
#RL
theory and/or Control theory, consider submitting to our ICML workshop
Deadline on May 27th :)
Please repost 🙏
1
24
60
0
0
2
@jasondeanlee
@MHop_Theory
@aminkarbasi
One example is the parities problem that Amin and others consider. Say my distribution puts most of its weight on a subspace V, such that this is the only subspace you need to be consistent with to get good accuracy, but there are a few orthogonal subspaces that still...
1
0
2
@michael_nielsen
If you ever wanted to ring tower bells, Trinity Change Ringers is open to visits/instruction for newcomers (I think they need to be arranged via their site though)
1
0
2
"Let's all agree on how we're going to tell time on the moon," is a solid closing quote.
0
0
2
@JHUCompSci
is hiring! I joined in July and it's already proved to be a great place to do research. Join us!
🚨 Johns Hopkins
@JHUCompSci
is hiring faculty at all ranks! 1) Data Science and AI 🤖; 2) All other areas of CS 💻. We will double in size to become one of the largest departments in the coming years! 🚀
Note: early action and spousal placement. 👥 🧵
2
48
170
0
0
2
@EugeneVinitsky
Thanks! I guess I have a whole system of high school research to learn and develop opinions about.
1
0
2
@DimitrisPapail
@jasondeanlee
@aminkarbasi
But for countably infinite domains, you could have the alg output a vector of the model's predictions on every element of the domain, and then you'd get the functional equivalence you mention
1
0
2
@TaliaRinger
It looks like you have 225lbs there, please don't call my 1rm light weight, you are hurting my feelings 😂😭
2
0
2
@sleeping4cat
Please see my point about flagging for ethics review not being punitive! Ethics review isn't holding back innovation, it's just asking that another pair of eyes take a look at new work with a focus on potential harms and legality rather than technical merit
2
0
2
@guerzhoy
from the community, and this seems much harder virtually.
These are some of the things that seem off to me about the new track, but still not convinced it's actively harmful enough to take up arms about it just yet.
0
0
2
@KyleMorgenstein
I'd recommend playing it for ~1 hour to see if it's your thing. The overall vibe is that life is, cosmically speaking, small and fragile, and the universe is a beautiful and fascinating place. Also the music is great imo
1
0
2
@thanhnguyentang
Here's the dropbox link. Not all of the slides are up yet, but should be soon.
1
1
2
The morning started with a panel on data sharing agreements, PETs, and obstacles to enabling useful privacy-preserving data analysis
1
0
2