[1/3] We are excited to open-source Langroid: – an intuitive, flexible framework to easily build LLM-powered applications.
You set up Agents, equip them with optional components: LLM, vector-store, methods.
I find the following type of review of most irritating: "I find the results in this paper not surprising." If you want to be entertained, please watch a movie:-)
Prospective graduate students. Free advice. If you have the choice, DO NOT STAY at your undergraduate institution for graduate school. Diversity of ideas and opinions is important. Being a bit uncomfortable is good.
Life begins at the end of your comfort zone.
As if 2020 couldn't get any worse, my advisor Ed Clarke passed away yesterday. He won the 2007 Turing Award for this work (). Really sad and heartbroken really. 😪
COVID situation in India hits close to home. Just lost a cousin sister. She was a gentle soul with beautiful singing voice. She is singing somewhere, but I just can't hear:-(
She was only 43 years old. Too soon...
Excellent talk by Prof. Tom Reps from
@WisconsinCS
on writing research papers. I will make it a required reading/watching for all students coming into my group.
Talk:
Slides:
I hear lot of grumbling about
#NeurIPS
reviews. The root cause is increased number of submissions and publications. As a community, we simply don't have the capacity to handle the volume of submissions. If we don't disincentivize the "numbers game", nothing will improve.
Wow, this was a really nice surprise. My years at
@iitdelhi
were absolutely wonderful. Really cherish this award.
Made by day. Now back to Zoom meetings:-)
Humbled and excited to receive MURI on the general
topic of robust learning. PIs are from US (UW, UCSD, CMU, PSU) and Australia (University of Melbourne, Macquarie University, University of Newcastle).
New group at Google with focus on ML debugging is looking for qualified candidates. Reach out to babic.domagoj
@google
.com if you are interested.
The group will be responsible for ML debugging infrastructure covering performance debugging,
Question for fellow researchers: I always try to submit my papers to top-tier venues few times before dropping tiers. However, I realize that the "revise resubmit" cycle can be frustrating for students. What is your strategy on submissions?
Congrats to Vinod Ganapathy for becoming the new chair of
@iiscbangalore
CS department. Vinod was among the first batch of students when I arrived at
@WisconsinCS
Perhaps the greatest joy in academia is to see your students succeed. Nice picture with the exchange of flowers.
Announcement about the NSF-Frontiers grant on adversarial ML (AML). The
grant is with Patrick McDaniel (PSU), Dan Boneh and Percy Liang (Stanford),
@kamalikac
(UCSD),
@dawnsongtweets
(Berkeley), and David Evans (Virginia).
Congrats to the entire team.
Building better human-bot cybersecurity teams: UW-Madison led team wins prestigious US Department of Defense MURI grant! Congratulations to CS professor and principal investigator
@jhasomesh
on this award!
@uwcdis
@UWMadisonLS
@DoDCTO
@CMUEngineering
I am deeply humbled to be part of this amazing group at
@UWMadison
Our department
@WisconsinCS
and school
@uwcdis
are incredible, and I am grateful to interact with great students, collaborators, and colleagues.
Success is the byproduct of you and the ecosystem you are part of!
Really proud of my students and collaborators. Three papers accepted at
@acm_ccs
[1] Strengthening Order Preserving Encryption with Differential Privacy
[2] EIFFeL: Ensuring Integrity for Federated Learning
[3] Federated Boosted Decision Trees with Differential Privacy
The response by
@JeffDean
was purely technical. Why don't you respond to the technical arguments point-by-point? Technical arguments are a bedrock of research, and I don't see it as toxic at all.
@JeffDean
@SashaMTL
Lol, seems totally unnecessary for a Google SVP to pick on my work in response to the LLaMA paper. Jeff, please stop. This is just one example of the toxic workplace culture that led me (and many others) to leave Google.
Perhaps one of the greatest joys in academia is one of your students placing well. Really excited that after a postdoc with
@kamalikac
,
@AmritaRoyChowd8
is joining
@UMichCSE
Congrats, Amrita!
She had go through some ups and down during the interview season, but powered through
I'm stoked to share that I'll be joining the University of Michigan, Ann Arbor, as an Assistant Professor in Fall 2024!!!
If thinking about privacy delights you as much as lapping up a butterscotch ice cream on a scorching summer day in June, then we should definitely chat🍦
Our paper "Revisiting Adversarial Robustness of Classifiers With a Reject Option" () received a Best Paper Award at the Workshop on Adversarial Machine Learning and Beyond at
@RealAAAI
2022 (). The work was lead by
@jiefengchen1
As CS community gets large, there is a big problem that we just don't have time to track all the results. When I started as a grad student (no, you can't ask, when), one roughly knew all major results in your area. Not possible now.
Thoughts?
Will be curious to read this and understand how this result isn't already implied by Nagarajan and Kolter and Negrea et al. (ICML). They cite neither as far as I can tell. :(
My favorite story related to international travel. At the LA airport, I show my US passport. The TSA agent asked me: "where are you from?". I was confused 😳 Another agent in the booth yells: "Mike, are you stupid? He just showed his US passport." and she added "welcome home".
This a first. A student asked a random combinatorics question during the review session. Like a nerd I answered it. Later I was told by another student that it was a question in on the homeworks from another class. Duh!!
Three papers at
@icmlconf
Congrats to my amazing co-authors. Will post videos soon.
1) Graphical models and privacy ()
2) From Attribution methods to causality
3) Concise explanations using adv training
@DrPorman
I much older than you, but whenever I feel like this, I take a break from everything for a few days and unplug. One advice, don't compare yourselves to others. Just follow your path. Comparing to others brings on anxiety and kills creativity.
Our department
@WisconsinCS
is looking for faculty at all levels. Reach out if you need more information.
FYI, Madison was rated as
#1
city livable city. Just saying:-)
Our paper presents the first provable watermarking scheme for large language models (LLMS) with public detectability or verifiability: we use a private key for watermarking and a public key for watermark detection. Work with Jaiden, Sanjam, Mingyuan, Mohammad, and Saeed.
Congrats to Vinod Ganapathy () on being promoted to the rank of full professor at IISc.
@iiscbangalore
Vinod was among the first group of PhD students when I arrived at
@WisconsinCS
, and thus was crucial in my success. Thanks Vinod and big congrats.
Two papers accepted to ICML 2018. Congrats to all my amazing co-authors. Both on adversarial ML. The arxiv
version of the papers are up, but we will update it soon based on reviewer comments.
Arxiv versions: and
Dual of an optimization problem can shed useful insights. What is favorite example of 👆? Please reply to this thread.
Mine is deriving a price of a stock option, which you can do from the dual (you have to assume complete markets).
Congrats to
@jiefengchen1
Jiefeng was co-advised by Prof. Yingyu Liang and me. He was a joy to have in the group, and is now a research scientist at
@Google
(). Just an amazing student with an amazing record.
Congrats Jiefeng. Now you are an alum of
Really excited to co-chair
@satml_conf
2025 with
@mlsec
We are really committed to keep growing this community. Please send email to me or Konrad if you have suggestions.
The 2nd edition of
@satml_conf
is a wrap! It was an absolute honour to co-chair the conference with
@carmelatroncoso
!
We are very excited to announce the co-chairs for the 3rd edition in 2025:
@jhasomesh
and
@mlsec
Follow
@satml_conf
for updates about the conference!
Really proud about the following paper () on exploring connections between model-extraction attacks and active learning. It is going to appear in Usenix-Security 2020. Congrats to my co-authors (Varun,
@kamalikac
, Irene and Songbai).
We are pleased to publish a paper based on the findings of the workshop held on June 27 and organized by Google, Stanford, and UW-Madison.
We hope that this paper will be useful for the community and thus welcome your feedback.
We all know reviewers that are super negative. Not just critical, but reviewers that reject everything on their path.:-) How should the research community deal with such reviewers? Thoughts?
Prof. Tom Reps attended his last faculty meeting at
@WisconsinCS
and there was a small reception after the meeting (photos attached). When I joined the department, I interacted a lot with Tom and late Susan Horwitz. I learned so much from both of them, and these interactions
There is a lot of discussion about tenure going on. My advice to young profs is not to think about it. Just do interesting things and let "chips fall" where they may. Remember in CS related fields not getting tenure is not disastrous, and..
TUE SEP 5 | 10:30 AM EST | LWSN 3102AB
As part of the excellence at
#PurdueCS
, join us with
@jhasomesh
as he delivers "Trustworthy Machine Learning and the Security Mindset" first in the Samuel D. Conte Distinguished Lecture Series.
#PurdueComputes
➡️
ML folks. There are papers that in the federated learning (FL) context that gradient updates (even with noise added) reveal information. What is the state of the art (SOTA) of the privacy leakage of gradient updates?
Great showing. The Wisconsin-PL group (I think) got 5+ papers into PLDI. Congrats to all. If you are a prospective PL student, this is the place to be!
When I joined
@WisconsinCS
, my greatest worry was "Can I turn students into researchers?"
The first 3 students I hired 4 years ago, Samuel Drews (coadvised w/
@awsTO
),
@HuQinheping
, and Kausik Subramanian (coadvised w/ Aditya Akella), just got a PLDI paper each! I'm so proud! 1/
🤣 I remember getting a recommendation letter from a very famous academic as a grad student. He asked me to write some material for the letter. His advice to start me off: "Don't be too humble".
CCS 2021 in Korea. I have visited Korea once, and it was a great visit. Really excited!! Hopefully, can travel by then.
Yongdae and Jong put together a great proposal.
* ACM CCS 2021 in Korea *
Thrilled to share that ACM CCS 2021 will be held in Seoul, Korea. Myself and Prof. Jong Kim (Postech) will serve as the General Co-chairs. Hopefully, we can take care of Covid-19 by then so that you can enjoy nice weather in Seoul, Korea.
Two papers in
@acm_ccs
Super excited about these results! Both these papers are with amazing collaborators. Will write a detailed description of them soon.
Paper
#1
: Zero-Knowledge Proofs of Training
Paper
#2
: Stateful Defenses for Machine Learning Models Are Not Yet Secure
This is my humble plea to
@NeurIPSConf
. Please extend the submission deadline to a few days after
@icmlconf
notification. This will make a lot of people happy!
👇 I really think this is harmful for a research community. I know some security-privacy venues have even a lower acceptance rate. I believe increasing acceptance rates will also decrease some of the randomness.
Discuss.
14% acceptance rate 😳
Admittedly there's a revision cycle this time, so a few more papers might get accepted after revision. But still. I find it hard to believe that so few papers are considered publishable.
Don't know why, but have been thinking of 👇 guy (Helmut Veith) today a lot. There are some folks who you just mesh with, and he was one of them. He was good friend and a collaborator. The CEGAR work in model checking was our first collaboration, but we had several more.
Happy Diwali to my friends. May this year be prosperous, successful, and peaceful.
The following music used to play in the morning on radios on Diwali when I was a kid . Shenai (the instrument) is played on momentous occasions in India.
It was exciting to host this workshop on the important topic of: Identifying risks and mitigation strategies of GenAI. The workshop was co-organized by Stanford, UW-Madison, and Google.
We need to standardize uploading of letters for grad-school and faculty positions. Right now everyone has slightly different submission form, and causes a lot pain angst (for me at least).
Happy Thanksgiving everyone.
The following message from a student in my class made my day.
Thank you for being such a great professor for my class this semester, I really enjoy watching you lecturelets and the smiley faces you draw are really cute, it makes my day every time.
Reviewer 3: I see you’ve solved global hunger, but it was always obvious that you could do that by working really hard, so we haven’t learned anything from your paper. Perhaps you could try solving global hunger using only purple foods? That would be novel.
End of an era! When I arrived at UW-Madison, I worked with Tom quite a bit on several topics (e.g. weighted pushdown systems). I learned so much from him. His clarity of thought and attention to detail are simply amazing! 🙏 Tom.
@WisconsinCS
20 positions open at OSU. That is like an entire new department:-)
Seriously, please apply. There are some great faculty at CSE
@OSU
. Plus I think they are leading two new AI institutes.
Great win for
@WisconsinCS
Welcome
@pdmcdan
to our department. Patrick and I have been collaborators and friends for a long time. Really excited about this hire.
I am happy to announce that after 18 great years at Penn State I will be moving to the University of Wisconsin-Madison (
@UWMadison
), joining as the Tsun-Ming Shih Professor in the Dept. of Computer Sciences (
@WisconsinCS
).
👇 Exactly right. In the same vein, some of my work that became popular (e.g. CEGAR, malware detection, attack graphs,...) I could not have predicted. You can never predict what catches on.
Professors at top universities are lottery winners, but rarely acknowledge the role of luck in their success. Be skeptical when they give you advice suggesting that the path they took is a repeatable one. If you aspire to an academic research career, have a backup plan.
Pet peeve of the day. N years back folks used to write journal papers expanding/combining work on one thread. Advantage: you can give the journal paper to a student and they get caught up with a sub-area. Why did we stop?:-( Two gems to showcase my reasoning:
Selective prediction can be used to improve the reliability of the LLMs by allowing them to abstain from making predictions when they are unsure of the answer. Read about this work at:
Google Research Blog:
Paper:
VentureBeat
Several positions open. My personal opinion: I have been very happy in this department for > n years, and I love the collaborative and friendly atmosphere.
If you thought an LLM can be protected against jailbreaks using yet another "guard LLM", you may want to read our paper -- we show that jailbreaks can be augmented with a carefully crafted prefix to foil the guard LLM.
Poignant story time. Every Sunday at my regular coffee shop, I see this guy have breakfast with his mom, and narrate all the happenings during the week (e.g. Emma had a great soccer game). It is pretty clear the mom has dementia. I don't know why, but it gets me every time
👇 This is a problem. If anything, the acceptance rate should be increased. Reviewers are not oracles who know everything. Increasing the acceptance rate reduces randomness.
From the
@icmlconf
Chairs: "...we plan to reduce the number of accepted papers. Please work with your SAC to raise the bar. AC/SAC do not have to accept a paper only because there is nothing wrong in it."
@MCAsche
Really like this message. Very positive. When I was in grad school, I was 2-3 yrs older than my peers and had a young kid. I really felt a bit out of place, but everything works out in the end. Cosmos always has a plan.
Modern cryptography is not possible without assuming that adversaries are computationally bounded. Can this assumption also help robust learning? Read our paper () to appear in ALT 2020. Congrats to my co-authors (Sanjam, Saeed, and Mohammad).
Born on this day in 1945, Edmund Melson Clarke received the 2007
#ACMTuringAward
together with E. Allen Emerson and Joseph Sifakis, for their role in developing Model-Checking into a highly effective verification technology.
#ACM
#Computing
#TuringAward
Congrats to my colleagues and collaborators
@awsTO
and Kassem Fawaz (get on Twitter Kassem:-)) for receiving the distinguished teaching award. It is humbling to be in presence of such great colleagues.
Two papers at
@IEEESSP
2020. Congrats to my amazing co-authors.
#1
: Towards Effective Differential Privacy Communication for Users' Data Sharing Decision and Comprehension
#2
: OAT: Attesting Operation Integrity of Embedded Devices
Last month or so I have written several letters. Took a huge amount of time. I do it it because I believe in "paying it forward", but there has to be a better way. For starters, why don't all universities use the same system, so that I have to upload my letter once.
This is super toxic! Reminds me of my favorite movie godfather. I am really sorry that you have to go through this
@roopshasamanta
It is sad and frankly quite pathetic.
Congrats to Tom Reps of
@WisconsinCS
for receiving the CAV award. The idea of context-bounded analysis for concurrent programs is so elegant and effective!
Also, Akash Lal (one the recipients of this award) was a PhD student at
@WisconsinCS
and now at MSR India.
Congrats to
Congratulations to Prof. Tom Reps, who recently earned the prestigious CAV Award (
@confCAV
) for a "simple and powerful" idea: the verification of concurrent programs through context-bounded analysis. Tom joins Prof.
@jhasomesh
, a 2015 CAV recipient.
Security/Privacy conferences have the revision option where reviewers ask you to do a bunch of stuff. Revision is reviewed again. It is quite tedious sometimes, but I think the paper comes out better at the end.
Would love to hear thoughts from others.
👇I agree with this. There are too many imperfections/randomness in the reviewing system. Having higher acceptance rate alleviates some of the randomness.
Thoughts from others?
Sorry
#AAAI2022
but a 15% acceptance rate is harmful to the community. Especially when I see lots of weak accept recommendations by the SPC+AC get overturned (~8% among my friend ACs). If SPC+AC, who are experts in the area, think it should be published, why waste everyone's time
Great move! I was on a fellowship while a grad student at CMU CS. As a grad student, I was married and had a kid. The fellowship had an allowance for each dependent (like 500/month). It really helped out. RA compensation needs improvement everywhere
This covid situation in India is really bad, which is a source of profound anxiety for me. This heart-wrenching article
captures the disastrous state of affairs.
Whenever I get an unfair review, I am reminded of the following ghazal:-) One of those days.
Kaise kaise log hamare, jee to ko jalane aa jatey hain
apne apne gam ke fasaane, hame sunane aa jate hain
different folks come to burn my heart
and they tell me there sob stories
Universities have started looking at faculty applications. I wanted to remind everyone that Tianhao Wang (
@bigflywth
) is on the market. He is absolutely top notch.