Gautam Kamath
@gautamkamath.com
Assistant Prof of CS at the University of Waterloo, Faculty and Canada CIFAR AI Chair at the Vector Institute. Joining NYU Courant in September 2026. Co-EiC of TMLR. My group is The Salon. Privacy, robustness, machine learning.
http://www.gautamkamath.com
http://www.gautamkamath.com
The last invited talk next Friday at #AAAI2026 will be Katerina Fragkiadaki (CMU) titled "Learning World Simulators from Data." It'll be Friday, January 23 at 5:35 PM! CC @aaai.org @hadihoss.bsky.social
January 13, 2026 at 4:31 PM
The last invited talk next Friday at #AAAI2026 will be Katerina Fragkiadaki (CMU) titled "Learning World Simulators from Data." It'll be Friday, January 23 at 5:35 PM! CC @aaai.org @hadihoss.bsky.social
Another invited talk at #AAAI2026 will be Yolanda Gil (USC), titled "From Workflows to Water Coolers: AI That Can Navigate Human Nature."
Check it out Friday, January 23 at 2 PM! CC @aaai.org @hadihoss.bsky.social
Check it out Friday, January 23 at 2 PM! CC @aaai.org @hadihoss.bsky.social
January 12, 2026 at 4:10 PM
Another invited talk at #AAAI2026 will be Yolanda Gil (USC), titled "From Workflows to Water Coolers: AI That Can Navigate Human Nature."
Check it out Friday, January 23 at 2 PM! CC @aaai.org @hadihoss.bsky.social
Check it out Friday, January 23 at 2 PM! CC @aaai.org @hadihoss.bsky.social
Across the pond this week -- visiting NYU (London!) for a joint workshop with researchers from Oxford. Will also be giving a talk at ICL on Thursday. Drop by if you're around!
January 12, 2026 at 9:57 AM
Across the pond this week -- visiting NYU (London!) for a joint workshop with researchers from Oxford. Will also be giving a talk at ICL on Thursday. Drop by if you're around!
The third invited talk at #AAAI2026 will be by Daniel Whiteson (@danielwhiteson.bsky.social, UC Irvine)!
His talk is titled "Fundamental Physics and Science Communication." Check it out at 8:30 AM on Friday, January 23 @aaai.org @hadihoss.bsky.social
His talk is titled "Fundamental Physics and Science Communication." Check it out at 8:30 AM on Friday, January 23 @aaai.org @hadihoss.bsky.social
January 9, 2026 at 4:48 PM
The third invited talk at #AAAI2026 will be by Daniel Whiteson (@danielwhiteson.bsky.social, UC Irvine)!
His talk is titled "Fundamental Physics and Science Communication." Check it out at 8:30 AM on Friday, January 23 @aaai.org @hadihoss.bsky.social
His talk is titled "Fundamental Physics and Science Communication." Check it out at 8:30 AM on Friday, January 23 @aaai.org @hadihoss.bsky.social
The second invited talk at #AAAI2026 will be by Bowen Zhou (Shanghai AI Lab, Tsinghua University)!
His talk is titled "Specialized Generalist: Towards High-Efficiency AGI." Check it out at 5:35PM on Thursday, January 22 @aaai.org @hadihoss.bsky.social
His talk is titled "Specialized Generalist: Towards High-Efficiency AGI." Check it out at 5:35PM on Thursday, January 22 @aaai.org @hadihoss.bsky.social
January 8, 2026 at 3:00 PM
The second invited talk at #AAAI2026 will be by Bowen Zhou (Shanghai AI Lab, Tsinghua University)!
His talk is titled "Specialized Generalist: Towards High-Efficiency AGI." Check it out at 5:35PM on Thursday, January 22 @aaai.org @hadihoss.bsky.social
His talk is titled "Specialized Generalist: Towards High-Efficiency AGI." Check it out at 5:35PM on Thursday, January 22 @aaai.org @hadihoss.bsky.social
If you are a senior researcher (i.e., associate prof level or later), and you're interested in moving to the University of Waterloo (best CS program in Canada) for the Canada Impact+ Research Chair (biggest chair position in Canada, $$$$$), email me. Discretion guaranteed.
January 6, 2026 at 3:26 PM
If you are a senior researcher (i.e., associate prof level or later), and you're interested in moving to the University of Waterloo (best CS program in Canada) for the Canada Impact+ Research Chair (biggest chair position in Canada, $$$$$), email me. Discretion guaranteed.
Some may remember this ACM guidance on inclusive terminology. E.g., as advocated by an anon ICLR reviewer, it recommends against the technical term Byzantine.
It was recently updated, and suggests avoiding "binary classification" and "stable marriage" (incorrectly defined)
It was recently updated, and suggests avoiding "binary classification" and "stable marriage" (incorrectly defined)
December 28, 2025 at 5:53 PM
Some may remember this ACM guidance on inclusive terminology. E.g., as advocated by an anon ICLR reviewer, it recommends against the technical term Byzantine.
It was recently updated, and suggests avoiding "binary classification" and "stable marriage" (incorrectly defined)
It was recently updated, and suggests avoiding "binary classification" and "stable marriage" (incorrectly defined)
This is not how mathematics works. For centuries, claiming a result obligates you to defend it. The bigger the claimed result, the larger the obligation.
This is particularly true in the era of AI models, where claiming a result is essentially free, but verifying is expensive
This is particularly true in the era of AI models, where claiming a result is essentially free, but verifying is expensive
December 22, 2025 at 7:02 PM
This is not how mathematics works. For centuries, claiming a result obligates you to defend it. The bigger the claimed result, the larger the obligation.
This is particularly true in the era of AI models, where claiming a result is essentially free, but verifying is expensive
This is particularly true in the era of AI models, where claiming a result is essentially free, but verifying is expensive
If you have a permanent job due to publishing in ML, you owe OpenReview a lot.
Please consider donating -- feel free to share donations below!
If you work in tech (e.g., MS, Google, etc.), your employer may match donations. We donated through my wife at Microsoft.
openreview.net/donate
Please consider donating -- feel free to share donations below!
If you work in tech (e.g., MS, Google, etc.), your employer may match donations. We donated through my wife at Microsoft.
openreview.net/donate
December 19, 2025 at 6:35 PM
If you have a permanent job due to publishing in ML, you owe OpenReview a lot.
Please consider donating -- feel free to share donations below!
If you work in tech (e.g., MS, Google, etc.), your employer may match donations. We donated through my wife at Microsoft.
openreview.net/donate
Please consider donating -- feel free to share donations below!
If you work in tech (e.g., MS, Google, etc.), your employer may match donations. We donated through my wife at Microsoft.
openreview.net/donate
I'd be a proud dad too
(Spotted on a blackboard at UWaterloo)
(Spotted on a blackboard at UWaterloo)
December 17, 2025 at 7:56 PM
I'd be a proud dad too
(Spotted on a blackboard at UWaterloo)
(Spotted on a blackboard at UWaterloo)
Ending a fantastic week at #NeurIPS2025. Saw a lot of friends (old and new), ate a lot of tacos, gave away a lot of stickers. See you at the next one!
December 8, 2025 at 5:52 PM
Ending a fantastic week at #NeurIPS2025. Saw a lot of friends (old and new), ate a lot of tacos, gave away a lot of stickers. See you at the next one!
IJCAI 2026 will charge $100 USD per submission. Funds will be used to compensate reviewers.
2026.ijcai.org/ijcai-ecai-2...
2026.ijcai.org/ijcai-ecai-2...
December 4, 2025 at 3:25 PM
IJCAI 2026 will charge $100 USD per submission. Funds will be used to compensate reviewers.
2026.ijcai.org/ijcai-ecai-2...
2026.ijcai.org/ijcai-ecai-2...
If you are a senior researcher at #NeurIPS2025 (i.e., roughly full professor level or later), and you're interested in moving to the University of Waterloo (best CS program in Canada) for a CERC (biggest chair position in Canada), email me. Discretion guaranteed.
December 3, 2025 at 2:51 PM
If you are a senior researcher at #NeurIPS2025 (i.e., roughly full professor level or later), and you're interested in moving to the University of Waterloo (best CS program in Canada) for a CERC (biggest chair position in Canada), email me. Discretion guaranteed.
With nearly 30,000 registrants (including virtual), 64% increase since last year, they're saying that #NeurIPS2025 is the largest conference at San Diego Convention Center
December 3, 2025 at 1:38 AM
With nearly 30,000 registrants (including virtual), 64% increase since last year, they're saying that #NeurIPS2025 is the largest conference at San Diego Convention Center
Had a great time as a mentor at the WIML workshop! Got to mentor a table with Tatiana Likhomanenko, who is both an Action Editor and Expert Reviewer for
@tmlrorg.bsky.social! #NeurIPS2025
@tmlrorg.bsky.social! #NeurIPS2025
December 2, 2025 at 8:52 PM
Had a great time as a mentor at the WIML workshop! Got to mentor a table with Tatiana Likhomanenko, who is both an Action Editor and Expert Reviewer for
@tmlrorg.bsky.social! #NeurIPS2025
@tmlrorg.bsky.social! #NeurIPS2025
We were then able to show that the domination number of this graph was O(k^1.5), leading to a non-interactive algorithm with the same sample complexity! Recall this is better than the naive O(k^2) sample algorithm, and "halfway" to the lower bound of Omega(k). 9/n
November 28, 2025 at 7:03 PM
We were then able to show that the domination number of this graph was O(k^1.5), leading to a non-interactive algorithm with the same sample complexity! Recall this is better than the naive O(k^2) sample algorithm, and "halfway" to the lower bound of Omega(k). 9/n
Until my student Matt Regehr came along! He came up with a brilliant generalization of the minimum distance estimator! With this framework, he reduced the problem to analyzing the domination number of a particular graph structure. 8/n
November 28, 2025 at 7:03 PM
Until my student Matt Regehr came along! He came up with a brilliant generalization of the minimum distance estimator! With this framework, he reduced the problem to analyzing the domination number of a particular graph structure. 8/n
But what if you consider the stronger constraint of local differential privacy? The bad news: a construction of Duchi and Rogers implies that you need Omega(k) samples. Exponentially more than in the centrally private case! 5/n
November 28, 2025 at 7:03 PM
But what if you consider the stronger constraint of local differential privacy? The bad news: a construction of Duchi and Rogers implies that you need Omega(k) samples. Exponentially more than in the centrally private case! 5/n
Hypothesis selection is the following problem: given n samples from a probability distribution, and k hypothesis distributions, how do you learn which one is (nearly) the closest? 2/n
November 28, 2025 at 7:03 PM
Hypothesis selection is the following problem: given n samples from a probability distribution, and k hypothesis distributions, how do you learn which one is (nearly) the closest? 2/n
🧵New paper, at #NeurIPS2025: "Query-Efficient Locally Private Hypothesis Selection via the Scheffe Graph" with Alireza F. Pour, Matthew Regehr (@matt19234.bsky.social), David P. Woodruff
Breaks the O(k^2) barrier I've been stuck on since 2019! arxiv.org/abs/2509.16180 1/n
Breaks the O(k^2) barrier I've been stuck on since 2019! arxiv.org/abs/2509.16180 1/n
November 28, 2025 at 7:03 PM
🧵New paper, at #NeurIPS2025: "Query-Efficient Locally Private Hypothesis Selection via the Scheffe Graph" with Alireza F. Pour, Matthew Regehr (@matt19234.bsky.social), David P. Woodruff
Breaks the O(k^2) barrier I've been stuck on since 2019! arxiv.org/abs/2509.16180 1/n
Breaks the O(k^2) barrier I've been stuck on since 2019! arxiv.org/abs/2509.16180 1/n
Wanna do a postdoc at NYU? We have postings for Faculty Fellows at both @nyudatascience.bsky.social and @nyucourant.bsky.social CS in the new School for Mathematics, Computing, and Data Science!
Come work with the best!
Courant CS: apply.interfolio.com/175537
CDS: apply.interfolio.com/174686
Come work with the best!
Courant CS: apply.interfolio.com/175537
CDS: apply.interfolio.com/174686
November 21, 2025 at 4:00 PM
Wanna do a postdoc at NYU? We have postings for Faculty Fellows at both @nyudatascience.bsky.social and @nyucourant.bsky.social CS in the new School for Mathematics, Computing, and Data Science!
Come work with the best!
Courant CS: apply.interfolio.com/175537
CDS: apply.interfolio.com/174686
Come work with the best!
Courant CS: apply.interfolio.com/175537
CDS: apply.interfolio.com/174686
A modest demonstration, but Nano Banana Pro with Gemini 3 Pro gets this simple task mostly right! At least no imaginary people this time.
(The aspect ratios are a bit off for some)
(The aspect ratios are a bit off for some)
November 21, 2025 at 12:50 PM
A modest demonstration, but Nano Banana Pro with Gemini 3 Pro gets this simple task mostly right! At least no imaginary people this time.
(The aspect ratios are a bit off for some)
(The aspect ratios are a bit off for some)
Thoughtful (as always) blog post from Nicholas Carlini. "Are large language models worth it?" A nice read giving his perspective on risks of ML models.
Post: nicholas.carlini.com/writing/2025...
For people who prefer, this is the video of the talk from @colmweb.org www.youtube.com/watch?v=PngH...
Post: nicholas.carlini.com/writing/2025...
For people who prefer, this is the video of the talk from @colmweb.org www.youtube.com/watch?v=PngH...
November 19, 2025 at 4:56 PM
Thoughtful (as always) blog post from Nicholas Carlini. "Are large language models worth it?" A nice read giving his perspective on risks of ML models.
Post: nicholas.carlini.com/writing/2025...
For people who prefer, this is the video of the talk from @colmweb.org www.youtube.com/watch?v=PngH...
Post: nicholas.carlini.com/writing/2025...
For people who prefer, this is the video of the talk from @colmweb.org www.youtube.com/watch?v=PngH...
At least if the reviewer ghosts you, you know you have to find another reviewer...
(The reviewer did come through at the 11th and a half hour though!)
(The reviewer did come through at the 11th and a half hour though!)
November 17, 2025 at 6:09 PM
At least if the reviewer ghosts you, you know you have to find another reviewer...
(The reviewer did come through at the 11th and a half hour though!)
(The reviewer did come through at the 11th and a half hour though!)