I do normal science.
New York, NY
Joined September 2009
- Tweets1.7K
- Following287
- Followers566
- Likes5.2K
Sanyam Kapoor retweeted
If we want to use LLMs for decision making, we need to know how confident they are about their predictions. LLMs don’t output meaningful probabilities off-the-shelf, so here’s how to do it 🧵
Paper: arxiv.org/abs/2406.08391
Thanks @psiyumm and @gruver_nate for leading the charge!
Sanyam Kapoor retweeted
Find us at burning man 2024 🔥
@gruver_nate @LotfiSanae @psiyumm @samuel_stanton_ @polkirichenko @Pavel_Izmailov @KuangYilun
@m_finzi @timrudner @ShikaiQiu @yucenlily @andrewgwils
Most likely functions and most likely parameters that describe the data may differ. How much does this matter?
Read on to learn more in our new #NeurIPS2023 paper!
When training machine learning models, should we learn most likely parameters—or most likely functions?
We investigate this question in our #NeurIPS2023 paper and made some fascinating observations!🚀
Paper: arxiv.org/abs/2311.15990
w/ @ShikaiQiu @psiyumm @andrewgwils
🧵1/10
Sanyam Kapoor retweeted
📢 I am recruiting Ph.D. students for my new lab at @nyuniversity! Please apply, if you want to work on understanding deep learning and large models, and do a Ph.D. in the most exciting city on earth.
Details on my website: izmailovpavel.github.io/. Please spread the word!
Sanyam Kapoor retweeted
LLMs aren't just next-word predictors, they are also compelling zero-shot time series forecasters! Our new NeurIPS paper:
arxiv.org/abs/2310.07820
w/ @gruver_nate, @m_finzi, @ShikaiQiu
1/7
Sanyam Kapoor retweeted
We're ecstatic to officially announce our new library, CoLA! CoLA is a framework for large-scale linear algebra in machine learning and beyond, supporting PyTorch and JAX.
repo: github.com/wilson-labs/cola
paper: arxiv.org/abs/2309.03060
w/ amazing @m_finzi, Andres Potap, Geoff Pleiss
🚨 Come join us at our poster “On Uncertainty, Tempering, and Data Augmentation in Bayesian Classification” at #NeurIPS2022 today (Dec 1) w/ Wesley, @Pavel_Izmailov @andrewgwils 11am-1pm Hall J #715 🚨 neurips.cc/virtual/2022/post…; (Paper: arxiv.org/abs/2203.16481)
We explore how to represent aleatoric (irreducible) uncertainty in Bayesian classification, with profound implications for performance, data augmentation, and cold posteriors in BDL.
arxiv.org/abs/2203.16481
w/@snymkpr, W. Maddox, @andrewgwils
🧵 1/16
Sanyam Kapoor retweeted
I'm so proud that our paper on the marginal likelihood won the Outstanding Paper Award at #ICML2022!!! Congratulations to my amazing co-authors @Pavel_Izmailov, @g_benton_, @micahgoldblum, @andrewgwils 🎉
Talk on Thursday, 2:10 pm, room 310
Poster 828 on Thursday, 6-8 pm, hall E
I'm happy that this paper will appear as a long oral at #ICML2022! It's the culmination of more than a decade of thinking about when the marginal likelihood does and doesn't make sense for model selection and hyper learning, and why. It was also a great collaborative effort.
Sanyam Kapoor retweeted
Pre-Train Your Loss: Easy Bayesian Transfer Learning with Informative Priors.
arxiv.org/abs/2205.10279
w/@ziv_ravid, @micahgoldblum, @HosseinSouri8, @snymkpr, @Eiri1114, @ylecun 1/6
How do we compare between hypotheses that are entirely consistent with the observations? See what @LotfiSanae has to say! 📈
Our next #MoroccoAI webinar will be taking place this Wednesday, the 27th of April! A webinar on 'The Promises and Pitfalls of the marginal likelihood', with Sanae LOTFI.
Please take a minute to RSVP to receive event Zoom link,
us02web.zoom.us/.../tZckdOyp…...
#MoroccoAI #AI #morocco
Sanyam Kapoor retweeted
Last Layer Re-Training is Sufficient for Robustness to Spurious Correlations. ERM learns multiple features that can be reweighted for SOTA on spurious correlations, reducing texture bias on ImageNet, & more!
w/ @Pavel_Izmailov and @andrewgwils
arxiv.org/abs/2204.02937
1/11
Sanyam Kapoor retweeted
New ICLR 2022 paper w/ @neiljethani @ianccovert @suinleelab and Rajesh Ranganath! Our ML interpretability method, FastSHAP, significantly speeds up Shapley value estimation by amortizing SHAP/KernelSHAP computations across a training dataset. [📜:openreview.net/forum?id=Zq2G…]
Many people in XAI prefer SHAP, but SHAP can be very slow in practice. Our new ICLR 2022 paper addresses this problem by introducing FastSHAP, a new method to estimate Shapley values in a single forward pass using a learned explainer model openreview.net/forum?id=Zq2G… 🧵⬇️
Sanyam Kapoor retweeted
We explore how to represent aleatoric (irreducible) uncertainty in Bayesian classification, with profound implications for performance, data augmentation, and cold posteriors in BDL.
arxiv.org/abs/2203.16481
w/@snymkpr, W. Maddox, @andrewgwils
🧵 1/16
Sanyam Kapoor retweeted
Contrary to expectations, energy conservation and symplecticity are not primarily responsible for the good performance of Hamiltonian neural networks! Our #ICLR2022 paper: arxiv.org/abs/2202.04836
with @m_finzi, @samscub, and @andrewgwils. 1/7
Sanyam Kapoor retweeted
Very excited to give a talk at AABI tomorrow (Feb 1st) at 5PM GMT / 12PM ET!
I will be talking about our recent work on HMC for Bayesian neural networks, cold posteriors, priors, approximate inference and BNNs under distribution shift. Please join!
Join us to discuss the latest advances in approximate inference and probabilistic models at AABI 2022 on Feb 1-2!
Webinar registration: approximateinference.org/sch…
We have an amazing line-up of speakers, panelists and papers👍
@vincefort @Tkaraletsos @s_mandt @ruqi_zhang
Sanyam Kapoor retweeted
1/2
A Russian mathematician is hired by a math department in the US, and is assigned to teach Calculus 1. On the day before her first lecture, she asks a colleague: "what am I supposed to teach in this class?"
The colleague says, "well, it's standard first-semester calculus...
Sanyam Kapoor retweeted
I heard a rumour there is this amazing Approximate Inference in Bayesian Deep Learning competition at #NeurIPS2021 tomorrow, starting at 1 pm ET. From what I understand, the winners will be revealing their solutions, and the link to join is neurips.cc/virtual/2021/comp…. 🤫
"It takes a long time to learn to live - by the time you learn your time is gone." - Paul R. Halmos amazon.com/I-Want-be-Mathema…
Sanyam Kapoor retweeted
We’re happy to share our new #NeurIPS2021 paper introducing Residual Pathway Priors (RPPs), which convert hard architectural constraints into soft inductive biases! arxiv.org/abs/2112.01388
Joint work with @m_finzi, @andrewgwils. 1/9
