Jayadev Acharya
@acharyajayadev
Associate Professor @Cornell. Interested in Information Theory, Machine Learning, and Statistics.
ID: 3078444975
http://people.ece.cornell.edu/acharya/ 08-03-2015 00:56:48
668 Tweet
2,2K Followers
177 Following
![[📜New preprint] Interested in statistical inference under Local Differential Privacy (LDP)?
Let's look at the basic problem, "simple binary hypothesis testing": Given n iid samples known to be sampled from either distribution p or distribution q, identify the true distribution](https://pbs.twimg.com/media/FmE5YCMWAAExz38.jpg "Ankit Pensia (@subgaussian) on Twitter photo [📜New preprint] Interested in statistical inference under Local Differential Privacy (LDP)?
Let's look at the basic problem, \"simple binary hypothesis testing\": Given n iid samples known to be sampled from either distribution p or distribution q, identify the true distribution")
🧵New paper led by Jimmy Di & Jack Douglas, co-advised with Jayadev Acharya Ayush Sekhari (on Job Market): "Hidden Poison: Machine Unlearning Enables Camouflaged Poisoning Attacks" The adaptive nature of machine unlearning opens a new attack vector for adversaries. arxiv.org/abs/2212.10717 1/n
I am so very excited to share that I have received the U.S. National Science Foundation CAREER award on causal reasoning with information-theoretic methods! This would not have been possible without my mentors' guidance, and my collaborators' endless effort on our papers together. nsf.gov/awardsearch/sh…
"Thirty academic and industry speakers walk into a bar..." "... (and I am one of them)." More info tomorrow morning: rtbevent.com/sydney Raising the Bar Engineering Sydney Uni
Michael J Barany Waiting to give an award to a statistician until he is 102 years old? en.wikipedia.org/wiki/C._R._Rao
Very happy about this short and sweet paper (only 20 pages!) on communication-constrained bandits. We ask "How does the regret scale when the reward feedback is supplied over a AWGN channel?". With: Jonathan Scarlett and Vincent Tan. Arxiv: arxiv.org/abs/2304.12680
![[Today 5pm poster 401 #NeurIPS2023]
Is your LLM inference too slow?
We achieve 2.13x wallclock speedup in sampling from SOTA LLMs with 𝐩𝐫𝐨𝐯𝐚𝐛𝐥𝐲 no quality sacrifice.
How?
We use a cheap LM to draft 𝐦𝐮𝐥𝐭𝐢𝐩𝐥𝐞 samples; scored with the LLM to accept/reject tokens.](https://pbs.twimg.com/media/GBPWStfX0AAzniS.png "Ziteng Sun (@sziteng) on Twitter photo [Today 5pm poster 401 #NeurIPS2023]
Is your LLM inference too slow?
We achieve 2.13x wallclock speedup in sampling from SOTA LLMs with 𝐩𝐫𝐨𝐯𝐚𝐛𝐥𝐲 no quality sacrifice.
How?
We use a cheap LM to draft 𝐦𝐮𝐥𝐭𝐢𝐩𝐥𝐞 samples; scored with the LLM to accept/reject tokens.")
![[📜preprint] New results for an old problem
We revisit the basic statistics problem "simple binary hypothesis testing":
Given (i) two distributions p & q and (ii) n data points promised to be sampled either i.i.d. from p or i.i.d. from q, identify the true distribution](https://pbs.twimg.com/media/GJn-RfHWkAAhfp9.jpg "Ankit Pensia (@subgaussian) on Twitter photo [📜preprint] New results for an old problem
We revisit the basic statistics problem \"simple binary hypothesis testing\":
Given (i) two distributions p & q and (ii) n data points promised to be sampled either i.i.d. from p or i.i.d. from q, identify the true distribution")
We invite expository articles on generative models for the IEEE BITS magazine: itsoc.org/sites/default/… Feel free to reach out to me or other guest editors Ahmad Beirami, Marco Mondelli , @sametoymac, Vatsal Sharan if you have any questions.