RT @karpathy: RT to help Simon raise awareness of prompt injection attacks in LLMs. Feels a bit like the wild west of early computing, wit….

RT @cseiitd: Looking to pursue a Ph.D. in Computer Science at one of India's top institutes?.📢 CSE IIT Delhi invites passionate researchers….

RT @florian_tramer: I'm thrilled to start Invariant Labs with amazing colleagues from @CSatETH. Our ambition is to build AI agents that so….

We believe that our work has broad implications to ML security in high dimensions (like training LLMs). You can find our paper here, I am glad to have collaborated with Sarthak Choudhary and Prateek Saxena on this work (. (4/4).

We show that there may be a fundamental computational hurdle that robust aggregators cannot bypass. Further, we show that the ones that seem to do in practice (see , are not adequate defenses by designing an optimal poisoning attack called HIDRA. (3/4).

Robust mean estimation is a fundamental problem with numerous applications, particularly for ML security. Robust aggregators are necessary to defend against poisoning attacks (backdoor, untargeted) during ML training. (2/4).

I will present our recent work on attacking byzantine robust aggregators at IEEE S&P'24 (@IEEESSP). HMU at the conference for discussing our work and anything related to ML security (backdoor attacks, defenses, adversarial attacks, and differential privacy)!. (1/4). #ieeesp2024.

RT @flwrlabs: 🥁 It's time to meet another Flower SoR contributor 💪. @aashishkolluri6, Ph.D. student @NUSingapore, implemented "#federatedle….

RT @DrJimFan: A neural network can smell like humans do for the first time!👃🏽. Digital smell is a modality that AI community has long ignor….

RT @ProfNoahGian: Here's a recap of some recent ChatGPT events that, even if you don't care about the moral dimensions, should give investo….

RT @mskumars1729: Glad to be part of the winning team from the UK! Our effort at the PETs prize challenge is an embodiment of everything th….

You’d think this is a joke until you get them. ChatGPT could probably give better reviews at this point. Oh, may be we can train LLMs for reviewing?.

RT @random_walker: TikTok users feel its algorithm is eerily accurate. But from everything we know about the algorithm, it's a pretty stand….

RT @Franc0Fernand0: System design is the most challenging and exciting type of tech interview. Here is a complete list of topics that will….

RT @teobaluta: Can you get a good utility-privacy tradeoff when you want to learn graph neural networks and keep the edges private? @aashis….

RT @thegautamkamath: Just want to highlight that this very cool work was based out of the Google Brain *Toronto* team. Always awesome to se….

RT @thedavidbrumley: Congrats to @ElaineRShi, Bryan Parno, @zardus, Sang Kil Cha, Manuel Egele, and others for being listed among the most….

RT @yufeizhao: Congratulations to Jinyoung Park and Huy Pham for proving the Kahn-Kalai conjecture---a central open problem in probabilisti….

RT @vardi: Nerd joke :-)

RT @yvesalexandre: 🚨 We analyzed in @NatureComms the guarantees of an anonymous “Differentially Private” mobility dataset of 300M @googlema….