Yao-Yuan Yang

@DeepMind | Ph.D. @UCSanDiego

[Thinking Machines @thinkymachines]

Today, we are introducing Inkling.

Inkling reasons efficiently across text, image, and audio modalities. We are making the full weights available.

https://t.co/Ghebq5mG30

Available today for fine-tuning on Tinker. Play with it in the Inkling Playground. 🧵

[Yao-Yuan Yang @yangarbiter]

Congrats to @AndrewDai and team on the launch. This is really a talented team. I worked with Andrew at Google DeepMind and have a lot of respect for his vision and leadership. Excited to see what’s ahead.

[Andrew M. Dai @AndrewDai]

After almost 12 years in Brain/DeepMind, I’ve finally decided to take the leap. My cofounders: @yinfeiy, Seth and I have kicked-off @ElorianAI. The first multimodal reasoning lab founded and led by former LLM pretraining, data and multimodal leads. https://t.co/XHcEtvl9F9 (1/n)

[Andrew M. Dai @AndrewDai]

After almost 12 years in Brain/DeepMind, I’ve finally decided to take the leap. My cofounders: @yinfeiy, Seth and I have kicked-off @ElorianAI. The first multimodal reasoning lab founded and led by former LLM pretraining, data and multimodal leads. https://t.co/XHcEtvl9F9 (1/n)

[Angel Hsing-Chi Hwang @angelhwang6]

@yangarbiter and I verified this concern by tracking the performance of 9,149 freelancers across two platforms (Upwork and Bēhance): Creators who declare the use of AI receive significantly lower pay, but non-creatives jobs earn more by labeling themselves as "AI Pros."

[Bogdan Kulynych @hiddenmarkov]

Presenting this paper in person at #NeurIPS2022 tomorrow (Tue) at 4pm. We apply the connections between privacy and generalization to (1) reduction of disparate impact of DP-SGD, (2) group DRO, and (3) adversarial training. Let's chat if you're around!
https://t.co/KqVfn4SU2W

[Bogdan Kulynych @hiddenmarkov]

A new preprint is out!
“What You See is What You Get: Distributional Generalization for Algorithm Design in Deep Learning” w/ @yangarbiter, @yaodong_yu, Jarek Błasiok, & @PreetumNakkiran, in which we show that differential privacy can ensure predictable model behavior.

[arXiv Daily @Arxiv_Daily]

Understanding Rare Spurious Correlations in Neural Networks
https://t.co/pKA1lsZ1r4
by @yangarbiter et al.
#NeuralNetwork #Classifier

[Michal Moshkovitz @ML_Theorist]

Interested in models which are both interpretable and robust to adversarial perturbation?

Joint work with @yangarbiter and @kamalikac
Accepted to #ICML2021
1/5

[Yao-Yuan Yang @yangarbiter]

Our paper "Robustness for Non-Parametric Classification: A Generic Attack and Defense" is on #AISTATS2020

live Q&A time at Aug 26, 27 10:00AM PDT

poster: https://t.co/I5Sm31LVau
blog post: https://t.co/T2ISgcG9Ha

Work w/ @CyrusRashtchian, Yizhen Wang and @kamalikac

[Cyrus Rashtchian 🇰🇷 ICML @CyrusRashtchian]

Real image datasets are well-separated, and in retrospect, we expect them to be because they are clean and curated. So what's up with adversarial examples?! A thread #MachineLearning #DataScience [1/10]

[Max Hopkins @MHop_Theory]

Finally, big thanks to @kamalikac, @yangarbiter, and @CyrusRashtchian for blog upkeep and help with the post!

[Kamalika Chaudhuri @kamalikac]

New blog post: https://t.co/5CDcuVucOr
Paper: https://t.co/wcDaChbSeK

TLDR; we give a general defense against adversarial examples for non-parametric methods and attacks to evaluate.

Work w/ @CyrusRashtchian @yangarbiter Y. Wang

[Russ Salakhutdinov @rsalakhu]

New blog post: Adversarial Robustness Through Local Lipschitzness:

https://t.co/a3WLT4OEWD

We posit that achieving both accuracy and robustness is possible by learning locally smooth classifiers.

Paper: https://t.co/7dH1b2hWv3

w/t Y Yang, @CyrusRashtchian, H Zhang, @kamalikac

[Stat.ML Papers @StatMLPapers]

Adversarial Robustness Through Local Lipschitzness. (arXiv:2003.02460v1 [cs.LG]) https://t.co/HPyaAa1SdN

[Machine Learning | arXiv @arXiv__ml]

#arXiv #machinelearning [cs.LG] Adversarial Robustness Through Local Lipschitzness. (arXiv:2003.02460v1 [cs.LG]) https://t.co/MfBxy7bBqK

A standard method for improving the robustness of neural networks is adversarial training, where the network is trained on adversarial examp…

[Cyrus Rashtchian 🇰🇷 ICML @CyrusRashtchian]

Tough questions from @kamalikac at the #ITA2020 ML panel: Is deep learning going to solve climate change? @rsalakhu defers to Turing award winners but also "obviously not"

[Cyrus Rashtchian 🇰🇷 ICML @CyrusRashtchian]

"Open problem: explain this plot" 😂Another good theory question from @roydanroy during the #ITA2020 plenary.

[Statistics (Machine Learning) Papers @StatsPapers]

Adversarial Examples for Non-Parametric Methods: Attacks, Defenses and Large Sample Limits. https://t.co/tSL73rWTtq

[Gautam Kamath @thegautamkamath]

When your poster isn't ready in time, you have to improvise. #NeurIPS2019