ZCODE0's profile picture. Incoming Assistant Professor @ AI Trust, HKUST (GZ)

Shu Yao

@ZCODE0

Incoming Assistant Professor @ AI Trust, HKUST (GZ)

Épinglé

Hi guys! I am extremely honored to join the AI Thrust at HKUST (GZ) as an Assistant Professor in April 2025. I am recruiting Ph.D. and MPhil students for Fall 2025. Visit my personal website yao.notion.site for more info and feel free to email for the application.


LLMs struggling with prompt variations? 🤔 PAFT (Prompt-Agnostic Fine-Tuning) to the rescue! 🚀 We fine-tune for robustness, achieving state-of-the-art performance. Read the paper: arxiv.org/abs/2502.12859 #AI #NLP #LLMs #Prompt #Robustness

ZCODE0's tweet image. LLMs struggling with prompt variations? 🤔 PAFT (Prompt-Agnostic Fine-Tuning) to the rescue! 🚀 We fine-tune for robustness, achieving state-of-the-art performance. Read the paper: arxiv.org/abs/2502.12859
#AI #NLP #LLMs #Prompt #Robustness

Follow is now a lifesaver for me!

ZCODE0's tweet image. Follow is now a lifesaver for me!

Shu Yao a reposté

The #EMNLP2024 @emnlpmeeting (findings) position paper of @michael_xinyi @WuZhaoxuan @ray_qiaorui @arun_v3rma @PangWeiKoh et al. proposes a data-centric viewpoint of AI research, focusing on #LLM #LLMs. Check it out @ arxiv.org/abs/2406.14473

bryanklow's tweet image. The #EMNLP2024 @emnlpmeeting (findings) position paper of @michael_xinyi @WuZhaoxuan @ray_qiaorui @arun_v3rma @PangWeiKoh et al. proposes a data-centric viewpoint of AI research, focusing on #LLM #LLMs.

Check it out @ arxiv.org/abs/2406.14473

Shu Yao a reposté

Visit the poster of @ZCODE0 @xiaoqiang_98 @Dai_Zh et al. on Federated #ZerothOrderOptimization at @icmlconf #ICML2024 Workshop on Differentiable Almost Everything (26 Jul, differentiable.xyz)! Paper: differentiable.xyz/papers-2024/pa… #FederatedLearning


Shu Yao a reposté

The #ICLR2024 @iclr_conf work of @he_zhenfeng @Dai_Zh @ZCODE introduces RoBoT🤖 to robustify and boost training-free #NeuralArchitectureSearch. Join us @ Poster Session 6 May 9 Thu 4:30PM Halle B #250 Paper: openreview.net/pdf?id=qPloNoD… Code: github.com/hzf1174/RoBoT


Shu Yao a reposté

Our research group & collaborators have put together 4 chapters in the #FederatedLearning: Theory and Practice book: fairness (ch.8), #DataValuation (ch.15) & incentives (ch.16) in #FederatedLearning, and federated sequential decision making (ch.14). sciencedirect.com/book/978044319… (1/n)

bryanklow's tweet image. Our research group & collaborators have put together 4 chapters in the #FederatedLearning: Theory and Practice book: fairness (ch.8), #DataValuation (ch.15) & incentives (ch.16) in #FederatedLearning, and federated sequential decision making (ch.14).
sciencedirect.com/book/978044319… (1/n)

Shu Yao a reposté

When using #ChatGPT, how do u decide what instruction to give it? xqlin98.github.io/INSTINCT/ Joint work on Automatic Prompting with @xiaoqiang_98 @WuZhaoxuan @Dai_Zh @_Hu_Wenyang @ZCODE0 see-kiong @pjaillet. #LLM #LLMs #PromptEngineering #GenerativeAI (1/n)


Shu Yao a reposté

Congrats to @ZCODE0 for winning the best Ph.D. thesis award in 2023 @NUSComputing! His thesis topic is on #NeuralArchitectureSearch. URL: comp.nus.edu.sg/~lowkh/pubs/ph…

bryanklow's tweet image. Congrats to @ZCODE0 for winning the best Ph.D. thesis award in 2023 @NUSComputing! 

His thesis topic is on #NeuralArchitectureSearch.

URL: comp.nus.edu.sg/~lowkh/pubs/ph…

Shu Yao a reposté

The #ICLR2023 work of @ZCODE0 @Dai_Zh weicong @arun_v3rma @pjaillet @bryanklow proposes a query-efficient Zeroth-Order Optimization algo w. trajectory-informed derivative est. #BayesianOptimization #GaussianProcess Paper: openreview.net/pdf?id=n1bLgxH… Present: iclr.cc/virtual/2023/p…


Shu Yao a reposté

The #ICLR2023 work of @Dai_Zh @ZCODE0 @arun_v3rma @Flint_xf_Fan @bryanklow @pjaillet proposes the first federated neural contextual bandit algorithm (1/N). #FederatedLearning #BayesianOptimization Paper: openreview.net/pdf?id=38m4h8H… Presentation: iclr.cc/virtual/2023/p…


This is the first unified theoretical study to explain why existing training-free NAS algorithms perform well in practice and how to further improve them. We believe this work can help/inspire existing training-free NAS and also the follow-ups to behave more soundly in practice.

To understand and boost gradient-based #TrainingFree #NeuralArchitectureSearch algorithms, the #NeurIPS2022 work of @ZCODE0 @Dai_Zh @WuZhaoxuan @bryanklow provides the first unified theoretical study and principled improvement for them based on theory of #NeuralTangentKernel.



Shu Yao a reposté

The #NeurIPS2022 work of @Dai_Zh @ZCODE0 @bryanklow @pjaillet introduces theoretically grounded batch #BayesianOptimization algorithms using #DeepNeuralNetworks (DNNs) as the surrogate function that can handle categorical, high-dimensional, or image inputs. #NeuralTangentKernel


Shu Yao a reposté

To fairly trade off betw payoff & model rewards in collaborative ML, the #NeurIPS2022 work of @qphong @bryanklow @pjaillet refines #ShapleyValue into a conditional variant representing pairwise payoff flows betw parties. #FederatedLearning #DataValuation


Shu Yao a reposté

Loading...

Something went wrong.


Something went wrong.