chang ma
@ma_chang_nlp
Followers
601
Following
2K
Statuses
106
Ph.D student @HKUNLP, previously @PKU1898, I work on the intersection of #AI4Science and NLP
Shanghai/Beijing
Joined May 2022
🥳 Introducing our #ICLR2025 paper: "Non-myopic Generation of Language Models for Reasoning and Planning" TLDR: We introduce an inference time decoding method, Predictive-Decoding, that takes inspiration from Model Predictive Control (MPC) to achieve optimal reasoning and planning with LLMs. 📰 paper link: 🧵[1/7]
1
17
63
RT @lockonlvange: Introducing CodeI/O (, a systematic way to condense diverse reasoning patterns via code input-out…
0
39
0
RT @_zhihuixie: Introducing CTRL, a new framework that trains LLMs to critique via RL without human supervision or distillation, enabling t…
0
55
0
RT @junxian_he: We replicated the DeepSeek-R1-Zero and DeepSeek-R1 training on 7B model with only 8K examples, the results are surprisingly…
0
667
0
Congratulations to @t_feyuan! KS-Lottery is accepted by #NAACL2025. We provide a theoretical guaranteed solution for finding lottery tickets for multilingual LLM. The results are stunning: fine-tuning the embedding of 18 tokens would be enough for learning new multilingual tasks. 📰 paper link:
0
2
14
Equally excited to share our ICLR rejected work: "Benchmarking and Enhancing Large Language Models for Biological Pathway Reasoning". 👨🔬This work by Haiteng @Syd59067213 is one major step towards benchmarking complex biological reasoning. 📰 link: This work unfortunately received no reply during the discussion period. But it is definitely one of my favourite work on LLM benchmarking -- Instead of testing on K-12 level scientific knowledge, we directly create a benchmark based on PubMed papers. We explore whether LLM could make scientific discovery like real biologists!
0
4
25
Thanks to my amazing collaborators @ikekong @junxian_he @Syd59067213 @junleiz0609 ! We have also proved that teaching LLMs to be less myopic could be a cheap solution for reasoning post-training. Stay tuned for our next release!
0
0
4
RT @linzhengisme: 🚀 Meet EvaByte: The best open-source tokenizer-free language model! Our 6.5B byte LM matches modern tokenizer-based LMs w…
0
83
0
RT @AndrewZeng17: 🚀 Excited to share our latest research: B-STAR! 💡 Tackling the stagnation in self-improvement, we present a framework th…
0
25
0
RT @qiushi_sun: 🎉Introducing our latest work on GUI Agents: "OS-Genesis: Automating GUI Agent Trajectory Construction via Reverse Task Synt…
0
42
0
RT @WeiLiu99: 🔔🎄Christmas Gift for Multimodal Reasoning: Introducing M-STaR 🎁 (1/6) How can we dive deeper to help Large Multimodal Models…
0
36
0
RT @yihengxu_: 1/ 🚀 Introducing AGUVIS: A unified, pure vision-based agent model for autonomous GUI interaction! It seamlessly operates acr…
0
60
0
I learned from my advisors (all Chinese) and friends how to keep the highest standards of integrity and placing integrity way ahead of academic success: showing failure cases and negative results, preprinting only something that works, maintaining open source, prizing paper solidness over quantity…. Chinese scholars may not be among the privileged, but undoubtedly passionate and truthful, which makes the accusation extra painful.
NeurIPS acknowledges that the cultural generalization made by the keynote speaker today reinforces implicit biases by making generalisations about Chinese scholars. This is not what NeurIPS stands for. NeurIPS is dedicated to being a safe space for all of us. We want to address the comment made during the invited talk this afternoon, as it is something that NeurIPS does not condone and it doesn't align with our code of conduct. We are addressing this issue with the speaker directly. NeurIPS is dedicated to being a diverse and inclusive place where everyone is treated equally.
0
2
52