Che-Ping Tsai @chepingt

[0/3] 🚀 Introducing Verlog – an open-source RL framework built specifically for training long-horizon, multi-turn LLM agents. 📊 Max episode length comparison: •VeRL / RAGEN → ~10 turns •verl-agent → ~50 turns •Verlog (ours) → 400+ turns 🔥 ⚙️ Technical foundation:…

Self-Questioning Language Models: LLMs that learn to generate their own questions and answers via asymmetric self-play RL. There is no external training data – the only input is a single prompt specifying the topic.

Yes, there is an official marking guideline from the IMO organizers which is not available externally. Without the evaluation based on that guideline, no medal claim can be made. With one point deducted, it is a Silver, not Gold.

Mikhail Samin @Mihonarium

2 months ago

Yes, there is an official marking guideline from the IMO organizers which is not available externally. Without the evaluation based on that guideline, no medal claim can be made. With one point deducted, it is a Silver, not Gold.

60 197 2K 462K 502

Download Image

I'll be at ICML this week to present our take on "what we're really learning in representation learning and why it works." Our central argument: "Representations are learned from the association between input 𝑋 and context variable 𝐴"

Thrilled to share our #ICML2025 paper! We introduce a variational approach for speech language models, automating speech attribute learning to deliver more natural, human-like speech. Joint work b/w @LTIatCMU and @Apple Read it: arxiv.org/abs/2506.14767

Can LLM solve PDEs? 🤯 We present CodePDE, a framework that uses LLMs to automatically generate solvers for PDE and outperforms human implementation! 🚀 CodePDE demonstrates the power of inference-time algorithms and scaling for PDE solving. More in 🧵: #ML4PDE #AI4Science

keyboard_arrow_left Previous keyboard_arrow_right Next

Introducing e3 🔥 Best <2B model on math 💪 Are LLMs implementing algos ⚒️ OR is thinking an illusion 🎩.? Is RL only sharpening the base LLM distrib. 🤔 OR discovering novel strategies outside base LLM 💡? We answer these ⤵️ 🚨 arxiv.org/abs/2506.09026 🚨 matthewyryang.github.io/e3/

🔥Unlocking New Paradigm for Test-Time Scaling of Agents! We introduce Test-Time Interaction (TTI), which scales the number of interaction steps beyond thinking tokens per step. Our agents learn to act longer➡️richer exploration➡️better success Paper: arxiv.org/abs/2506.07976

Excited to share our work with my amazing collaborators, @Goodeat258, @SimulatedAnneal, @zicokolter, and Kaiming. In a word, we show an “identity learning” approach for generative modeling, by relating the instantaneous/average velocity in an identity. The resulting model,…

Data selection and curriculum learning can be formally viewed as a compression protocol via prequential coding. New blog (with @AllanZhou17 ) about this neat idea that motivated ADO but didn’t make it into the paper. yidingjiang.github.io/blog/post/curr…

In our #AISTATS2025 paper, we ask: when it is possible to recover a consistent joint distribution from conditionals? We propose path consistency and autoregressive path consistency—necessary and easily verifiable conditions. See you at Poster session 3, Monday 5th May.

Check out Runtian's thesis on contexture theory, which shows that many representation learning methods perform eigendecomposition on the context-induced linear operators. More papers coming soon—stay tuned!

Runtian Zhai @RuntianZhai

4 months ago

Check out Runtian's thesis on contexture theory, which shows that many representation learning methods perform eigendecomposition on the context-induced linear operators. More papers coming soon—stay tuned!

2 30 164 22K 114

Heading to #ICLR2025! Looking forward to discussions on LLMs (for tabular data), interpretability, and representation learning. I'll be presenting my internship project on LLMs for tabular anomaly detection — catch our poster on Sat, April 26 at 10am! Come say hi! @iclr_conf

Are current reasoning models optimal for test-time scaling? 🌠 No! Models make the same incorrect guess over and over again. We show that you can fix this problem w/o any crazy tricks 💫 – just do weight ensembling (WiSE-FT) for big gains on math! 1/N

Excited to share new work from my internship @GoogleAI ! Curious as to how we should measure the similarity between examples in pretraining datasets? We study the role of similarity in pretraining 1.7B parameter language models on the Pile. arxiv: arxiv.org/abs/2502.02494 1/🧵

Model-free deep RL algorithms like NFSP, PSRO, ESCHER, & R-NaD are tailor-made for games with hidden information (e.g. poker). We performed the largest-ever comparison of these algorithms. We find that they do not outperform generic policy gradient methods, such as PPO. 1/N

keyboard_arrow_left Previous keyboard_arrow_right Next

To trust LLMs in deployment (e.g., agentic frameworks or for generating synthetic data), we should predict how well they will perform. Our paper shows that we can do this by simply asking black-box models multiple follow-up questions! w/ @m_finzi and @zicokolter 1/ 🧵

Through 2024, scaling test-time compute has become key. But, what does it mean to use test-time compute effectively & efficiently + how to do it? 🤔 We wrote a blog post ✍️ with a conceptual perspective on this: blog.ml.cmu.edu/2025/01/08/opt… 🎯Answer: meta reinforcement learning 🧵⤵️

Introducing Content-Adaptive Tokenizer (CAT) 🐈! An image tokenizer that adapts token count based on image complexity, offering flexible 8x, 16x, or 32x compression! Unlike fixed-length tokenizers, CAT optimizes both representation efficiency and quality. Importantly, we use just…

We have just released SenSet, a novel list of 106 senescence marker genes. We hope this resource accelerates discoveries in aging research, cancer biology, and regenerative medicine. #senescence #aging #pulearning #gene-set #SenNet biorxiv.org/content/10.110…