Zheng Chen

Excited to share two milestones in our EEG research 🏄‍♂️—TFM-Token and ODEBrain have been accepted to ICLR 2026! Looking forward to pushing this direction further with the team and collaborators. #ICLR2026

Meta reviewer: 🧐all reviewers keep their negative original scores. (Pic. 1) All reviewers: 🤔Didn’t I raise my scores at all? (Pic. 2-4) openreview.net/forum?id=gr8tc… @iclr_conf @BharathHarihar3 @LerrelPinto @Diyi_Yang #ICLR26

Project Page: chenzrg.github.io/project/mlomics ⚙️ Key Features: • 🧬 8,314 multi-omics TCGA samples across 32 cancers • 🧪 ML-ready datasets covering 20 learning tasks • 🧠 Benchmarking reproducible baseline models • 🔄 Multiple feature scales • 📈 Diverse biological evaluation

Beyond models, data and AI infrastructure are equally important for understanding how far AI is from biomedical practice. This resonates with our MLOmics (Nature Scientific Data), a ML-ready multi-omics dataset to enable reproducible evaluation of AI models in cancer research.

Can we model brain activity as a time-evolving network, rather than a static graph? EvoBrain answers this by grounding dynamic EEG graph modeling in theory and validating it on real intracranial EEG seizure prediction. Spotlight @NeurIPS2025 🎉 Github: github.com/Kotoge/EvoBrain

@44nb_k Hi Prof, this is a very nice paper. It touches on frequency bias, which is closely related to our KDD 2024 work(Fredformer), where we formally defined the frequency learning issue of Transformers in TSF. Glad to see further discussion of this line of work in future studies.

Our paper "Dualformer: Time-Frequency Dual Domain Learning for Long-term Time Series Forecasting" (first author: Jingjing Bai (UOsaka)) has been accepted at #AISTATS2026 (arXiv: arxiv.org/abs/2601.15669). We tackle Transformers' low-pass bias in long-term time series forecasting via a time/frequency dual-branch model with layer-wise band allocation.

Are you curious about how SE researchers apply Reinforcement Learning (RL) in their SE tasks? We recently conducted a survey of RL4SE, including the trends, research topics, used RL algorithms, model design and optimization, and more! Please check it! arxiv.org/abs/2507.12483

Welcome to our presentation and poster! Oral Session 2E Thu 24 Apr 3:42-3:54 pm Poster Hall 3 + Hall 2B #16 Thu 24 Apr 10 - 12:30 am.

SODor was accepted as an Oral at #AAAI2025. 🎉

🚀Thrilled to share that our work "SODor: Long-Term EEG Partitioning for Seizure Onset Detection" has been accepted at AAAI 2025! arxiv.org/abs/2412.15598

Data&Code: github.com/chenzRG/GeSubN… Paper: arxiv.org/abs/2410.13178

Can we integrate biological knowledge bases, experimental data, and AI models to unravel gene-disease associations?🤔 Excited to share that our paper “GeSubNet: Gene Interaction Inference for Disease Subtype Network Generation”has been accepted for an Oral at #ICLR2025 🚀🚀

arxiv.org/abs/2410.11200 This paper introduces a self-supervised learning method with a splittable architecture tailored for single-channel EEG, which holds potential for portable EEG monitoring and cross-channel/device deployment.

My Ph.D. student, Rikuto, will be at #ICDM2024 to present his first paper "SplitSEE: A Splittable Self-supervised Framework for Single-Channel EEG Representation Learning" on Day 2, December 10, at 14:45 (DM385). Feel free to chat about anything!

Rule-based rewards (RBRs) use model to provide RL signals based on a set of safety rubrics, making it easier to adapt to changing safety policies wo/ heavy dependency on human data. It also enables us to look at safety and capability in a more unified lens as a more capable grader model gives us higher quality RL signals.

Here is our paper: arxiv.org/abs/2406.09009

Congratulations to my first Ph.D. student, Xihao, for having his first paper, Fredformer, accepted by the #KDD2024 Research Track! We provide empirical analyses of frequency bias in Transformer-based time series forecasting and propose a feasible solution to solve this problem.

Checkout a new ICML paper from our group led by @ori_press! 🚨 We show why entropy minimization improves the accuracy of classifiers on input data, and how it could be used to estimate the accuracy of a classifier on an arbitrary dataset, without labels. Detailed🧵👇

科研費若手研究が採択されました！ Thrilled to receive my first research grant from JSPS!

I finished the manuscript of the funding application. This is the first time I can complete an application (or conference paper) before one week of DDL.