Chenyou Guo

Academic Homepage

Chenyou Guo 郭宸佑

Undergraduate Student, Ocean University of China

I am a first-year undergraduate student in Computational Physics at Ocean University of China. My research interests include domain-specific large language models, AI for Science, computer vision, high-performance computing, and research infrastructure.

Email OpenReview Qingdao, Shandong Panzhihua, Sichuan
About

Research profile

I am interested in building intelligent systems that can reason over scientific domains. My current work focuses on domain adaptation, expert-level evaluation, and robust multimodal reasoning. I also enjoy the engineering side of research, including Linux environments, multi-GPU scheduling, reproducible experiments, and clean technical writing.

Domain-Specific LLMs AI for Science Computer Vision HPC AI Infrastructure
News

News and highlights

July 2026 SIGHT: From Imagination to Evidence via Active Spatial Hypothesis Testing was submitted to ICLR 2027.
Jun 2026 Multi-view Semantic-Guided Domain Adaptive Semantic Segmentation through Cross-modality Motif Matching was accepted by Pattern Recognition.
Jun 2026 The patent “A Sea Surface Temperature Completion Method and System Based on Spatiotemporal Priors and Flow Matching Generative Models” was accepted.
Jun 2026 Geo-Expert was accepted by the ICML 2026 AI for Science Workshop.
May 2026 Motif Matching with Structure-Semantic Signatures was submitted to NeurIPS 2026.
Research

Selected research and projects

SIGHT

SIGHT: From Imagination to Evidence via Active Spatial Hypothesis Testing

First Author · ICLR 2027

We introduce SIGHT (Spatial Inference through Generative Hypothesis Testing), a training-free test-time framework that treats world-model-assisted spatial reasoning as active hypothesis testing. SIGHT converts answer choices into falsifiable hypotheses and selects camera probes predicted to distinguish them.

Under Review
Geo-Expert

Geo-Expert: Expert-Level Geological Reasoning via Parameter-Efficient Fine-Tuning

First Author · ICML 2026 AI for Science Workshop

We present Geo-Expert, a parameter-efficient geological LLM framework, and establish Geo-Eval, a benchmark for geological reasoning.

Accepted
Motif Matching

Motif Matching with Structure-Semantic Signatures for Robust CLIP Test-Time Adaptation

Second Author (First Student) · NeurIPS 2026

Motif-based reasoning enhances test-time adaptation by integrating semantic and structural information for robust out-of-distribution and cross-domain generalization.

Under Review
Pattern Recognition

Multi-view Semantic-Guided Domain Adaptive Semantic Segmentation through Cross-modality Motif Matching

Third Author · Pattern Recognition

This work studies semantic-guided domain adaptive semantic segmentation through cross-modality motif matching.

Accepted
Education

Academic training

Ocean University of China

Sep 2025 - Present

B.S. in Computational Physics · First-year undergraduate study

GPA: 3.937 / 4.00 · Average score: 95.480 / 100 · Rank: 3

Experience

Research experience

GeoWorldModel

Jun 2026 - Present

Core Leader

Brings together Quaternary fossil records, modern biodiversity observations, present-day climate, and paleoclimate simulations. The goal is a unified representation of taxa, time, space, and climate for life–Earth system research.

SIGHT Project

May 2026 - Present

Core Leader

SIGHT turns generated views into evidence by actively testing spatial hypotheses and revising answers only with pose-grounded counterevidence.

Geo-Expert Project

Nov 2025 - May 2026

Core Leader

Led a domain-specific LLM project for expert-level geological reasoning, including framework design, benchmark construction, and experimental iteration.

Motif Matching Project

Dec 2025 - May 2026

Student Member

Contributed to a robust CLIP test-time adaptation method using structure-semantic signatures for out-of-distribution and cross-domain generalization.

Skills

Technical skills

  • Programming and frameworks: Python, PyTorch, MATLAB
  • Engineering: Linux server deployment, multi-GPU computing resource scheduling
  • Research tools: LaTeX, Microsoft Office, benchmark design, technical writing
  • Workflow: Vibe Coding Workflow, collaborative documentation, experiment organization