avatar

Junyan Cheng

Ph.D. Candidate
Dartmouth College
jc.th@dartmouth.edu



I am a 4th-year Ph.D. candidate at Dartmouth College, advised by Prof. Peter Chin; and a fully funded PhD Innovation Fellow (NSF Research Traineeship, Thayer Engineering & Tuck Business collaboration), supervised by Prof. Eric R. Fossum on research translation and entrepreneurship. I hold an M.S. in CS (Intelligent Robotics) from USC. I interned at Microsoft Research AIIL group and the Allen Institute for AI (Ai2) Aristo team. I’m interning at Google Research this summer.

About My Research

Can we build a general thinking machine that really creates knowledge beyond imitation in the real world? Towards this goal, I study the theoretical feasibility from representation level (ICLR’24 “TDL”). The design of “thinking” agents follows AIMA definition, environment modeling and framework (ICLR’24 Spotlight “SocioDojo”), lifelong memory (FinAI@ICLR’25 “AAPM”), adaptation and personalization (ACL’26 Findings “Apeiron), and reasoning (ICLR’26 “Analytica”). Finally, the evolution and discovery (NeurIPS’25 Spotlight “Genesys”). Read more in my blog “RCM”.

Autonomous Agentic Systems

I developed 🧊LLLM, an modular, minimalistic, and high-performance open-source framework for building agentic systems. Check live demos of my flagship systems below (click to explore, swipe left-right). In particular, I built an 🧠Analytica web app for LLM-agentic analysis (may slow in first load).

Analytica

Robust and efficient wide and deep LLM-agentic analysis and forecasting with soft logic and parallel data analysts.

ICLR'26

Genesys

Distributed discovery system for novel LM architectures with autonomous design, pre-training, and evaluation.

NeurIPS'25 Spotlight

Apeiron

Large-scale full-lifecycle autonomous app synthesizer, optimizing user demands with multi-threaded CUA evaluators.

ACL'26 Findings

SocioDojo

LLM agentic lifelong societal modeling and analysis with news feed and real-time data flow. A "Human World Model".

ICLR'24 Spotlight

Selected Publications

  1. ICLR
    Analytica: Soft Propositional Reasoning for Robust and Scalable LLM-Driven Analysis
    Junyan Cheng, Kyle Richardson, Peter Chin
    The Fourteenth International Conference on Learning Representations (ICLR), 2026.
    PDF Code Demo

  2. ACL Findings
    Apeiron: A Scalable LLM-agentic Framework for Autonomous Full-lifecycle Demand-optimized Application Synthesis
    Junyan Cheng, Ankit Srivastava, Jessie Zeng, Milenko Drinic, Jack W. Stokes
    Findings of The 64th Annual Meeting of the Association for Computational Linguistics (ACL), 2026.
    PDF Demo

  3. NeurIPS
    Language Modeling by Language Models
    Junyan Cheng, Peter Clark, Kyle Richardson
    The Thirty-ninth Annual Conference on Neural Information Processing Systems (NeurIPS), 2025.
    PDF Code Demo Spotlight (3.19% acceptance rate)

  4. FinAI@ICLR
    Empirical Asset Pricing with Large Language Model Agents
    Junyan Cheng, Peter Chin
    ICLR 2025 Advances in Financial AI Workshop (Long Paper track), 2025.
    PDF Code

  5. ICLR
    SocioDojo: Building Lifelong Analytical Agents with Real-world Text and Time Series
    Junyan Cheng, Peter Chin
    The Twelfth International Conference on Learning Representations (ICLR), 2024.
    PDF Code Demo Spotlight (5.02% acceptance rate)

  6. ICLR
    Bridging Neural and Symbolic Representations with Transitional Dictionary Learning
    Junyan Cheng, Peter Chin
    The Twelfth International Conference on Learning Representations (ICLR), 2024.
    PDF Code

  7. EMNLP
    Multimodal Phased Transformer for Sentiment Analysis
    Junyan Cheng*, Iordanis Fostiropoulos*, Barry Boehm, Mohammad Soleymani
    Conference on Empirical Methods in Natural Language Processing (EMNLP), 2021.
    PDF Code

More Publications

  1. JCISE
    Physics Informed Synthetic Image Generation for Deep Learning-Based Detection of Wrinkles and Folds
    Omey M Manyar, Junyan Cheng, Reuben Levine, Vihan Krishnan, Jernej Barbič, Satyandra K Gupta
    Journal of Computing and Information Science in Engineering
    PDF Code Best Paper Award at ASME IDETC'22

  2. arXiv
    Graph Conditioned Sparse-Attention for Improved Source Code Understanding
    Junyan Cheng, Iordanis Fostiropoulos, Barry Boehm
    arXiv preprint, arXiv:2112.00663
    PDF Code

  3. arXiv
    GN-Transformer: Fusing Sequence and Graph Representation for Improved Code Summarization
    Junyan Cheng, Iordanis Fostiropoulos, Barry Boehm
    arXiv preprint, arXiv:2111.08874
    PDF Code

  4. RNA biology
    Predicting microRNA-disease associations using bipartite local models and hubness-aware regression
    Xing Chen*, Junyan Cheng*, Jun Yin
    RNA biology, 2018, VOL. 15, NO. 9, 1192-1205
    PDF Code