Ruotong Zhao

4Machine learning papers tracked by Lacuna

1 other author named Ruotong Zhao: Ruotong Zhao (2 papers)

Is this you?

Research Overview

Ruotong Zhao focuses on generative agent simulations, automated scientific discovery, and the evaluation of large language models. His work explores how multi-agent systems can model complex social phenomena and accelerate research workflows through human-AI collaboration. Currently, he investigates robust evaluation frameworks for AI-generated literature reviews and automated model discovery systems.

Research Trajectory

Zhao’s research trajectory began with a focus on integrating large language models into complex, multi-agent environments to simulate both scientific and social ecosystems. In 2025, he contributed to OmniScientist: Toward a Co-evolving Ecosystem of Human and AI Scientists, a framework designed to shift artificial intelligence agents from isolated task executors to active participants in scientific research. This work modeled the social infrastructure of science, including peer review and citation networks, to facilitate cumulative discovery. This project intersected with the research direction Complementary Strengths in Human AI Diagnostic Systems and the broader area of Verbosity Bias in Automated Language Model Evaluation.

That same year, he collaborated on CrimeMind: Simulating Urban Crime with Multi-Modal LLM Agents. This project integrated large language models with agent-based modeling to simulate urban crime patterns. By embedding Routine Activity Theory—a classic criminological framework—directly into the agents' cognitive architecture, the system improved hotspot prediction and enabled realistic policy evaluation. This research advanced several fields , including the research direction Generative Agent Based Modeling with Large Language Models, the area of Scaling Generative Agent Simulations, and the broader area of Cognitive Architectures for Goal-Directed Artificial Agents. It also touched upon the direction Forecasting Beyond Historical Patterns with Event Driven Reasoning.

Moving into 2026, Zhao's focus on automating research workflows continued with his participation in the development of AutoSOTA: An End-to-End Automated Research System for State-of-the-Art AI Model Discovery. This system was designed to automate the replication, debugging, and optimization phases of AI research, aiming to free human researchers to focus on conceptual innovation rather than laborious implementation details.

Building on the need to rigorously assess these automated research tools, he subsequently introduced LitReview Arena: Evaluating Literature Review Agents with Battle-style Peer Review Platform. This evaluation framework utilized expert pairwise comparisons to measure the quality of AI-generated literature reviews. The study highlighted that while language models excel at retrieval, they struggle with higher-order synthesis compared to human experts. To address this gap, the work introduced a calibrated automated evaluator to better align with human expert standards.

Current Directions

The most recent papers in the evidence, published in 2026, indicate a strong focus on the automation and rigorous evaluation of scientific research workflows. Recent work investigates how end-to-end systems can autonomously discover state-of-the-art models and how the outputs of these systems—such as literature reviews—can be systematically evaluated. A central theme in this current research is bridging the gap between AI retrieval capabilities and the higher-order synthesis required for true scientific discovery, ensuring that automated agents can reliably augment human researchers in complex academic tasks.

Profile updated

See something wrong? Report a mistake