Sitong Zhang

Sitong Zhang

postdoc · aalto university · espoo, finland

Hi — I’m Sitong Zhang, a Postdoctoral Researcher at Aalto University, Department of Computer Science, working with Prof. Bo Zhao.

Before Aalto I was a Postdoctoral Researcher at CityU-Oxford Joint CIMDA, City University of Hong Kong, working with Prof. Hong Yan (IEEE Fellow). I received my PhD (2023) and BEng (2018) from Harbin Engineering University, advised by Prof. Yibing Li.

Currently building
HarnessKit — A control plane for your AI coding agents — see, secure, and manage every extension and config from one place. ↗ ★—
Paper Viewer — A self-hosted paper workspace for research reading groups — daily arXiv digest with AI intros, shared library, PDF annotation, and per-paper AI chat. ↗
csconf-papers — Accepted-paper lists for eight top systems and database conferences, rebuilt from DBLP on the first of every month. ↗

Research Interests

Deep Reinforcement Learning MLSys LLM Infra

My research lies at the intersection of machine learning and systems. I currently focus on infrastructure for distributed reinforcement learning and LLM post-training, with particular interest in the runtime designs that make these workloads fast, adaptive, and cost-aware at cluster scale.

This builds on my doctoral work in reinforcement learning itself, where I developed deep reinforcement learning methods for UAV autonomous navigation. After years inside the training loop, I now work on the systems that run it at scale.

RL Training Stack
Algorithm 2018–2024
policy design, learning objectives
Runtime 2025–
▸ scheduling when
▸ placement where
▸ orchestration how
hover to explore
Hover a layer to see what I work on there.
past work · 2018–2024
Deep reinforcement learning for UAV autonomous navigation — obstacle avoidance, long-distance trajectory planning, and human-in-the-loop motion planning.
current focus · 2025–
Runtime systems for distributed RL and LLM post-training. The right balance of compute, memory, and interconnect shifts continuously with model scale, workload characteristics, and hardware conditions. Scheduling, placement, and orchestration are the runtime levers that keep the system responsive to these dynamics at cluster scale.

Selected Publications

view all publications on Google Scholar →
    2. 2026
      Conduit: An Experience Data Plane for Distributed Reinforcement Learning
      Sitong Zhang, Tuo Shi, Mario Di Francesco, Zeke Wang, Bo Zhao
      arXiv preprint arXiv:2609.24456[html][bibtex]
    2. 2026
      Nereus: Adaptive Parallelism for LLM Post-Training
      Songlin Jiang, Tuo Shi, Sitong Zhang, Zeke Wang, Mario Di Francesco, Bo Zhao
      arXiv preprint arXiv:2609.34645[html][bibtex]
    2. 2022
      Autonomous navigation of UAV in multi-obstacle environments based on a deep reinforcement learning approach
      Sitong Zhang, Yibing Li, Qianhui Dong
      Applied Soft Computing[html][code][video][bibtex]
    2. 2023
      A hybrid human-in-the-loop deep reinforcement learning method for UAV motion planning for long trajectories with unpredictable obstacles
      Sitong Zhang, Yibing Li, Fang Ye, Xiaoyu Geng, Zitao Zhou, Tuo Shi
      Drones[html][code][bibtex]