Ruoshi Liu

Papers

Total Citations

H-Index

About

Ruoshi Liu is an emerging researcher at the intersection of computer vision, machine learning, and robotics, with work spanning generative modeling, neural network optimization, and autonomous systems. His research tackles some of the field's most challenging inverse problems — situations where a desired output must be traced back to its underlying cause — with a particular focus on making these computations more tractable and reliable. Among his most recognized contributions is work on extreme monocular dynamic novel view synthesis, which explores how cameras can simulate complex cinematic movements from minimal input data, garnering 9 citations since its 2024 publication. His "Landscape Learning" framework addresses a fundamental bottleneck in neural network inversion: the treacherous non-convex loss landscapes that plague gradient descent optimization, offering a more principled path through inference-time challenges. Liu has also pushed boundaries in physical robotics, developing AquaBot, an autonomous underwater manipulation system that learns to improve itself despite the formidable complexities of fluid dynamics and unstructured marine environments. His creative project "Controlling the World by Sleight of Hand" further demonstrates a broad intellectual curiosity about human-machine interaction. Though early in his career, Liu's cross-disciplinary ambition marks him as a researcher worth following closely.