Feng Liu

Papers

Total Citations

H-Index

About

Feng Liu is a prominent researcher working at the intersection of computer vision and natural language processing, with a particular focus on image captioning and visual intelligence systems. His most recognized contribution, "Actor-Critic Sequence Training for Image Captioning" (2017), addresses a fundamental challenge in AI: enabling machines to generate meaningful natural language descriptions of visual scenes — a critical capability for robots and AI agents that must communicate with human users about their visual environment. In this influential work, Liu tackled the notorious exposure bias problem in sequence generation by applying actor-critic reinforcement learning methods, moving beyond conventional likelihood maximization to optimize directly for evaluation metrics. The paper has accumulated 99 citations, reflecting its meaningful impact on the research community and its relevance to both the computer vision and natural language generation fields. Liu's work sits at a fascinating and practically important frontier — bridging perception and communication in AI systems — with clear real-world implications for assistive robotics, autonomous agents, and human-computer interaction. His contributions help lay the groundwork for AI systems capable of not just seeing the world, but meaningfully describing it.