Export
Yifeng Huang
· 2 previous affiliations
- Stony Brook UniversitySince 2021
- PhD student · Since 2021
- Undergrad student · 2015–2020
Research Overview
Yifeng Huang focuses on graph neural networks, human-AI interaction, and applied machine learning for complex data structures. Early work explored deep reinforcement learning for molecular graph generation, while subsequent research addressed structural feature co-embedding in graph pooling and crowd-assisted computer vision for disaster response. The most recent available evidence indicates an interest in integrating algorithmic efficiency with human-in-the-loop systems.
Research Trajectory
The available evidence for Yifeng Huang's research trajectory spans a concentrated period between 2019 and 2020, highlighting contributions to graph representation learning and applied artificial intelligence. Because the documented publication record is sparse, this trajectory reflects a snapshot of collaborative projects rather than a decades-long independent research program.
In 2019, Huang participated as a middle author in research targeting computational drug discovery. This work, detailed in Molecular Graph Generation with Deep Reinforced Multitask Network and Adversarial Imitation Learning, addressed the challenges of delayed rewards and mode collapse in generating discrete molecular graphs. The project proposed a framework combining deep reinforcement learning with adversarial imitation and multitask learning to optimize multiple biochemical properties simultaneously without conflicting objectives.
By 2020, Huang's focus within graph-based machine learning deepened, taking a more prominent second-author role in methodological improvements for graph neural networks (GNNs). A major open problem in GNN architecture is graph pooling, which often relies on empirical heuristics rather than rigorous topological guidance. In Topology Discriminative Pooling via Graph Kernel Derived Nodes and Edges Co-Embedding, Huang co-developed a topology discriminative pooling (TD-Pool) approach. This method utilized graph kernels—specifically integrating the Weisfeiler-Lehman graph kernel with neural networks—to capture and co-embed both semantic and structural features across nodes and edges, thereby improving graph coarsening.
Concurrently in 2020, Huang applied machine learning techniques to high-stakes, real-world computer vision tasks, specifically in the domain of emergency response. As a second author on Crowd-Assisted Disaster Scene Assessment with Human-AI Interactive Attention, Huang contributed to systems that evaluate damage severity from social media imagery. Recognizing that fully automated AI algorithms often fall short of the accuracy required for prioritizing rescue operations, this research investigated human-AI interactive attention mechanisms. By leveraging crowdsourcing platforms, the work sought to bridge the gap between algorithmic efficiency and human perceptual reliability in disaster scene assessment.
Current Directions
Based on the provided corpus, the most recent papers in the evidence are from 2020. These publications suggest a dual interest in rigorous structural machine learning and applied, human-centered AI systems. On the theoretical side, the work addresses fundamental architectural bottlenecks in graph neural networks, particularly the integration of topological information into graph pooling mechanisms. On the applied side, the research investigates how human-in-the-loop frameworks—such as interactive attention models and crowdsourcing—can augment computer vision models in complex, low-resource environments like disaster response.
Profile updated
Loading new papers…
Loading research neighbors…
See something wrong? Report a mistake