Cheng Xin

Cheng Xin

Postdoc Scholar @ Rutgers University [on Job Market].

About me

I am currently a Postdoctoral Researcher in the Department of Computer Science at Rutgers University, advised by Dr. Jie Gao. I design principled geometric and topological methods for trustworthy machine learning, with a focus on interpretability, robustness, and structural understanding in modern AI systems.

I earned my Ph.D. in Computer Science from Purdue University, advised by Dr. Tamal K. Dey. My doctoral research focused on topological data analysis (TDA) and the theory and algorithms of multiparameter persistent homology, with contributions to both foundational theory and efficient computation.

I am actively seeking faculty and research positions.

Research Overview

My research develops geometric and topological foundations for modern machine learning and AI.
While contemporary learning systems achieve impressive empirical performance, they often lack structural guarantees, interpretability, and robustness. My work addresses these limitations by integrating tools from algebraic topology, computational geometry, and metric geometry into learning frameworks that are both theoretically principled and practically effective.

A recurring theme in my research is structure preservation: designing representations and algorithms that respect the intrinsic geometry and topology of data, particularly in non-Euclidean, graph-structured, and scientific domains.

Current Research Directions

My recent and ongoing work includes:

  • Multiparameter Persistent Homology and Stability Theory
    Algorithmic foundations and distance computation for multiparameter persistence modules, with applications to structured data analysis.

  • Topologically-Enhanced and Interpretable Graph Learning
    Differentiable topological representations for graph neural networks, including principled frameworks for explainability and rationale discovery.

  • Non-Euclidean and Pseudo-Euclidean Representations
    Extensions of classical dimensionality reduction and embedding theory (e.g., MDS, Johnson–Lindenstrauss) beyond Euclidean settings.

  • Geometric and Topological Methods for AI for Science
    Applications to scientific machine learning, including biological networks, molecular and physical systems, and 3D / video understanding (e.g., DL3DV).

These projects aim to bridge theoretical foundations and real-world impact, enabling learning systems that are more robust, interpretable, and aligned with underlying data structure.

Research Philosophy

I view geometry and topology not as auxiliary tools, but as core organizing principles for learning systems.
By characterizing invariance, stability, and expressivity at a mathematical level, my work seeks to provide explanatory power beyond black-box performance metrics, particularly for high-stakes scientific and societal applications.

This website is built using the academicpages template and is periodically updated to reflect my current research and teaching activities.