Xiaotong (Denny) Cui

• Co-developing a benchmark that evaluates how well VLM-based agents ground visual perception into decisions and actions in complex, interactive video-game environments.
• Deployed the benchmark across multiple games using the PyBoy Game Boy emulator, enabling multimodal agents to be evaluated in diverse interactive environments.
• Collaborating with Ph.D. researchers on task design, agent evaluation, and experiments; manuscript under review at ICLR 2027.

• Proposed SCAD, a lightweight model using capsule-network modules to dynamically extract multiple interest representations per user from behavior sequences; code released open-source.
• Co-developed ACNNS, a unified multi-interest framework integrating neighbor-sequence enhancement with attention-based capsule networks.
• Thesis recognized as an Outstanding Undergraduate Thesis of Heilongjiang University.

• Developed DNN components that learn correlations in high-dimensional data to group similar objects, reducing the search cost of kNN and range queries.
• Introduced a multi-module DNN framework plus a linear-time data layout refinement algorithm that bounds candidate size, achieving superior time and space efficiency with high accuracy.

Courses: Deep Learning and Its Applications, Natural Language Processing, Analysis of Algorithms, Database Systems

Thesis: “SCAD: A Lightweight Recommendation Model Based on Multiple Interests” (Outstanding Undergraduate Thesis)

• Analyzed large-scale advertising datasets with Python, PyTorch, TensorFlow and MySQL to identify patterns that improve campaign performance and client ROI.
• Assisted in designing algorithms to improve ad targeting accuracy and user engagement; automated data preprocessing and reporting, and built real-time KPI dashboards.

• Designed and maintained server-side logic and RESTful APIs; worked with cross-functional teams to improve system stability.

• Built prototypes and proofs of concept, performed system testing and validation, and optimized database structures and queries.