RandSlot: Learning Compact Visual Document Representations with Random Soft Tokens
arXiv:2609.32699 · 2026 · Preprint [Paper] [PDF]
Compact multi-vector representations for efficient visual document retrieval.
张凌霄
Ant Group | Tsinghua University
Research interests
LLMs & Agents · Search & Recommendation · Multimodal Retrieval
Google Scholar: Lingxiao Zhang
GitHub: OwenMasculinity
Contact:
owenmasculinity@gmail.com
I am an algorithm engineer at Ant Group, working on large language model agents, search and recommendation, and multimodal retrieval. My work connects machine learning research with production applications, with a focus on understanding user intent and making intelligent systems more effective and reliable in real business settings.
My current work focuses on core algorithms for agents and search, spanning complex task execution, memory and tool use, as well as retrieval and ranking. My interests center on bringing the reasoning and representation capabilities of large models into interactive systems, while balancing task quality, computational efficiency, and the demands of production deployment.
I received my master’s degree in Computer Science and Technology from Tsinghua University in 2020 and my bachelor’s degree in Electronic Information Engineering from Wuhan University in 2018. Before joining Ant Group, I worked as an algorithm engineer at Huawei and Alibaba, focusing on search advertising and intelligent recruitment.
Listed in reverse chronological order.
arXiv:2609.32699 · 2026 · Preprint [Paper] [PDF]
Compact multi-vector representations for efficient visual document retrieval.
arXiv:2607.13420 · 2026 · Preprint [Paper] [PDF]
Disentangling content interest from access habits to reduce recommendation bias.
ICONIP 2020 · pp. 491–503 · 2020 · Conference paper [Paper] [PDF]
Combining local intent and global preference for sequential recommendation.
For research discussions and collaboration, please get in touch by email.
owenmasculinity@gmail.com