Sequence Analysis and Modeling Lab · Rutgers University

Making multimodal AI reliable, efficient and agentic.

We build probabilistic and deep learning methods for images, video, language and time series: from vision-language models that don’t hallucinate, to agents that remember what they see, to one-step generative models.

Removing three image tokens eliminates hallucinated objects in an LVLM caption
Featured: removing three image tokens eliminates hallucinated objects in a vision-language model (ICCV 2025).

CVPR 2026

Prof. Pavlovic served as Program Chair of the largest computer vision conference

~300 papers

Peer-reviewed publications at NeurIPS, CVPR, ICCV, ECCV, ICLR and more

19 Ph.D. alumni

Graduates of the lab now in research and academia

What we work on

All research →

EAZY hallucination figure

Vision-language models

Finding where LVLMs hallucinate and fail to reason, and fixing it with lightweight, training-free interventions.

ECCV 2026 · ICCV 2025

MemEye benchmark overview

Agents with long-term memory

Benchmarks and self-optimizing memory policies for agents that live through hours of video and interaction.

LVMH · MemEye

Rectified MeanFlow figure

Efficient generative models

Few-step and one-step diffusion and flow models for images, language and trajectories.

NeurIPS 2026 · Re-MeanFlow

Visual puzzles generated by GenVP

Visual reasoning

Models that generate and solve visual puzzles with interpretable, structured reasoning.

ICLR 2025 · ICPR 2022 Best Student Paper

ECTraj trajectory predictions

Crowds and trajectories

Predicting how people and agents move through complex spaces, for safer buildings and cities.

ECCV 2026 · CVPR 2024

News

Nov 2026

Keynote at the AAAI Fall Symposium on Trustworthy Agentic Systems: “Seeing Is Not Remembering.”

Oct 2026

CVPR 2026 Compute Reporting findings announced by the IEEE Computer Society.

2026

Few-step diffusion language models accepted to NeurIPS 2026; two papers at ECCV 2026.

Sep 2026

Prof. Pavlovic returns to Rutgers after four years as an NSF Program Director.

All news →

The team

VP

Vladimir Pavlovic

AM

Alen Mrdovic

TL

Tony Qingze Liu

XZ

Xinxi Zhang

LC

Liwei (Levi) Che

SK

Sooyoung Kim

Meet the lab and our alumni →

Join SEQAM Lab

Ph.D. applicants: apply to the Rutgers Computer Science Ph.D. program and name SEQAM Lab in your statement. Read our guide first: what we look for, open directions and how to reach us.