Coverage matters: MarginMerge for compressing multi-vector visual document retrievers
arXiv:2608.02969
AI Scientist / Research Engineer
Versed in language models, information retrieval, computer vision, and neural graphics.

I’m a Machine Learning Research Engineer based in Vancouver, with nearly five years of professional experience.
At Global Relay, I develop financial AI systems, focusing on knowledge distillation, quantization, and scalable inference. Previously at Huawei, I led neural texture compression and built segmentation and real-time AI rendering pipelines.
My research areas include LLMs, decision models, multi-modal embedding and retrieval, computer vision, and neural graphics. I hold an M.A.Sc. from the University of British Columbia.
arXiv:2608.02969
ICML 2026 Workshop on Resource-Adaptive Foundation Model Inference
SIGGRAPH Asia 2023 Conference Papers, 1–9
2021 IEEE International Conference on Systems, Man, and Cybernetics (SMC)
University of British Columbia
Large language models (LLMs), decision models, and efficient embedding models for financial AI.
Neural graphics, computer vision, and optimized inference for real-time rendering.
Instance segmentation, synthetic data, and perception for robotics.
Graduate research in 3D reconstruction through object motion tracking.
For research, engineering roles, or project inquiries.
aria.salari@hotmail.com ↗