Multimodal perception & intelligent agents

Anna Min 闵安娜

Hi! I received B.E. at Tsinghua University, where I study computer science and statistics.

Email: anna.min1754@gmail.com (if you prefer an edu email, annamin@csail.mit.edu)

Recent Updates

Feb. 2025: One paper got accepted to CVPR 2025. See you in Nashville TN. Feel free to reach out!
Sep. 2024: I gave a talk at MIT ML Tea Time titled "Multi-sensory Perception from Top to Down." [link]
Feb. 2024: The feature for multi-modal generation based on the algorithm and demo that I designed gained over 246k views on Twitter during my internship at Pika Labs .

Research

I study intelligent agents through multimodal perception, generative models, and world modeling. My previous work explores how different modalities, including vision, audio, and language, interact and emerge in intelligent systems.

Selected Publications

SEDM: Selective Self-Evolving Distributed Memory for Agents

Yichen Liu*, Anna Min*, Yinzhi Wang, Yuhang Yao, Yichen Han, Rui Zhang, Jiantao He, Zhiheng Li, Yingying Qi, Haoran Xu, Zhimin Ding, Yuxin Tang, Qi You, Lynn Ai, Eric Yang, Lewei He, Tianyu Shi

MERA: Model Evolution and Routing with Skill Adaptation for Agentic Systems at Scale

Yuhang Yao*, Anna Min*, Tianyu Shi

In submission to ICLR 2027

See, Infer, Intervene: Proactive World Modeling for Goal-Oriented Social Intelligence

Anna Min*, Honghui Zhang*, Chenmeinian Guo, Yichen Yu, Guanyu Liu, Yujia Zhang, Yongming Qin, Chongguo Song, Mengyue Yang, Lei Yu, Tianyu Shi

NeurIPS@VLM4RWD / In submission to ICLR 2027

Supervising Sound Localization by In-the-wild Egomotion

Supervising Sound Localization Using In-the-wild Ego-motion

Anna Min, Ziyang Chen, Hang Zhao, Andrew Owens

CVPR 2025 (Highlight -- 2.98% accept rate) / [Paper]

Learn spatial sound sources in the wild using ego-motion signals derived from visual cues with limited perspectives.

A Unit-based System and Dataset for Expressive Direct Speech-to-Speech Translation

Anna Min*, Chenxu Hu* , Yi Ren , Hang Zhao

Interspeech 2024 / [Paper]

Propose a dataset and pipeline with aligned bilingual audio tracks sharing similar emotions without using text as an intermediate for lesser-spoken languages and dialects.

Projects

Pika 1.5

Anna Min (Core Contributor)

Generative Multi-modal Model

Core contributor to Pika 1.5, a generative video model enabling text-to-video creation and video editing.

Other Works

Quantifying Geometrical Associations Across Multi-modal Perception: An Information-theoretic Perspective

Anna Min, Hang Zhao

NeurIPS 2024 Wi@ML Workshop

Study geometric relationships across multimodal representations from an information-theoretic perspective.

When End-to-End is Overkill: Rethinking Cascaded Speech-to-Text Translation

Anna Min, Chenxu Hu , Yi Ren , Hang Zhao

Preprint

Revisit cascaded speech translation systems and analyze when end-to-end approaches are unnecessary.

Selected Awards

2025, Bright Lite Scholarship, Finalist
2024, National Science Fund for Distinguished Undergraduate Researchers (130 undergraduate STEM researchers selected from across China, $50000), National Natural Science Foundation of China (NSFC)
2023, Tsinghua University Academic Excellence Award (2/103)
2022-2024, Tsinghua University Research Excellence Award (2/103)
2022, Spark Innovative Talent Cultivation Program (top 1%, 50/3900 undergraduates in Tsinghua for research performance)
2021, Meritorious Winner of Mathematical Contest in Modeling (top 0.18%, ranked 89th/49424)

Service

Reviewer: ICASSP 2025, CVPR 2025, NeurIPS 2024
Support underrepresented students: Volunteered with the Program Buddy Group at Tsinghua University, working with underrepresented students interested in coding, 2022

Misc: Art Creation

In the past I designed and developed video games (eg: A Player vs AI Strategy Game ). I play the piano and make music as well. My Chinese name is pronounced as Ān Nà, which is similar to Anna.