CV

Ph.D. Candidate in Streaming Video Understanding

Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany

Advisor: Prof. Dr. Juergen Beyerer and Prof. Dr. Juergen Gall

M.Sc. in Mechatronics and Robotics

Leibniz University Hannover, Hannover, Germany

M.Sc. in Mechatronics and Robotics (with distinction) — Final Grade: 1.2 (Excellent, German Scale)

B.Eng. in Process Engineering and Energy Technology

University of Applied Sciences and Arts Hannover, Hannover, Germany

B.Eng. in Process Engineering and Energy Technology — Final Grade: 1.7 (Good, German Scale)

Research Assistant

IES, Karlsruhe Institute of Technology (KIT) & Fraunhofer IOSB, Karlsruhe, Germany

• Proposed E²-TTT to efficiently learn long context beyond the fixed-window constraints of transformer models, sustaining >90% passkey-retrieval accuracy at 8× the training context.
• Developed FlowNar, an efficient streaming video narration framework that dynamically manages the visual context, supporting 10× longer videos at 3× higher throughput than prior online LMMs.
• Designed several novel action anticipation architectures achieving superior performance on large-scale benchmarks, including Epic-Kitchens and Ego4D.
• Authored a comprehensive survey of 90+ deep learning action anticipation methods, introducing a taxonomy by contribution and consolidating datasets, metrics, and future directions.
• KIMoS project: led grant acquisition (€180K) for real-time assembly assistance, securing three industrial partners, and represented Fraunhofer IOSB at Hannover Messe 2026.
• JuBot project: built and deployed a multimodal action recognition system that identifies multiple people's actions in real time on the KIT ARMAR humanoid robot.

Master Thesis & Research Assistant

IPI, Leibniz University Hannover, Hannover, Germany

• Designed probabilistic models for uncertainty estimation in dense stereo matching (grade 1.0).

Research Intern

Volkswagen AG, Wolfsburg, Germany

Research on adversarial attacks and robustness for semantic segmentation.

Development Intern & Bachelor Thesis

Robert Bosch GmbH, Schwieberdingen, Germany

Simulation of gasoline engine exhaust systems.

Rethinking Expressivity and Efficiency in Test-Time Training

Under Review

Z. Zhong, et al.

FlowNar: Scalable Streaming Narration for Long-Form Videos

ICML 2026

Z. Zhong, et al.

Scalable Video Action Anticipation with Cross Linear Attentive Memory

WACV 2026

Z. Zhong, et al.

A Survey on Deep Learning Techniques for Action Anticipation

TPAMI (Under Review)

Z. Zhong, et al.

Anticipative Feature Fusion Transformer for Multi-Modal Action Anticipation

WACV 2023

Z. Zhong*, D. Schneider*, et al.

2nd Place (out of 15 teams), Ego4D Long-Term Action Anticipation Challenge @ CVPR 2024

CVPR 2024

6th Place (out of 30 teams), KIT Neuland Innovation Contest

Karlsruhe Institute of Technology (KIT)