Zhakshylyk Nurlanov

News & Highlights

• [Seeking Position] I am actively looking for full-time research scientist/engineer positions starting May 2026. My expertise spans AI security and safety (content protection and provenance), robustness, computer vision, and large language models. Please reach out if you have opportunities!
• [2025.01] New preprint on Effective Transferable Attacks via Random Local Search - demonstrating that jailbreaking LLMs does not require any prior information nor gradients.
• [2024.09] Our Adaptive Certified Training accepted at ECML PKDD 2024! Achieving better accuracy and certified robustness tradeoffs.
• [2023.02] Our Universe Points Representation Learning for Partial Multi-Graph Matching accepted at AAAI 2023! SOTA semantic keypoint matching by building on graph neural networks (3D-SplineCNN) and on the object-to-universe formulation.
• [2022.06] Our Sublabel-Accurate Energy Minimization paper accepted at ICPR 2022! Extending classic discrete alpha-expansion algorithm to sublabel-accurate energy minimization.
• [2021.12] Graduated with MSc in Computer Science from TUM, Germany! Thesis: Deep Learning for Multi‑Graph Matching (Grade: 1.0), supervised by Prof. Florian Bernard.

About

I am a PhD student specializing in AI Security and Safety, with focus on adversarial robustness, content protection and provenance, computer vision, and large language models. My research develops reliable AI systems through problem-specific optimization methods and robust training techniques. I have contributed to cutting-edge projects at Bosch, TUM, and Samsung, with publications in top-tier venues.

I am currently based in Berlin, Germany, and actively seeking full-time positions in AI Security & Safety, Computer Vision, and Large Language Models.

Research Interests

• AI Security and Safety: adversarial robustness, verifiable AI systems, content protection and provenance, jailbreaking attacks and defenses
• Computer Vision: correspondence problems, 3D vision, optimization methods, visual odometry & SLAM, robust vision systems
• Large Language Models: security, safety, and alignment of LLMs

Academic Service

Conference Reviewer: NeurIPS'25, CVPR'25, ICLR'25, ECCV'24, ICML'23, ICCV'23, GCPR'23

Journal Reviewer: IEEE TNNLS, IEEE TPAMI