Adriano Koshiyama

Researcher, Department of Computer Science, University College London, University of London

Joined

October 2023

Names

Adriano Koshiyama (Preferred)

Emails

****@ucl.ac.uk (Confirmed)

Personal Links

Google Scholar

Career & Education History

Researcher

Department of Computer Science, University College London, University of London (cs.ucl.ac.uk)

2023 – Present

Advisors, Relations & Conflicts

No relations added

Expertise

No areas of expertise listed

Publications

Probing Implicit Bias Risk Framing in Language Models
Rishi Kalra, Andrea Dhelpra, Seonglae Cho, Adriano Koshiyama
- AFAA 2026 Poster
- Readers: Everyone
Mind the Gap: Evaluating Model- and Agentic-Level Vulnerabilities in LLMs with Action Graphs
Ilham Wicaksono, Zekun Wu, Rahul Patel, Theo King, Adriano Koshiyama, Philip Colin Treleaven
- AFAA 2026 Poster
- Readers: Everyone
Mind the Gap: Evaluating Model- and Agentic-Level Vulnerabilities in LLMs with Action Graphs
Ilham Wicaksono, Zekun Wu, Rahul Patel, Theo King, Adriano Koshiyama, Philip Colin Treleaven
- ICLR 2026 AIWILD
- Readers: Everyone
The Confidence Manifold: Geometric Structure of Correctness Representations in Language Models
Seonglae Cho, Zekun Wu, Kleyton Da Costa, Adriano Koshiyama
- Agentic AI in the Wild: From Hallucinations to Reliable Autonomy Poster
- Readers: Everyone
CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Features
Seonglae Cho, Zekun Wu, Adriano Koshiyama
- ICLR 2026 Re-Align Workshop
- Readers: Everyone
Control Reinforcement Learning: Interpretable Token-Level Steering of LLMs via Sparse Autoencoder Features
Seonglae Cho, Zekun Wu, Adriano Koshiyama
- ICLR 2026 Trustworthy AI
- Readers: Everyone
Control Reinforcement Learning: Interpretable Token-Level Steering of LLMs via Sparse Autoencoder Features
Seonglae Cho, Zekun Wu, Adriano Koshiyama
- Submitted to ICLR 2026
- Readers: Everyone
CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Features
Seonglae Cho, Zekun Wu, Adriano Koshiyama
- Submitted to ICLR 2026
- Readers: Everyone
Personality as a Probe for LLM Evaluation: Method Trade-offs and Downstream Effects
Gunmay Handa, Zekun Wu, Adriano Koshiyama, Philip Colin Treleaven
- NeurIPS 2025 LLM Evaluation Workshop Poster
- Readers: Everyone
Knowledge Collapse in LLMs: When Fluency Survives but Facts Fail under Recursive Synthetic Training
Figarri Keisha, Zekun Wu, Ze Wang, Adriano Koshiyama, Philip Colin Treleaven
- NeurIPS 2025 LLM Evaluation Workshop Poster
- Readers: Everyone

View all 22 publications

Co-Authors

View all 27 co-authors