Papers by Jun Sakuma
When Benchmarks Leak: Inference-Time Decontamination for LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | a large number of large language models (LLMs) are being evaluated for their performance, but their reliability is threatened by test set contamination. |
| Approach: | They propose a framework that decontaminates large language models by applying small perturbations to the input embedding space. |
| Outcome: | The proposed framework achieves strong decontamination effectiveness while incurring minimal degradation in benign utility. |
Differentially Private Synthetic Text Generation for Retrieval-Augmented Generation (RAG) (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing private RAG methods rely on query-time differential privacy (DP) Existing studies have identified significant privacy risks when their databases contain sensitive information. |
| Approach: | They propose a framework that generates differentially private RAG databases using LLMs . Unlike prior methods, the synthetic text can be reused once created . |
| Outcome: | Experiments show that DP-SynRAG achieves superior performance to state-of-the-art RAG systems while maintaining a fixed privacy budget. |
Pattern Enhanced Multi-Turn Jailbreaking: Exploiting Structural Vulnerabilities in Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing multi-turn methods for large language models exploit conversational context to bypass safety constraints gradually. |
| Approach: | They propose a framework of five conversation patterns to construct multi-turn jailbreaks through natural dialogue. |
| Outcome: | The proposed framework exploits conversational contexts to construct multi-turn jailbreaks . it reveals that models exhibit distinct weakness profiles and model families share similar failure modes . |