Papers by Enyan Dai
CAGenMol: Condition-Aware Diffusion Language Model for Goal-Directed Molecular Generation (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to optimize target-directed molecular generation fail to reconcile conflicting objectives without compromising structural validity. |
| Approach: | They propose a condition-aware discrete diffusion framework that allows for conditional denoising guided by heterogeneous structural and property signals. |
| Outcome: | The proposed framework improves on structure-conditioned, property-conditioned and dual-conditioned benchmarks in binding affinity, drug-likeness, and success rate. |
DuFFin: A Dual-Level Fingerprinting Framework for LLMs IP Protection (2026.findings-eacl)
Copied to clipboard
| Challenge: | Large language models are valuable intellectual property due to the computational cost of training. |
| Approach: | They propose a dual-level fingerprinting framework that extracts trigger patterns and knowledge-level signatures to verify black-box ownership. |
| Outcome: | The proposed framework verifies the copyright of protected LLMs on their variants, achieving an IP-ROC greater than 0.99. |
Route to Rome Attack: Directing LLM Routers to Expensive Models via Adversarial Suffix Optimization (2026.acl-long)
Copied to clipboard
| Challenge: | Existing routing attacks depend on white-box access or heuristic prompts, rendering them ineffective in real-world black-box scenarios. |
| Approach: | They propose a cost-aware routing strategy that routes queries to the least-cost model . they propose heuristic prompts that are ineffective in real-world black-box scenarios . |
| Outcome: | The proposed approach significantly increases the routing rate to expensive models on queries of different distributions. |