Papers by Elias Bassani
Denoising Attention for Query-aware User Modeling (2024.findings-naacl)
Copied to clipboard
| Challenge: | Recent work has proposed to build user models at query time by leveraging the Attention mechanism, which allows weighing the contribution of the user-related information w.r.t. the current query. |
| Approach: | They propose to use the Attention mechanism to build user models at query time by weighing the contribution of the user-related information w.r.t. the Attention variant adopts a robust normalization scheme and introduces . filtering mechanism to better discern among the user related data those helpful for personalization. |
| Outcome: | The proposed approach improves MAP, MRR, and NDCG above 15% w.r.t. other Attention variants at the state-of-the-art. |
GuardBench: A Large-Scale Benchmark for Guardrail Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Lack of a standard benchmark for guardrail models poses significant evaluation issues . lack of standardized benchmark makes it hard to compare results across scientific publications. |
| Approach: | They propose a large-scale benchmark for guardrail models comprising 40 safety evaluation datasets. |
| Outcome: | The proposed model achieves competitive results without specific fine-tuning without the need for specific fine tuning. |
On Guardrail Models’ Robustness to Mutations and Adversarial Attacks (2025.findings-emnlp)
Copied to clipboard
| Challenge: | generative AI systems providing unsafe information has raised significant concerns, emphasizing the need for safety guardrails. |
| Approach: | They propose to evaluate 15 state-of-the-art guardrail models to assess their robustness to input mutations and adversarial attacks designed to bypass models’ safety alignment. |
| Outcome: | The proposed models are robust to input mutations and adversarial attacks that bypass models’ safety alignment. |