Papers by Thy Thy Tran
\mathsf{Con Instruction}: Universal Jailbreaking of Multimodal Large Language Models via Non-Textual Modalities (2025.acl-long)
Copied to clipboard
| Challenge: | Existing attacks communicate instruction through text, accompanied by a toxic image or audio . a novel gray-box attack method generates adversarial images or audio to convey harmful instructions to MLLMs . |
| Approach: | They propose a gray-box attack method that generates adversarial images or audio to convey specific harmful instructions to MLLMs by following non-textual instruction. |
| Outcome: | The proposed method achieves highest success rates on visual and audio-language models . larger models are more susceptible toCon Instruction, compared to their underlying models - the results will be released . |
Towards Automated Error Discovery: A Study in Conversational AI (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent work shows that LLMs require information about the nature of an error or hints about its occurrence for accurate detection. |
| Approach: | They propose an encoder-based approach to detect and define errors in conversational AI. |
| Outcome: | The proposed framework outperforms baselines across multiple error-annotated dialogue datasets and shows strong generalization to unknown intent detection. |
The Devil is in the Details: On Models and Training Regimes for Few-Shot Intent Classification (2023.eacl-main)
Copied to clipboard
| Challenge: | Recent methods for task-oriented dialog (ToD) intent classification use pretrained language models . but lack of informative ablations prevents identification of factors that drive performance . |
| Approach: | They propose a framework to evaluate components of Few-Shot Intent Classification . they propose to combine cross-encoder architecture and episodic meta-learning . |
| Outcome: | The proposed framework evaluates cross-encoder architecture and episodic meta-learning . it also shows that splitting episodes into support and query sets outperforms non-episodic counterparts. |
Revisiting Unsupervised Relation Extraction (2020.acl-main)
Copied to clipboard
| Challenge: | Unsupervised relation extraction (URE) extracts relations between named entities from raw text without manually-labelled data and existing knowledge bases (KBs). |
| Approach: | They compare unsupervised relation extraction methods to generative and discriminative approaches . they conclude that entity types provide a strong inductive bias for URE . |
| Outcome: | The proposed method outperforms generative and discriminative approaches on two popular datasets. |