Papers by Wenya Guo
MTSA: Multi-turn Safety Alignment for LLMs through Multi-round Red-teaming (2025.acl-long)
Copied to clipboard
| Challenge: | Existing jailbreak techniques rely on single-round interactions, pro-Corresponding author. |
| Approach: | They propose a multi-turn safety alignment framework to address the challenge of securing large language models in multi-round interactions. |
| Outcome: | The proposed framework exhibits state-of-the-art attack capabilities while improving safety performance on safety benchmarks. |
Licon: A Diverse, Controllable and Challenging Linguistic Concept Learning Benchmark (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for Concept Learning focus on visual information, but visual information cannot present abstract concepts exactly, which struggles the introduction of novel concepts related to known concepts. |
| Approach: | They propose a benchmark where concepts in diverse forms are defined by linguistic descriptions and an entailment-based concept learning method to model the relationship among concepts. |
| Outcome: | The proposed benchmark is based on the existing visual concepts learning benchmarks and will be released to the public soon. |
AoM: Detecting Aspect-oriented Information for Multimodal Aspect-Based Sentiment Analysis (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to extract aspects from text-image pairs and recognize their sentiments are noisy and coarsely establishing image-aspect alignment will interfere with aspect-relevant semantic and sentiment information. |
| Approach: | They propose an Aspect-oriented method to detect aspect-relevant semantic and sentiment information by selecting textual tokens and image blocks that are semantically related to the aspects. |
| Outcome: | The proposed method is superior to existing methods in the field of sentiment analysis. |
Look before You Leap: Dual Logical Verification for Knowledge-based Visual Question Generation (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing methods for visual question generation focus on leveraging the semantics of inputs to propose questions, ignoring the logical coherence between generated questions and images. |
| Approach: | They propose a logical verification method that checks logical structure between Q, images, answers and acquired outside knowledge by incorporating logical coherence between Q and Q twice in the whole procedure. |
| Outcome: | The proposed method can generate diverse and insightful knowledge-based visual questions on two common datasets. |
HyperPELT: Unified Parameter-Efficient Language Model Tuning for Both Language and Vision-and-Language Tasks (2023.findings-acl)
Copied to clipboard
| Challenge: | Pretraining and fine-tuning are the dominant paradigms in natural language processing. |
| Approach: | They propose a parameter-efficient multitask learning framework that takes trainable hyper-embeddings and visual modality as input and outputs weights for different modules in a pretrained language model. |
| Outcome: | The proposed framework adds fewer trainable parameters in multi-task learning while achieving superior performances and transfer ability compared to state-of-the-art methods. |
From Competition to Synergy: Unlocking Reinforcement Learning for Subject-Driven Image Generation (2026.acl-long)
Copied to clipboard
| Challenge: | a naive application of GRPO leads to conflicting gradient signals and a misalignment with the temporal dynamics of the diffusion process. |
| Approach: | They propose a framework that uses synergy-aware reward shaping to penalize conflicted reward signals and amplify synergies to provide a sharper and decisive gradient. |
| Outcome: | The proposed framework outperforms naive GRPO and Time-Aware Dynamic Weighting (TDW) on DreamBench, and achieves a state-of-the-art balance between ID preservation and prompt adherence. |
A Span-based Multimodal Variational Autoencoder for Semi-supervised Multimodal Named Entity Recognition (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for named entity recognition on social media are not efficient for semi-supervised MNER because of the mismatch between the posted text and image. |
| Approach: | They propose a novel method to fuse the text and image features for multimodal named entity recognition under semi-supervised setting by exploiting modal-specific VAEs. |
| Outcome: | The proposed method outperforms baselines under supervised setting and improves performance with less labeled data than existing semi-supervised methods. |