Papers by Ting Deng

10 papers
Beware of Your Po! Measuring and Mitigating AI Safety Risks in Role-Play Fine-Tuning of LLMs (2025.acl-long)

Copied to clipboard

Challenge: Existing role-play fine-tuning techniques improve role adaptability but may degrade safety performance, especially for villainous characters.
Approach: They propose safety-aware Role-Play Fine-Tuning (SaRFT) to balance role-playing capabilities and safety.
Outcome: The proposed method outperforms state-of-the-art baselines under both LoRA and full-parameter fine-tuning settings.
Lightweight Spatial Modeling for Combinatorial Information Extraction From Documents (2023.findings-eacl)

Copied to clipboard

Challenge: Existing datasets do not cover documents with complex spatial structures and a lack of spatial information for document entity classification.
Approach: They propose a new spatial bias in attention calculation based on the K-nearest-neighbor graph of document entities that limits entities’ attention to their local radius.
Outcome: The proposed model outperforms baselines in most entity types and is highly parameter-efficient compared to existing methods.
On Safety Risks in Experience-Driven Self-Evolving Agents (2026.findings-acl)

Copied to clipboard

Challenge: Experience-driven self-evolution has emerged as a promising paradigm for improving the autonomy of large language model agents, yet its reliance on self-curated experience introduces underexplored safety risks.
Approach: They investigate how experience accumulation and utilization in self-evolving agents affect safety performance across web-based and embodied environments.
Outcome: The findings expose inherent limitations of current self-evolving agents and call for more principled strategies to ensure safe and reliable adaptation.
MPO: Multilingual Safety Alignment via Reward Gap Optimization (2025.acl-long)

Copied to clipboard

Challenge: Existing preference learning methods for safety alignment are monolingual and struggle with noisy multilingual data.
Approach: They propose a multilingual reward gaP optimization approach that leverages the well-aligned safety capabilities of the dominant language to improve safety alignment across multiple languages.
Outcome: Extensive experiments on three LLMs, LLaMA-3.1, Gemma-2 and Qwen2.5, validate MPO’s efficacy in multilingual safety alignment without degrading general multilingual utility.
Context Attribution with Multi-Armed Bandit Optimization (2026.findings-acl)

Copied to clipboard

Challenge: Existing approaches to augmenting attribution with retrieval-augmented generation (RAG) focus on training models to explicitly cite context segments during generation, but their reliability remains unverifiable.
Approach: They propose a framework that formulates context attribution as a combinatorial multi-armed bandit problem by using Linear Thompson Sampling to efficiently identify the most influential context segments while minimizing the number of model queries.
Outcome: The proposed method reduces model queries by 30% while matching or exceeding the attribution quality of existing approaches.
PromptBERT: Improving BERT Sentence Embeddings with Prompts (2022.emnlp-main)

Copied to clipboard

Challenge: Existing research shows that BERT and RoBERTa are poor in sentence embeddings due to static token embeddable bias and ineffective BERT layers.
Approach: They propose a novel contrastive learning method for better sentence embeddings by using a template denoising technique.
Outcome: The proposed method achieves 2.29 and 2.58 points of improvement compared to SimCSE and RoBERTa in the unsupervised setting.
Chain of Strategy Optimization Makes Large Language Models Better Emotional Supporter (2025.findings-emnlp)

Copied to clipboard

Challenge: Existing supervised fine-tuning (SFT) fails to address these issues, as it trains models on single gold-standard responses without modeling nuanced strategy trade-offs.
Approach: They propose a two-stage framework that optimizes strategy selection preferences at each dialogue turn.
Outcome: The proposed framework improves strategy selection preferences at each dialogue turn.
AdaSteer: Your Aligned LLM is Inherently an Adaptive Jailbreak Defender (2025.emnlp-main)

Copied to clipboard

Challenge: Activation steering offers training-free defense but relies on fixed steering coefficients, resulting in suboptimal protection and increased false rejections of benign inputs.
Approach: They propose an adaptive activation steering method that dynamically adjusts model behavior based on input characteristics.
Outcome: The proposed method outperforms baseline methods across multiple jailbreak attacks with minimal impact on utility.
Path Spuriousness-aware Reinforcement Learning for Multi-Hop Knowledge Graph Reasoning (2023.eacl-main)

Copied to clipboard

Challenge: Multi-hop reasoning is a common approach for query answering, but can be biased to spurious paths which coincidentally lead to the correct answer with poor explanation.
Approach: They propose a method that quantitatively estimates to what extent a path is spurious by a metric called Path Spuriousness (PS) they propose KG reasoning, which infers new facts along existing paths in KGs.
Outcome: The proposed model significantly improves the agent’s ability to prevent spurious paths while keeping comparable to state-of-the-art performance.
DisCo: Distilled Student Models Co-training for Semi-supervised Text Mining (2023.emnlp-main)

Copied to clipboard

Challenge: Existing text mining models are fine-tuned by fine-timing a large pre-trained language model (PLM) in downstream tasks.
Approach: They propose a semi-supervised learning framework for fine-tuning a cohort of small student models generated from a large pre-trained language model using knowledge distillation.
Outcome: The proposed framework outperforms baseline models on semi-supervised text classification and extractive summarization tasks while maintaining comparable performance.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations