Papers by Shang Ma

13 papers
Interpretable Safety Alignment via SAE-Constructed Low-Rank Subspace Adaptation (2026.acl-long)

Copied to clipboard

Challenge: Prior work has shown that safety behaviors are governed by low-rank structures . Low-Rank Adaptation (LoRA) consistently underperforms full fine-tuning and reinforcement learning on safety benchmarks .
Approach: They propose a safety alignment system that disentangles safety-relevant directions into monosemantic features and constructs an interpretable safety subspace from SAE directions.
Outcome: Empirically, the proposed model achieves 99.6% safety rates across multiple model families and scales . low-rank Adaptation consistently underperforms full fine-tuning and reinforcement learning on safety benchmarks compared with previous methods .
PsyScam: A Benchmark for Psychological Techniques in Real-World Scams (2025.findings-emnlp)

Copied to clipboard

Challenge: PTs are employed by scammers to manipulate victims and cause lasting psychological trauma.
Approach: They propose a benchmark to capture the PTs employed in real-worldscam reports and investigate how LLMs can be utilized to generate variants of scams based on the pts and the contexts provided by thesescams.
Outcome: The proposed model can generate variants of scams based on the PTs employed in real-world scam reports and the contexts provided by these scams.
BASS: Batched Attention-optimized Speculative Sampling (2024.findings-acl)

Copied to clipboard

Challenge: Speculative decoding has emerged as a powerful method to improve latency and throughput in hosting large language models.
Approach: They propose a batched speculative decoding system that generates sequences at an average speed of 5.8ms per token and a batch size of 8 at a 2.15 speed-up over optimized regular decoding.
Outcome: The proposed system achieves state-of-the-art latency and speed-up over optimized regular decoding.
Pair-Based Joint Encoding with Relational Graph Convolutional Networks for Emotion-Cause Pair Extraction (2022.emnlp-main)

Copied to clipboard

Challenge: Emotion cause pair extraction (ECPE) aims to extract emotion clauses and corresponding cause clauses.
Approach: They propose a novel task called emotion-cause pair extraction to extract emotion clauses and corresponding cause clauses.
Outcome: The proposed task can extract emotion clauses and cause clauses, and achieve state-of-the-art performance on the Chinese benchmark corpus.
A Span-based Dynamic Local Attention Model for Sequential Sentence Classification (2021.acl-short)

Copied to clipboard

Challenge: Existing methods for sentence classification ignore latent segment structure of document, in which contiguous sentences have coherent semantics.
Approach: They propose a span-based dynamic local attention model that captures structural information by supervised dynamic local focus.
Outcome: The proposed model outperforms state-of-the-art models on two benchmark datasets.
Crowd Comparative Reasoning: Unlocking Comprehensive Evaluations for LLM-as-a-Judge (2025.acl-long)

Copied to clipboard

Challenge: Existing methods rely on majority voting or criteria expansion to capture detailed and detailed details, often leading to incomplete outcomes.
Approach: They propose a method which introduces additional crowd responses to compare with the candidate responses, thereby exposing deeper and more comprehensive details within the candidate answers.
Outcome: Experiments show that the proposed method improves evaluation reliability and achieves an average gain of 6.7% across five benchmarks.
Joint Constrained Learning with Boundary-adjusting for Emotion-Cause Pair Extraction (2023.acl-long)

Copied to clipboard

Challenge: Emotion-Cause Pair Extraction (ECPE) aims to identify the document’s emotion clauses and corresponding cause clauses.
Approach: They propose a constrained learning framework with boundary-adjusting for Emotion-Cause Pair Extraction that summarizes prior rules and forces the model to take them into consideration in optimization.
Outcome: The proposed framework achieves competitive results compared with state-of-the-art methods on unbalanced data and proves robustness on unbalancing data.
Unraveling and Mitigating Safety Alignment Degradation of Vision-Language Models (2025.findings-acl)

Copied to clipboard

Challenge: LLaVA-7B demonstrated a decline in safety alignment ability on multi-modal inputs compared to its LLM backbone.
Approach: They propose a method to recover alignment ability from LLM backbone while preserving functional capabilities of VLMs.
Outcome: The proposed framework recovers alignment ability that is inherent in the LLM backbone with minimal impact on fluency and linguistic capabilities of pre-trained VLMs.
Few-Shot Data-to-Text Generation via Unified Representation and Multi-Source Learning (2023.acl-long)

Copied to clipboard

Challenge: Existing methods for data-to-text generation focus on specific types of structured data.
Approach: They propose a method that provides a unified representation that can handle various forms of structured data such as tables, knowledge graph triples, and meaning representations.
Outcome: The proposed method improves zero-shot and few-shot scenarios and can adapt to new structured data.
Preserving Commonsense Knowledge from Pre-trained Language Models via Causal Inference (2023.acl-long)

Copied to clipboard

Challenge: Existing studies attribute catastrophic forgetting to fine-tuning, and they retain pre-trained knowledge indiscriminately without identifying what knowledge is transferable.
Approach: They propose a unified objective for fine-tuning to retrieve the causality back from pre-trained data and use it to mitigate negative transfer while preserving knowledge.
Outcome: The proposed method outperforms state-of-the-art fine-tuning methods on commonsense QA datasets and can be implemented as a plug-in module to inflate the performance of existing QA models.
CodeFort: Robust Training for Code Generation Models (2024.findings-emnlp)

Copied to clipboard

Challenge: Existing research efforts to improve code generation models are inadequate . code generation model performance is degraded under small perturbations .
Approach: They propose a framework to improve the robustness of code generation models by generalizing code perturbations to enrich training data and enabling various robust training strategies.
Outcome: The proposed framework increases pass rates and robustness drop rate against code-syntax perturbations.
MATCH: Modulating Attention via In-Context Retrieval for Long-Context Transformers (2026.acl-long)

Copied to clipboard

Challenge: Existing approaches to improve efficiency often enforce rigid structural constraints such as local attention windows.
Approach: They propose a framework that augments sparse-attention mechanisms with dynamically integrated in-context information through an efficient retrieval system.
Outcome: Empirical results show that MATCH significantly improves the performance of sparse-attention models on synthetic and real-world natural-language tasks.
Enhancing Agentic Textual Graph Retrieval with Synthetic Stepwise Supervision (2026.acl-long)

Copied to clipboard

Challenge: Existing methods for integrating textual graphs with LLMs are limited by symbolic inference and high annotation costs.
Approach: They propose a textual graph reasoning framework that integrates textual diagrams with large language models.
Outcome: The proposed approach achieves 15.6% accuracy and 17.2% in F1 score on three common datasets.

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