Elamur Research⌕
☰
home / authors / Lei Sha

Lei Sha

Lei Sha has 7 papers in our library, in 2015–2026, most of them on Bias, fairness & safety, Language models & pretraining, Reasoning; often with Hao Li, Rui Li, Hao Wang.

7 papers in our library · 2015–2026

OpenAlex · ORCID · h-index 17 · 50 works · 1,360 citations OpenAlex, 2026-10-14

📄 7 works in our library · 🔗 Cited 77 times in our library · 3 more in our library without a page yet · ⭐ Paper Stars ranking: #7438

The profile is the library's match of the name; two people with one name are not always told apart.

Elsewhere: Google Scholar · Semantic Scholar · dblp · ACL Anthology

 Affiliations

🎓Beihang UniversityBeijing · China2022–2026 · 12 papersEmployment since 2022 · Professor, Institute of Artificial Intelligencecurrent🎓Research Institute of Tsinghua University in ShenzhenShenzhen · China2026 · 1 papercurrent🎓Peking UniversityBeijing · China2013–2020 · 24 papersEducation 2013–2018 · PhD, EECS🎓University of OxfordOxford · United Kingdom2020–2024 · 8 papersEmployment 2020–2022 · Research Associate, Department of Computer Science🏢AppleCupertino · United States2018–2020Employment 2018–2020 · Senior NLP Research Scientist, Siri Understanding

From the bylines of their papers (OpenAlex, 2026-10-14), ORCID and Wikidata; institutions on papers are matched by machine. 4 more institutions on single bylines are not listed.

 Fields, topics and concepts

Of its works in our library, with the number of works on each.

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Artificial Intelligence · 7 worksArts and Humanities · 7 worksComputer Science · 7 worksLanguage and Linguistics · 7 worksSafety Research · 2 worksSocial Sciences · 2 works

🧭 Topics 3Research subjects and languages the work is about.Bias, fairness & safety · 6 worksReasoning · 3 worksLanguage models & pretraining · 3 works

 Works by year

2020s

20263 papers20253 papers

2010s

20151 paper

 Papers

The most cited first (OpenAlex and our library).

  1. Reasoning-to-Defend: Safety-Aware Reasoning Can Defend Large Language Models from Jailbreaking (2025 · 21 citations)
    Making LLMs safer usually depends on external filters or on supervised tuning that teaches blunt refusals. These approaches ignore the models' own reasoning abilities, and blunt refusals tend to generalize poorly. The authors propose Reasoning-to-Defend (R2D), a training paradigm with two stages:
  2. LLMs know their vulnerabilities: Uncover Safety Gaps through Natural Distribution Shifts (2025 · 18 citations)
    The paper investigates a safety weakness of aligned LLMs. Seemingly harmless prompts that are semantically connected to a harmful request can steer a model into producing unsafe output. The authors call this exploitation of "natural distribution shift" and contrast it with attacks that deliberately distort the toxic…
  3. Layer-Aware Representation Filtering: Purifying Finetuning Data to Preserve LLM Safety Alignment (2025 · 11 citations)
    Fine-tuning a safety-aligned LLM can weaken its refusal behaviour even when the training data look harmless.
  4. Recognizing Textual Entailment Using Probabilistic Inference (2015 · 10 citations)
    The paper addresses Recognizing Textual Entailment (RTE), which asks whether a hypothesis H follows from a text T. The authors see two problems with earlier work. Shallow overlap systems miss facts that are only implied. Converting sentences into strict logic loses information and makes proofs fragile.
  5. HAUNTATTACK: When Attack Follows Reasoning as a Shadow (2026 · 3 citations)
    Large Reasoning Models (LRMs) such as DeepSeek-R1 and o1 are strong at multi-step problem solving, but it is unclear whether this skill becomes a safety weakness when harmful intent is woven into the reasoning itself.
  6. Be Your Own Red Teamer: Safety Alignment via Self-Play and Reflective Experience Replay (2026 · 1 citation)
    LLM safety training usually depends on fixed red-teaming resources, such as static defense prompts or pre-collected jailbreak datasets.
  7. SafetyMem: Adaptive Jailbreak Defense via Dual-Component Safety Memory (2026)
    LLM jailbreak defenses tend to fall into two camps. Methods that change model weights, such as fine-tuning or trained detectors, are costly and slow to adapt. Inference-time filters are flexible, but they start from scratch on every prompt. SafetyMem is a training-free framework that lets a model accumulate defense…

All works

10 works of theirs in our library, the most cited first; 7 with a page here, the others link their original until we summarise them.

  1. Encoding Temporal Information for Time-Aware Link Prediction ↗Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing · no summary yetTingsong Jiang et al.2016 · 148 citations
  2. Attentive Interactive Neural Networks for Answer Selection in Community Question Answering ↗Proceedings of the AAAI Conference on Artificial Intelligence · no summary yetXiaodong Zhang et al.2017 · 97 citations
  3. Reasoning-to-Defend: Safety-Aware Reasoning Can Defend Large Language Models from JailbreakingConference on Empirical Methods in Natural Language Processing (EMNLP)Junda Zhu, Lingyong Yan, Shuaiqiang Wang, et al.2025 · 21 citations
  4. LLMs know their vulnerabilities: Uncover Safety Gaps through Natural Distribution ShiftsAnnual Meeting of the Association for Computational Linguistics (ACL)Qibing Ren, Hao Li, Dongrui Liu, et al.2025 · 18 citations
  5. Layer-Aware Representation Filtering: Purifying Finetuning Data to Preserve LLM Safety AlignmentConference on Empirical Methods in Natural Language Processing (EMNLP)Hao Li, Lijun Li, Zhenghao Lu, et al.2025 · 11 citations
  6. Benchmarking Multi-Step Legal Reasoning and Analyzing Chain-of-Thought Effects in Large Language Models ↗arXiv:2511.07979 · no summary yetWenhan Yu et al.2025 · 11 citations
  7. Recognizing Textual Entailment Using Probabilistic InferenceConference on Empirical Methods in Natural Language Processing (EMNLP)Lei Sha, Sujian Li, Baobao Chang, et al.2015 · 10 citations
  8. HAUNTATTACK: When Attack Follows Reasoning as a ShadowFindings of the Association for Computational LinguisticsJingyuan Ma, Rui Li, Zheng Li, et al.2026 · 3 citations
  9. Be Your Own Red Teamer: Safety Alignment via Self-Play and Reflective Experience ReplayFindings of the Association for Computational LinguisticsHao Wang, Yanting Wang, Hao Li, et al.2026 · 1 citation
  10. SafetyMem: Adaptive Jailbreak Defense via Dual-Component Safety MemoryAnnual Meeting of the Association for Computational Linguistics (ACL)Hao Wang, Ziyi Ni, Huacan Wang, et al.2026

 Co-authors here

12 of 20 co-authors with a page, the most papers together first.

Venues

Where their works here were published, the most works first, with the editions of those that have a page.

 Cited by

Papers of our library cite their works 77 times (of the 149,439 papers whose bibliographies we read); 15 citing papers with a page here →.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/authors/lei-sha/