Nakyeong Yang

Nakyeong Yang (양낙영)

Postdoctoral Researcher
Max Planck Institute for Software Systems

About

I am a postdoctoral researcher in Artificial Intelligence at Max Planck Institute for Software Systems. My research focuses on Large Language Models (LLMs) and Interpretable & Safe AI, particularly from a Mechanistic Interpretability perspective. I study how knowledge is represented at the neuronal level and manifests in model behavior, and develop interventions such as neuron-level pruning, targeted post-training, and activation steering to improve model safety. Building on this foundation, I am extending my research toward safe and trustworthy LLM agents, with the goal of understanding and improving how agentic systems reason, act, and interact with external environments in a reliable and controllable manner.


News

  • [2026.08] One paper (Personalization) is accepted to EMNLP 2026. NEW
  • [2026.06] I joined Max Planck Institute for Software Systems as a Postdoctoral Researcher. NEW
  • [2026.06] Two papers (Agent, Personalization) are accepted to ICML 2026 Workshops.
  • [2026.04] Three papers (Pretraining, Unlearning, Self-consistency) are accepted to ACL 2026.
  • [2025.08] Two papers (Unlearning, Pretraining) are accepted to ICLR 2026.
  • [2026.01] One paper (Persona) is accepted to EACL 2026.
  • [2025.12] I defended my Ph.D!
  • [2025.11] One paper (Math reasoning) is accepted to AAAI 2026.
  • [2025.10] One paper (Persona) is accepted to IJCNLP-AACL 2025.
  • [2025.08] Two papers (Unlearning, Story generation) are accepted to EMNLP 2025.
  • [2025.08] One paper (Compression) is accepted to CIKM 2025.
  • [2024.08] I joined Max Planck Institute for Security and Privacy as a research intern.
  • [2024.07] One paper (Video moment retrieval) is accepted to CIKM 2024.
  • [2024.05] One paper (AI Safety) is accepted to ACL 2024.
  • [2023.09] I joined LG AI Research as a research intern.

Publication

2026
[21] Personalized Privacy Control in LLMs via Attention Head Intervention PDF
Junseok Kim*, Nakyeong Yang*, Kyomin Jung (*Equal Contribution)
EMNLP 2026 Findings (& ICML 2026 FoGen)
[20] Tool Selection Bias Amplifies in Multi-turn User–Agent Interactions PDF
Nakyeong Yang, Junseok Kim, Kyomin Jung
ICML 2026 AIWILD
[19] Reliability-Aware Adaptive Self-Consistency for Efficient Sampling in LLM Reasoning PDF Code
Junseok Kim, Nakyeong Yang, Kyungmin Min, Kyomin Jung
ACL 2026 Findings
[18] How Training Data Shapes the Use of Parametric and In-Context Knowledge in Language Models PDF Code
Minsung Kim, Dong-kyum Kim, Jea Kwon, Nakyeong Yang, Kyomin Jung, Meeyoung Cha
ACL 2026
Max Planck Institute
[17] Rethinking Post-Unlearning Behavior of Large Vision-Language Models PDF
Minsung Kim, Nakyeong Yang, Kyomin Jung
ACL 2026 Findings
[16] Erase or Hide? Suppressing Spurious Unlearning Neurons for Robust Unlearning PDF
Nakyeong Yang, Dong-kyum Kim, Jea Kwon, Minsung Kim, Kyomin Jung, Meeyoung Cha
ICLR 2026
Max Planck Institute
[15] Bilinear representation mitigates reversal curse and enables consistent model editing PDF
Dong-kyum Kim, Minsung Kim, Jea Kwon, Nakyeong Yang, Meeyoung Cha
ICLR 2026
Max Planck Institute
[14] Persona Switch: Mixing Distinct Perspectives in Decoding Time PDF Code
Junseok Kim, Nakyeong Yang, Kyomin Jung
EACL 2026 Findings
[13] Confidence-Guided Stepwise Model Routing for Cost-Efficient Reasoning PDF
Sangmook Lee, Dohyung Kim, Hyukhun Koh, Nakyeong Yang, Kyomin Jung
AAAI 2026
2025
[12] Persona is a Double-edged Sword: Rethinking the Impact of Role-play Prompts in Zero-shot Reasoning Tasks PDF Code
Junseok Kim, Nakyeong Yang, Kyomin Jung
IJCNLP-AACL 2025 Findings
[11] FaithUn: Toward Faithful Forgetting in Language Models by Investigating the Interconnectedness of Knowledge PDF Code
Nakyeong Yang, Minsung Kim, Seunghyun Yoon, Joongbo Shin, Kyomin Jung
EMNLP 2025
Adobe Research
[10] Avoidance Decoding for Diverse Multi-Branch Story Generation PDF
Kyeongman Park, Nakyeong Yang, Kyomin Jung
EMNLP 2025
[9] Unplug and Play Language Models: Decomposing Experts in Language Models at Inference Time PDF
Nakyeong Yang, Jiwon Moon, Junseok Kim, Yunah Jang, Kyomin Jung
CIKM 2025 (Oral)
2024
[8] MVMR: A New Framework for Evaluating Faithfulness of Video Moment Retrieval against Multiple Distractors PDF Code
Nakyeong Yang, Minsung Kim, Seunghyun Yoon, Joongbo Shin, Kyomin Jung
CIKM 2024 (Oral)
Adobe Research
[7] Mitigating Biases for Instruction-following Language Models via Bias Neurons Elimination PDF Code
Nakyeong Yang, Taegwan Kang, Jungkyu Choi, Honglak Lee, Kyomin Jung
ACL 2024
LG AI Research
[6] LongStory: Coherent, Complete and Length Controlled Long story Generation PDF
Kyeongman Park, Nakyeong Yang, Kyomin Jung
PAKDD 2024 (Oral)
2023
[5] Task-specific Compression for Multi-task Language Models using Attribution-based Pruning PDF
Nakyeong Yang, Yunah Jang, Hwanhee Lee, Seohyeong Jeong, Kyomin Jung
EACL 2023 Findings
[4] Multi-View Zero-Shot Open Intent Induction from Dialogues: Multi Domain Batch and Proxy Gradient Transfer PDF
Hyukhun Koh, Haesung Pyun, Nakyeong Yang, Kyomin Jung
SIGDial 2023 Workshop (DSTC 11)
2022
[3] Deriving Explainable Discriminative Attributes Using Confusion About Counterfactual Class PDF
Nakyeong Yang, Taegwan Kang, Kyomin Jung
ICASSP 2022 (Oral)
[2] Semantic and Explainable Research-related Recommendation System based on Semi-supervised Methodology using BERT and LDA models PDF
Nakyeong Yang, Jeongje Jo, Myeongjun Jeon, Wooju Kim, Juyoung Kang
Expert Systems with Applications (2022)
2021
[1] RABERT: Relation-Aware BERT for Target-Oriented Opinion Words Extraction PDF
Taegwan Kang, Minwoo Lee, Nakyeong Yang, Kyomin Jung
CIKM 2021

Education

Research Experience