The lab's research is committed to understanding humanity and society to further develop human-like and trustworthy Artificial Intelligence. Recent primary interests include building trustworthy and safe Large Language Models (LLMs), with a focus on: (1) construction of safety datasets, benchmarks, and evaluation metrics; (2) controllable language generation; (3) LLM security, including adversarial attack and red-teaming; (4) safety alignment and learning methods.

HAILO Lab (Human-centered AI and Language for All)의 연구는 인간과 사회를 이해하여 인간과 유사하고 신뢰할 수 있는 인공지능을 더욱 발전시키는 데 전념하고 있습니다. 최근 주요 관심사는 다음과 같은 사항에 중점을 둔 신뢰할 수 있고 안전한 대규모 언어 모델(LLM) 구축입니다: (1) 안전 데이터셋, 벤치마크 및 평가 지표 구축; (2) 제어 가능한 언어 생성; (3) 적대적 공격 및 레드팀을 포함한 LLM 보안; (4) 안전 정렬 및 학습 방법.

Alignment Data Map

BLEnD-Vis

Drift

CSRT

CSCL

GUARD

AdvisorQA

MAQA

BLEnD

SWEET

APRICOT

KorNAT

TRAP

TimeChara

LifeTox

KoBBQ

Prometheus

ProPILE

SQuARe

KoSBi

Bayesian Red Teaming

Critic-Guided Decoding

ClaimDiff

Why Knowledge Distillation Amplifies Gender Bias and How to Mitigate - from the Perspective of DistilBERT

Masked Summarization to Generate Factually Inconsistent Summaries for Improved Factual Consistency Checking

Plug-and-Play Adaptation for Continuously-updated QA

TaleBrush: Sketching Stories with Generative Pretrained Language Models

Reasoning Visual Dialog with Sparse Graph Learning and Knowledge Transfer

SUMBT+LaRL: Effective Multi-domain End-to-end Neural Task-oriented Dialog System

Alignment Data Map for Efficient Preference Data Selection and Diagnosis

BLEnD-Vis: Benchmarking Multimodal Cultural Understanding in Vision Language Models

Drift: Decoding-time Personalized Alignments with Implicit User Preferences

Code-Switching Curriculum Learning for Multilingual Transfer in LLMs

CSRT: Evaluation and Analysis of LLMs using Code-Switching Red-Teaming Dataset

AdvisorQA: Towards Helpful and Harmless Advice-seeking Question Answering with Collective Intelligence

MAQA: Evaluating Uncertainty Quantification in LLMs Regarding Data Uncertainty

Guaranteed Generation from Large Language Models

BLEnD: A Benchmark for LLMs on Everyday Knowledge in Diverse Cultures and Languages

HyperCLOVA X Technical Report

TimeChara: Evaluating Point-in-Time Character Hallucination of Role-Playing Large Language Models

Calibrating Large Language Models Using Their Generations Only

TRAP: Targeted Random Adversarial Prompt Honeypot for Black-Box Identification

KorNAT: LLM Alignment Benchmark for Korean Social Values and Common Knowledge

Who Wrote this Code? Watermarking for Code Generation

LifeTox: Unveiling Implicit Toxicity in Life Advice

Prometheus: Inducing Fine-grained Evaluation Capability in Language Models

KoBBQ: Korean Bias Benchmark for Question Answering

ProPILE: Probing Privacy Leakage in Large Language Models

KoSBi: A Dataset for Mitigating Social Bias Risks Towards Safer Large Language Model Applications

SQuARe: A Large-Scale Dataset of Sensitive Questions and Acceptable Responses Created through Human-Machine Collaboration

Query-Efficient Black-Box Red Teaming via Bayesian Optimization

Critic-Guided Decoding for Controlled Text Generation

ClaimDiff: Comparing and Contrasting Claims on Contentious Topics

TaleBrush: Visual Sketching of Story Generation with Pretrained Language Models

Method and Apparatus for Providing Hybrid Intelligent Customer Consultation

Method and Apparatus for Dialogue State Tracking for Use in Goal-oriented Dialog System

SUMBT: Slot-Utterance Matching for Universal and Scalable Belief Tracking

Unpaired Speech Enhancement by Acoustic and Adversarial Supervision for Speech Recognition

Neural representations for speech recognition and natural language generation

Rescoring of N-best Hypotheses using Top-down Selective Attention for Automatic Speech Recognition

A Deep Chatbot for QA and Chitchat

Compositional Sentence Representation from Character within Large Context Text

Deep CNNs Along the Time Axis With Intermap Pooling for Robustness to Spectral Variations

Hierarchical committee of deep CNNs with exponentially-weighted decision fusion for static facial expression recognition

Active Learning for Large-scale Object Classification: from Exploration to Exploitation

Learning Tonotopically Organized Auditory Feature-map from Speech by an Intermap Pooling Layer in a Deep CNN

Spoken Sentence Embedding from Character by Jointly Learning Character-level Compositional Word Model and RNN Sentence Encoder

A Calibration Method for Eye-Gaze Estimation Systems Based on 3D Geometrical Optics

Smart User Interface for Mobile Consumer Devices Using Model-Based Eye-Gaze Estimation

As large language models become increasingly integrated into daily life, detecting implicit toxicity across diverse contexts is crucial. To this end, we introduce LifeTox, a dataset designed for identifying implicit toxicity within a broad range of advice-seeking scenarios. Unlike existing safety datasets, LifeTox comprises diverse contexts derived from personal experiences through open-ended questions. Experiments demonstrate that RoBERTa fine-tuned on LifeTox matches or surpasses the zero-shot performance of large language models in toxicity classification tasks. These results underscore the efficacy of LifeTox in addressing the complex challenges inherent in implicit toxicity. We open-sourced the dataset and the LifeTox moderator family; 350M, 7B, and 13B.

Hwaran Lee

Assistant Professor

HAILO Lab (Human-centered AI and Language for All)

교수님 소개

이화란

연구실 소개

연구실 키워드

알럼나이가 간 회사

연구실 지원 방법

대우 조건

연도별 논문 인용수

저널 & 학회 논문 출판 수

연구실 논문 목록

Alignment Data Map

BLEnD-Vis

Alignment Data Map for Efficient Preference Data Selection and Diagnosis

연구실 보유 특허

대학별 입시끈

우리 연구실 모집

교수님 소개

이화란

연구실 키워드

알럼나이가 간 회사

연구실 지원 방법

대우 조건

연도별 논문 인용수

저널 & 학회 논문 출판 수

연구실 논문 목록

Alignment Data Map

BLEnD-Vis

Alignment Data Map for Efficient Preference Data Selection and Diagnosis

연구실 보유 특허

최근 3년간 연구실적 높은 연구실

대학별 입시끈

우리 연구실 모집