Mitigating Hallucination in Vision-Language Model with Depth and Spatial-aware Key-Value Refinement

BeaconKV: Key-Value Cache Compression Guided by Beacon Queries for Efficient Large Reasoning Model Inference

Mitigating Object and Relationship Hallucination in Large Vision Language Model with Multi-Agent Guidance

ATO: Adaptive Target Optimization for Semi-Supervised Domain Adaptation via Deep Reinforcement Learning

Adaptive Capacity Allocation for Vision-Language-Action Fine-Tuning

Feedback Adaptation for Retrieval-Augmented Generation

Encoder-Free Style Controllable Text-to-Speech with Voice Attribute Vectors

Famigo: A Privacy-Preserving Hybrid Voice Assistant for Multi-User Family Environments

Mask2Flow-TSE: Two-Stage Target Speaker Extraction with Masking and Flow Matching

Whisper-CD: Accurate Long-Form Speech Recognition using Multi-Negative Contrastive Decoding

WAND: Windowed Attention and Knowledge Distillation for Efficient Autoregressive Text-to-Speech Models

Unlocking Fine-Grained and Within-Utterance Speaking Style Control in Prompt-Based Text-to-Speech Models

DialBGM: A Benchmark for Background Music Recommendation from Everyday Multi-Turn Dialogues

Revealing Multi-View Hallucination in Large Vision-Language Models

Visually Guided Decoding: Gradient-free Hard Prompt Inversion with Language Models

Chain-of-Rank: Enhancing Large Language Models for Domain-specific RAG in Edge Device

Learning Primitive Relations for Compositional Zero-shot Learning

Learning Contextual Retrieval for Robust Conversational Search

Towards Comprehensive Scene Understanding: Integrating First and Third-Person Views for LVLMs

InfiniPot-V: Memory-Constrained KV Cache Compression for Streaming Video Understanding

Unlocking Transfer Learning for Open-World Few-shot Recognition

P2VA: Converting Persona Descriptions into Voice Attributes for Fair and Controllable Text-to-Speech

Evaluating Hallucinations in Multimodal LLMs with Spoken Queries under Diverse Acoustic Conditions

Leveraging Adapter for Parameter-efficient ASR Encoder

InfiniPot: Infinite Context Processing on Memory-constrained LLMs

SToRI: Semantic Token Reweighting for Interpretable and Controllable Text Embeddings in Vision-language Models

Preserving Pre-trained Representation Space: On Effectiveness of Prefix-tuning for Large Multi-modal Models

Crayon: Customized On-device LLM via Instant Adapter Blending and Edge-server Hybrid Inference

Expand-and-Quantize: Unsupervised Semantic Segmentation Using High-dimensional Space and Product Quantization

Knowledge Distillation from Non-streaming to Streaming ASR Encoder using Auxiliary Non-streaming Layer

Depth-relative Self-attention for Monocular Depth Estimation

Teacher Intervention: Improving Convergence of Quantization Aware Training for Ultra-low Precision Transformers

Improving Small Footprint Few-shot Keyword Spotting with Supervision on Auxiliary Data

Vision Transformer-based Feature Extraction for Generalized Zero-shot Learning

Semantic-preserving Augmentation for Robust Image-text Retrieval

Spatial Cross-attention for Transformer-based Image Captioning

Task-agnostic Open-set Prototype for Few-shot Open-set Recognition

Korean Tokenization for Beam Search Rescoring in Speech Recognition

Understanding the Role of Self-attention for Efficient Speech Recognition

Similarity and Content-based Phonetic Self-attention for Speech Recognition

Semantic Feature Extraction for Generalized Zero-Shot Learning

Layer-wise Pruning of Transformer Attention Heads for Efficient Language Modeling

TernGEMM: General Matrix Multiply Library with Ternary Weights for Fast DNN Inference

SVD-Softmax: Fast Softmax Approximation on Large Vocabulary Neural Networks

Role of Sensing and Computer Vision in 6G Wireless Communications

Toward Intelligent Millimeter and Terahertz Communication for 6G: Computer Vision-aided Beamforming

Pseudo-label-free Weakly Supervised Semantic Segmentation using Image Masking

Optimization of General Matrix Multiply Library for Ternary Weight for Fast DNN Inference

In this paper, we reveal that self-attention in speech recognition models performs two distinct roles: phonetic and linguistic localization. Especially, we show that phonetic localization in the lower layers extracts phonologically meaningful features from speech and reduces the phonetic variance in the utterance for proper linguistic localization in the upper layers . From this understanding, we discover that attention maps can be reused as long as their localization capability is preserved. We propose the layer-wise attention map reuse technique and achieve 2x speedup without performance loss.

We propose a novel architecture for efficient speech recognition. The model utilizes the parameter-sharing technique and a unique adapter module for Conformer-based ASR encoders. The proposed model reduces approximately 50% of parameters and 20% of computations.

Leveraging Adapter for Parameter-Efficient ASR Encoder

To improve the performance of streaming ASR, we propose a layer-to-layer knowledge distillation (KD) from non-streaming teacher's encoder to streaming student's encoder.

Handling long input contexts remains a significant challenge for Large Language Models (LLMs), particularly in resource-constrained environments such as mobile devices. Our work aims to address this limitation by introducing InfiniPot, a novel KV cache control framework designed to enable pre-trained LLMs to manage extensive sequences within fixed memory constraints efficiently, without requiring additional training.

InfiniPot: Infinite Context Processing on Memory-Constrained LLMs

The customization of large language models (LLMs) for user-specified tasks becomes important. Yet, the performance of on-device LLMs is inherently constrained by the limitations of small-scale models. To overcome these restrictions, we propose Crayon, a novel approach for on-device LLM customization that 1) instantly blends adapters into a customized adapter without extra training, and 2) adopts device-server hybrid inference.

Crayon: Customized On-Device LLM via Instant Adapter Blending and Edge-Server Hybrid Inference

We introduce Visually Guided Decoding (VGD), a gradient-free approach that leverages LLMs and CLIP-based guidance to generate coherent and semantically aligned prompts for text-to-image (T2I) generative models.

We propose a framework of Semantic Token Reweighting to build Interpretable text embeddings (SToRI), facilitating the interpretative analysis of vision tasks through natural language. SToRI can be used for few-shot image classification, data-driven insight extraction, and image retrieval tailored to user preferences.

Semantic Token Reweighting for Interpretable and Controllable Text Embeddings in CLIP

음성 및 자연어처리 연구실

교수님 소개

심규홍

연구실 소개

연구실 키워드

알럼나이가 간 회사

연구실 지원 방법

대우 조건

연도별 논문 인용수

저널 & 학회 논문 출판 수

연구실 논문 목록

Mitigating Hallucination in Vision-Language Model with Depth and Spatial-aware Key-Value Refinement

BeaconKV: Key-Value Cache Compression Guided by Beacon Queries for Efficient Large Reasoning Model Inference

Mitigating Object and Relationship Hallucination in Large Vision Language Model with Multi-Agent Guidance

연구실 보유 특허

대학별 입시끈

우리 연구실 모집

교수님 소개

심규홍

연구실 키워드

알럼나이가 간 회사

연구실 지원 방법

대우 조건

연도별 논문 인용수

저널 & 학회 논문 출판 수

연구실 논문 목록

Mitigating Hallucination in Vision-Language Model with Depth and Spatial-aware Key-Value Refinement

BeaconKV: Key-Value Cache Compression Guided by Beacon Queries for Efficient Large Reasoning Model Inference

Mitigating Object and Relationship Hallucination in Large Vision Language Model with Multi-Agent Guidance

연구실 보유 특허

최근 3년간 연구실적 높은 연구실

대학별 입시끈

우리 연구실 모집