Page 29 - Paper Library - AiPaper

CRS-Que: A User-centric Evaluation Framework for Conversational Recommender Systems

Published:11/2/2023

Conversational Recommender SystemsUser-Centric Evaluation FrameworkUser Experience Evaluation MetricsMusic Exploration RecommendationMobile Phone Purchase Recommendation

This paper presents CRSQue, a usercentric evaluation framework for conversational recommender systems, built on ResQue. It integrates conversationrelated UX metrics and validates its effectiveness and reliability across different scenarios, highlighting the interaction between

Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection

Published:10/18/2023

Self-Reflective Retrieval-Augmented GenerationQuality Improvement for Large Language ModelsControllable Generation and Reflection MechanismOpen-Domain Question Answering TasksFact Verification and Citation Accuracy

The SelfRAG framework enhances large language models' quality and factual accuracy through adaptive retrieval and reflection tokens. Experiments show it significantly outperforms existing models in opendomain QA, reasoning, and fact verification tasks.

The Efficacy of Conversational Artificial Intelligence in Rectifying the Theory of Mind and Autonomy Biases: Comparative Analysis

Published:6/20/2024

Application of Conversational AI in Mental Health InterventionsCognitive Bias RectificationTherapeutic Chatbot Effectiveness EvaluationAffect Recognition in Human-AI InteractionComparison of General-Purpose and Therapeutic Language Models

This study evaluated the effectiveness of conversational AI in correcting Theory of Mind and autonomy biases, showing that generalpurpose chatbots outperform therapeutic ones in identifying and rectifying these biases and recognizing emotional responses.

BOSS: Blocking algorithm for optimizing shuttling scheduling in Ion Trap

Published:12/5/2024

Ion Trap Scheduling OptimizationQuantum Computing Scheduling AlgorithmHigh-Fidelity Quantum GatesQuantum Hardware Performance EnhancementEfficient Scheduling Algorithms

The BOSS algorithm enhances shuttling efficiency in ion traps, addressing challenges in fidelity and execution times, achieving up to 96.1% reduction in shuttle operations. It includes simulations that improve fidelity and execution time estimates.

From IDs to Semantics: A Generative Framework for Cross-Domain Recommendation with Adaptive Semantic Tokenization

Published:11/11/2025

Cross-Domain Recommendation SystemGenerative Cross-Domain Recommendation FrameworkDomain-Adaptive TokenizationUser Preference ModelingMulti-Domain Joint Training

This paper presents GenCDR, a novel generative crossdomain recommendation framework that overcomes limitations of traditional methods by using domainadaptive tokenization for disentangled semantic IDs, significantly improving recommendation accuracy and generalization across mu

Quantum-Inspired Evolutionary Algorithm for a Class of Combinatorial Optimization

Published:12/1/2002

Quantum-Inspired Evolutionary AlgorithmCombinatorial Optimization ProblemsQ-bit RepresentationSuperposition of StatesKnapsack Problem Optimization

This paper presents a novel QuantumInspired Evolutionary Algorithm (QEA) leveraging principles of quantum computing, showing superior performance in combinatorial optimization problems without premature convergence, even with small populations.

SONIC: Supersizing Motion Tracking for Natural Humanoid Whole-Body Control

Published:11/11/2025

Motion Tracking Foundation ModelNatural Humanoid ControlLarge-Scale Motion Capture DatasetReal-Time Motion PlanningVision-Language-Action Model

The SONIC framework scales model capacity, data, and compute for natural humanoid control, utilizing diverse motioncapture data for dense supervision. It features realtime motion planning and multiinterface support, demonstrating significant performance gains from scaling.

Long-Sequence Recommendation Models Need Decoupled Embeddings

Published:10/3/2024

Long-Sequence Recommendation SystemsDecoupled Attention and Representation EmbeddingsEmbedding Learning in Recommendation SystemsApplication of Attention Mechanism in RecommendationsBehavior Sequence Search and Aggregation

The study highlights the limitations of existing recommendation models using a single embedding for attention and representation. The authors propose the DARE model, which employs decoupled embeddings, improving behavior search accuracy and overall performance on various datasets

屠宰废水处理工程设计与调试

Slaughterhouse Wastewater TreatmentMBBR ProcessConstructed Wetland TreatmentWastewater Treatment Engineering DesignMeat Processing Water Pollutant Discharge Standards

The study presents a wastewater treatment process for slaughterhouse wastewater, incorporating extended hydraulic retention time in the regulating tank to address intermittent discharge issues, ultimately achieving water quality that meets national discharge standards.

FlowRL: Matching Reward Distributions for LLM Reasoning

Published:9/19/2025

RL Training for Large Language ModelsFlow-Balanced Optimization MethodsReward Distribution MatchingKL Divergence-Based Policy OptimizationMathematical reasoning tasks

FlowRL introduces a novel method that matches full reward distributions via flow balancing, enhancing diversity in reasoning. Experiments show FlowRL improves performance by 10% over GRPO and 5.1% over PPO in math tasks, demonstrating the significance of reward distribution match

A Token is Worth over 1,000 Tokens: Efficient Knowledge Distillation through Low-Rank Clone

Published:5/19/2025

Low-Rank Clone ModelKnowledge Distillation for Small Language ModelsEfficient Pre-Training MethodKnowledge Transfer OptimizationActivation Clone Alignment

The study introduces LowRank Clone (LRC), an efficient knowledge distillation method that significantly enhances training efficiency for small language models. By compressing teacher model weights and aligning student activations, LRC achieves comparable performance using only 2

Enhancing Graph Contrastive Learning with Reliable and Informative Augmentation for Recommendation

Published:9/9/2024

Graph Contrastive Learning Enhanced RecommendationDiscrete Code Mapping for Users and ItemsVirtual Neighbor Augmentation StrategyMulti-Level Vector QuantizationNeighborhood Structure-based Contrastive View Generation

The CoGCL framework enhances graph contrastive learning for recommendations by using discrete codes for stronger collaborative information. It employs a multilevel quantizer to map users and items, generating reliable contrastive views through virtual neighbor augmentation and s

Interactive Recommendation Agent with Active User Commands

Published:9/25/2025

Interactive Recommendation System with Active User CommandsNatural Language Commands in Recommendation StrategiesDual-Agent Architecture for Recommendation SystemsSimulation-Augmented Knowledge DistillationReal-Time Policy Adjustment

This paper introduces the Interactive Recommendation Feed (IRF), enabling users to actively adjust recommendations via natural language commands. The developed RecBot employs a dualagent architecture to understand user intent and optimize strategies, significantly enhancing user

Quantum thermodynamics and semi-definite optimization

Published:5/7/2025

Quantum ThermodynamicsSemi-Definite OptimizationFree Energy MinimizationGradient Ascent MethodsHybrid Quantum-Classical Algorithms

The paper addresses energy minimization in quantum thermodynamics by minimizing free energy instead, transforming it into a dual chemical potential maximization problem solved via stochastic gradient ascent methods, ensuring fast convergence. It also applies to classical and hybr

WeatherDiffusion: Weather-Guided Diffusion Model for Forward and Inverse Rendering

Published:8/9/2025

Weather-Guided Diffusion ModelInverse Rendering in Autonomous DrivingSynthetic and Real Weather DatasetsIntrinsic Map-Aware Attention MechanismDiffusion Models in Autonomous Driving

WeatherDiffusion is a novel framework for forward and inverse rendering in autonomous driving, addressing challenges under complex weather and lighting. It introduces an intrinsic mapaware attention mechanism for accurate estimation of material properties, scene geometry, and co

011

Olympian: Scheduling GPU Usage in a Deep Neural Network Model Serving System

Multi-Model GPU SchedulingDeep Neural Network Serving SystemFair Resource SharingTensorFlow-Serving ExtensionGPU Resource Usage Modeling

The paper presents Olympian, which extends TensorFlowServing to enable fair sharing of a single GPU among multiple large DNNs with low overhead, achieving interleaved execution in 12 ms and greatly reducing latency unpredictability.

Gallant: Voxel Grid-based Humanoid Locomotion and Local-navigation across 3D Constrained Terrains

Published:11/19/2025

Voxel-based Humanoid Locomotion PlanningNavigation in 3D Constrained EnvironmentsStructured Perception with LiDAR DataHigh-Fidelity LiDAR SimulationEnd-to-End Optimized Control Policy

Gallant is a voxelgridbased framework for humanoid locomotion and navigation that utilizes voxelized LiDAR data for accurate 3D perception, achieving near 100% success in challenging terrains through endtoend optimization.

Qwen2.5-Omni Technical Report

Published:3/26/2025

Multimodal Large Language ModelThinker-Talker ArchitectureTime-aligned Multimodal RoPEAudio-Video Synchronization ProcessingStreaming Audio Generation

The report presents Qwen2.5Omni, an endtoend multimodal model that perceives text, images, audio, and video while generating text and natural speech in a streaming manner, utilizing interleaved audiovideo sequencing and the ThinkerTalker architecture for optimal performance.

036

Temporally Averaged Regression for Semi-Supervised Low-Light Image Enhancement

Published:6/1/2023

Semi-Supervised Low-Light Image EnhancementMulti-Consistency Regularization LossFeature Dependency Learning in Image SpaceImage Enhancement NetworkProgressive Supervised Loss Function

This study presents a deep learning model that integrates spatial and layerwise dependencies for lowlight image enhancement, addressing the challenges of annotated dataset construction. The incorporation of MultiConsistency Regularization and progressive supervised loss signif

Retinexformer: One-stage Retinex-based Transformer for Low-light Image Enhancement

Published:3/13/2023

Low-Light Image EnhancementRetinex-Based TransformerIllumination-Guided ModelLong-Range Dependency ModelingImage Quality Assessment

This paper introduces a novel onestage Retinex framework (ORF) for lowlight image enhancement. By estimating illumination and restoring corruptions, combined with an IlluminationGuided Transformer (IGT), Retinexformer outperforms stateoftheart methods significantly across b

561 - 580 / 982

Papers