Page 24 - Paper Library - AiPaper

Killing Two Birds with One Stone: Unifying Retrieval and Ranking with a Single Generative Recommendation Model

Published:4/23/2025

Generative Recommendation SystemsUnified Generative Recommendation FrameworkRetrieval and Ranking in Recommendation SystemsInformation Sharing and OptimizationDynamic Balancing Optimization Mechanism

The study introduces the Unified Generative Recommendation Framework (UniGRF) to unify retrieval and ranking stages in recommendation systems, enhancing performance through information sharing and dynamic optimization, outperforming existing models in extensive experiments.

DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning

Self-Verifiable Mathematical ReasoningLLM-based Theorem ProvingReinforcement Learning for Math ReasoningProof Generator and VerifierQuantitative Reasoning Capability Enhancement

The DeepSeekMathV2 model addresses the effectiveness of large language models in mathematical reasoning. By training a theorem prover verifier, it enables selfverification, producing more accurate proofs and achieving excellent results in competitions, demonstrating the potenti

042

MEGCF: Multimodal Entity Graph Collaborative Filtering for Personalized Recommendation

Published:6/13/2022

Multimodal Entity Graph Collaborative FilteringPersonalized Recommendation SystemMultimodal User Preference ModelingDeep Feature ExtractionE-commerce Recommendation

The MEGCF model addresses the mismatch between multimodal feature extraction and user interest modeling by extracting semantic entities, constructing a useritem graph, and employing a sentimentweighted graph convolution network to enhance recommendation accuracy.

Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM

Published:4/9/2021

RL Training for Large Language ModelsLarge Language Model Fine-TuningTransformer-Based Efficient Forward PredictionGPU Cluster TrainingPipeline Parallel Training

This paper introduces a novel interleaved pipeline parallelism schedule, combining tensor, pipeline, and data parallelism, to enhance the training efficiency of large language models on GPU clusters, achieving 502 petaFLOP/s on 3072 GPUs with over 10% throughput improvement.

RWKV-7 "Goose" with Expressive Dynamic State Evolution

Published:3/19/2025

RWKV ArchitectureLanguage Model Performance EvaluationMultilingual TasksOpen Source Pre-trained DatasetState Tracking and Language Recognition

RWKV7 "Goose" is a novel sequence modeling architecture that achieves constant memory usage and inference time. This 2.9 billion parameter model sets new stateoftheart performance on multilingual tasks and matches existing benchmarks in English, while introducing generalized

011

Semantics Meet Signals: Dual Codebook Representationl Learning for Generative Recommendation

Published:11/15/2025

Generative Recommendation SystemsUniversal Recommendation SystemsDual Codebook Representation LearningBalancing Collaborative Filtering and Semantic UnderstandingFlexCode Framework

This paper introduces FlexCode, a dual codebook framework that balances collaborative filtering and semantic understanding in generative recommendation systems, enhancing performance with adaptive token allocation and a lightweight mixtureofexperts model, outperforming strong b

RoboFAC: A Comprehensive Framework for Robotic Failure Analysis and Correction

Published:5/18/2025

Robotic Failure Analysis and Correction FrameworkVision-Language-Action ModelTask Understanding for Failure CorrectionRoboFAC DatasetOpen-World Robotic Manipulation

The RoboFAC framework enhances robotic failure analysis and correction for VisionLanguageAction models in openworld scenarios. It includes a large dataset and a model capable of task understanding, with experimental results showing significant performance improvements.

Adversarial Data Collection: Human-Collaborative Perturbations for Efficient and Robust Robotic Imitation Learning

Published:3/15/2025

Adversarial Data CollectionHuman-Collaborative PerturbationsRobotic Imitation LearningData Efficiency in ExperimentsRobust Task Performance

The paper introduces the Adversarial Data Collection (ADC) framework, emphasizing data quality over quantity in robotic imitation learning. By leveraging realtime humanrobot interactions and collaborative perturbations, ADC enhances task performance and robustness with minimal

AHA: A Vision-Language-Model for Detecting and Reasoning Over Failures in Robotic Manipulation

Published:10/1/2024

Vision-Language ModelsFailure Detection in Robotic ManipulationAHA DatasetTask and Motion PlanningNatural Language Failure Reasoning

AHA is an opensource visionlanguage model designed for detecting and reasoning about failures in robotic manipulation using natural language. By framing failure detection as freeform reasoning, it provides adaptable explanations across various tasks, demonstrating strong effec

Robotic World Model: A Neural Network Simulator for Robust Policy Optimization in Robotics

Published:1/17/2025

Robotic World ModelAutoregressive MechanismLong-Horizon PredictionModel-Based Reinforcement LearningSelf-Supervised Training

This paper presents a novel framework for a robotic world model using dualautoregressive mechanisms and selfsupervised training, enabling reliable longhorizon predictions without domainspecific biases. It supports efficient policy optimization and seamless deployment in real

Agility Meets Stability: Versatile Humanoid Control with Heterogeneous Data

Published:11/22/2025

Humanoid Robot Dynamic Tracking and Balance ControlHeterogeneous Data-Driven Control FrameworkHybrid Reward SchemeHuman Motion Capture DatasetUnified Agility and Stability

The AMS framework unifies dynamic motion tracking and extreme balance maintenance for humanoid robots, leveraging heterogeneous data sources. A hybrid reward scheme and adaptive learning strategy enhance training efficiency, demonstrating potential for diverse motion applications

ggy423.pdf

Deep Neural Network Seismic Arrival Time PickingAutomated Seismic Phase PickingSeismic Monitoring and AnalysisSeismic Sensor Data ProcessingSeismic Phase Identification Technology

PhaseNet, a deep learning method, significantly improves the accuracy of seismic P and S wave arrival time picking compared to existing automated methods, addressing the challenges posed by increasing sensor numbers and aiding earthquake monitoring and localization.

Thinking in 360°: Humanoid Visual Search in the Wild

Published:11/25/2025

Humanoid Visual Search360° Panoramic Image ProcessingVisual-Spatial Reasoning CapabilityH* Bench BenchmarkOpen-Source Model Optimization

The study introduces humanoid visual search (HVS), where agents use head movements in immersive 360° images. The new benchmark H Bench emphasizes advanced visualspatial reasoning. Experiments reveal low success rates for top models, though posttraining significantly enhances p

Learning to Execute

Published:10/17/2014

LSTM TrainingSequence-to-Sequence LearningProgram EvaluationNew Variant of Curriculum LearningCharacter-Level Representation Mapping

The paper demonstrates that LSTMs can learn to evaluate simple computer programs in a sequencetosequence framework, mapping characterlevel representations to outputs. A new curriculum learning strategy significantly improved performance, achieving 99% accuracy in adding two 9

Black-Box On-Policy Distillation of Large Language Models

Published:11/14/2025

Black-Box Policy DistillationGenerative Adversarial DistillationLLM DistillationGenerator-Based Feedback MechanismStudent LLM Training

This study introduces Generative Adversarial Distillation (GAD) for extracting knowledge from a teacher LLM in a blackbox setting. By framing a minimax game, it trains a discriminator that evolves with the student, offering onpolicy feedback, outperforming traditional sequence

Scaling Latent Reasoning via Looped Language Models

Published:10/30/2025

Looped Language ModelsLatent Space ReasoningPre-Training Model OptimizationEntropy RegularizationLarge-Scale Language Model Training

This study introduces Looped Language Models (LoopLM), embedding reasoning in the pretraining phase through iterative latent space computation. The Ouro model, trained on 7.7T tokens, outperforms SOTA LLMs with 12B parameters, showing more consistent reasoning traces.

030

SCoTER: Structured Chain-of-Thought Transfer for Enhanced Recommendation

Published:11/24/2025

LLM-based Recommendation SystemsStructured Chain-of-Thought TransferPattern Discovery MechanismEfficient Model Structure IntegrationOnline Recommendation System Optimization

SCoTER is a framework designed to enhance recommendation systems by efficiently integrating Large Language Models' reasoning capabilities. It addresses key challenges through automated pattern discovery and structurepreserving integration, demonstrating improved performance thro

Graph Segmentation and Contrastive Enhanced Explainer for Graph Neural Networks

Published:4/11/2025

Graph Neural Network ExplainersGraph Segmentation TechniquesContrastive Learning Enhanced ExplainersGraph-Structured Data ModelingModel Interpretability

This paper introduces a novel GNN explainer that combines graph segmentation and contrastive learning, enhancing explanation fidelity by distinguishing between explanatory and redundant subgraphs. Extensive experiments validate its effectiveness in graph and node classification t

Physics informed WNO

Published:10/30/2023

Physics-Informed Neural NetworksWavelet Neural OperatorOperator LearningStochastic ProjectionPhysics-Informed Learning

This paper introduces a Physicsinformed Wavelet Neural Operator (PIWNO) to address the datahungry nature of traditional PDE solutions. This method effectively learns solution operators for parametric PDE families without labeled data, validated through four relevant nonlinear s

LightRAG: Simple and Fast Retrieval-Augmented Generation

Published:10/8/2024

Retrieval-Augmented Generation (RAG)Graph-Based Text Indexing and RetrievalDual-Level Retrieval SystemIncremental Update AlgorithmOpen-Source Retrieval System

LightRAG is a novel RetrievalAugmented Generation (RAG) system that integrates graph structures and a duallevel retrieval system to enhance comprehensive information retrieval. It utilizes an incremental update algorithm for efficient, contextually relevant responses.

461 - 480 / 982

Papers