Page 26 - Paper Library - AiPaper

The paper introduces UniDex, a novel approach to inverted indexing that employs unified semantic modeling to enhance retrieval. Featuring components UniTouch and UniRank, it significantly improves semantic generalization and retrieval effectiveness, validated in Kuaishou's short

RNA stability enhancers for durable base-modified mRNA therapeutics

Published:11/7/2025

RNA Stability Enhancement for TherapeuticsmRNA TherapeuticsViral Sequence ScreeningStability Enhancing ElementsRNA Modifications and Translation

This study identifies 11 RNA stability enhancers from screening 196,277 viral sequences, improving mRNA stability and translation by extending the poly(A) tail via TENT4 recruitment. Element A7 shows exceptional robustness, achieving stability comparable to circRNA with enhanced

010

MiroThinker: Pushing the Performance Boundaries of Open-Source Research Agents via Model, Context, and Interactive Scaling

Published:11/14/2001

MiroThinker v1.0, an opensource research agent, enhances reasoning and information retrieval through interaction scaling, unlike traditional model expansions. Utilizing systematic training, it achieves significant performance boosts in benchmarks, highlighting interaction depth

Recent Advances in Speech Language Models: A Survey

Published:10/2/2024

Speech Language ModelsAutomatic Speech Recognition (ASR)Text-to-Speech (TTS)End-to-End Speech GenerationSpeech Model Evaluation Metrics

This survey provides a comprehensive overview of recent methodologies for constructing Speech Language Models (SpeechLMs), emphasizing their advantages as endtoend models that generate speech directly, overcoming challenges like information loss, latency, and error accumulation

MUSS-TI: Multi-level Shuttle Scheduling for Large-Scale Entanglement Module Linked Trapped-Ion

Published:9/30/2025

Large-Scale Trapped-Ion Quantum ComputingQuantum Circuit CompilerMulti-Level Scheduling MethodQuantum Charge-Coupled Device (QCCD)Quantum Operation Efficiency Optimization

MUSSTI is a scalable compiler designed for largescale trappedion quantum architectures, utilizing multilevel scheduling to reduce shuttling overhead in Entanglement Module Linked QCCD systems. It achieves up to 73.38% reduction in shuttling operations, enhancing quantum opera

AnyBimanual: Transferring Unimanual Policy for General Bimanual Manipulation

Published:12/9/2024

Bimanual Policy TransferUnimanual Policy GeneralizationBimanual Manipulation ModelHigh-Dimensional Action SpaceMachine Learning Application

AnyBimanual is introduced to transfer unimanual policies to general bimanual manipulation using minimal demonstration data. It employs a Skill Manager and a Visual Aligner, achieving a 17.33% success rate increase in simulated tasks and 84.62% in realworld tasks.

Mooncake: A KVCache-centric Disaggregated Architecture for LLM Serving

Published:6/24/2024

LLM Serving ArchitectureKVCache-Centric Scheduling StrategyDisaggregated Cache SystemLoad Prediction and Rejection PolicyLong-Context Processing Optimization

Mooncake features a KVCachecentric disaggregated architecture that significantly enhances effective throughput for LLM serving. By separating prefill and decoding stages and utilizing idle GPU cluster resources, it achieves up to a 525% increase in throughput in longcontext sce

A physics-informed transformer neural operator for learning generalized solutions of initial boundary value problems

Published:12/12/2024

Physics-Informed Neural OperatorInitial Boundary Value Problem SolvingNonlinear Partial Differential EquationsCross-Attention MechanismSimulation-Free Training

This paper introduces PINTO, a physicsinformed transformer neural operator for solving initial boundary value problems. It efficiently generalizes to unseen conditions using only physics loss in a simulationfree setting, enhancing solution accuracy and efficiency.

Semi-Supervised Subspace Clustering via Tensor Low-Rank Representation

Published:5/21/2022

Semi-Supervised Subspace ClusteringLow-Rank Tensor RepresentationAffinity Matrix LearningLaplacian Graph RegularizationMulti-Benchmark Dataset Experiments

This paper introduces a novel semisupervised subspace clustering approach that constructs a discriminative affinity matrix while enhancing supervisory information via a global lowrank constraint on a stacked tensor. Local geometric structures further improve affinity learning,

其他文献.pdf

Social Media Algorithm CensorshipTwitter Shadow Banning StudyText Analysis and Machine Behavior FeaturesPlatform Auditing and Accountability MechanismsImpact of Algorithms on Online Attention

This study audits Twitter's phenomenon of shadowbanning, analyzing algorithms' roles in directing online attention. Testing 25,000 U.S. Twitter accounts revealed shadowbanning is rare; botlike accounts are more affected, while verified ones are less so, particularly those postin

Tokenize Once, Recommend Anywhere: Unified Item Tokenization for Multi-domain LLM-based Recommendation

Published:11/17/2025

Multidomain LLM-based Recommendation SystemsUnified Item Tokenization FrameworkMixture-of-Experts ArchitectureMutual Information Calibration MechanismDomain-Specific Knowledge Capture

The paper introduces , a unified item tokenization framework that addresses the limitations of separate models for different item domains in LLMbased recommendation systems. It employs a mixtureofexperts architecture and a mutual information calibration mechanism to pr

015

ThinkBot: Embodied Instruction Following with Thought Chain Reasoning

Published:12/12/2023

Instruction Completion Based on Large Language ModelsThought Chain ReasoningAction Planning in Human-Agent Cooperative EnvironmentsComplex Goal CompletionObject Localization and Interaction

ThinkBot addresses Embodied Instruction Following by using Chain of Thought reasoning to fill in missing action steps in human commands, enhancing coherence. It leverages a Large Language Model for instruction completion and a multimodal Transformer for accurate object localizati

Unified Multimodal Understanding and Generation Models: Advances, Challenges, and Opportunities

Published:5/5/2025

Multimodal Understanding and Generation ModelsGenerative Adversarial ModelsFusion of Diffusion and Autoregressive ModelsText-to-Image GenerationMultimodal Datasets and Benchmarks

This paper provides a comprehensive survey on unified multimodal understanding and generation models, exploring the challenges posed by architectural differences between autoregressive and diffusion models. It highlights three main unified framework paradigms and offers tailored

LucidFusion: Reconstructing 3D Gaussians with Arbitrary Unposed Images

Published:10/21/2024

3D Reconstruction with Relative Coordinate MapUnposed Image Reconstruction MethodsLucidFusion FrameworkHigh-Quality 3D Object GenerationDifferentiable Rasterization Techniques

LucidFusion redefines 3D reconstruction as imagetoimage translation using Relative Coordinate Maps, eliminating camera pose reliance. The extension to Relative Coordinate Gaussians ensures consistent geometry and pose recovery, enabling robust 3D reconstructions from arbitrary

ViewCrafter: Taming Video Diffusion Models for High-fidelity Novel View Synthesis

Published:9/4/2024

video diffusion modelsHigh-Fidelity Novel View SynthesisPoint-Based RepresentationCamera Trajectory Planning3D Reconstruction and Synthesis

This study introduces , a method that synthesizes highfidelity novel views from single or sparse images using video diffusion models, overcoming the dependence on dense multiview captures. It incorporates coarse 3D clues and camera pose control while featuring an i

Efficient, continuous mutagenesis in human cells using a pseudo-random DNA editor

Continuous Mutagenesis TechnologyGene Editing in Human CellsT7 Polymerase-Driven EditingCell Viability ScreeningMutant Screening Methods

This study introduces TRACE, a novel method for continuous targeted mutagenesis in human cells, achieving high mutagenesis rates over multiple generations and identifying mutations correlated with MEK1 inhibitor resistance.

Self-supervised Graph Learning for Recommendation

Published:10/21/2020

Self-Supervised Graph Learning for RecommendationLong-Tail Item RecommendationGraph Convolution Network OptimizationRobustness Against Interaction NoisesUser-Item Graph Representation Learning

This paper introduces Selfsupervised Graph Learning (SGL) to address GCN limitations in recommendation systems, particularly improving longtail item recommendation and noise robustness by generating multiple views through selfsupervised tasks.

$π^{*}_{0.6}$: a VLA That Learns From Experience

Published:11/19/2025

Vision-Language-Action ModelRL Training for Large Language ModelsExperience-Based Reinforcement LearningRobotic Data Collection and OptimizationAdvantage-Conditioned Policies

The study presents RECAP, a method for training VisionLanguageAction models through realworld learning. The

π^{}{0.6}

model, pretrained using offline reinforcement learning, demonstrates significant performance improvements on various tasks, such as laundry folding and es

013

Seer: Online Context Learning for Fast Synchronous LLM Reinforcement Learning

Published:11/19/2025

Online Context LearningSynchronous LLM Reinforcement LearningLong-Tail Latency OptimizationDynamic Load BalancingAdaptive Grouped Speculative Decoding

Seer is a novel online context learning system that optimizes synchronous reinforcement learning for large language models by addressing performance bottlenecks in the rollout phase. It significantly improves throughput by 74% to 97% while reducing longtail latency and enhancing

Are Graph Augmentations Necessary? Simple Graph Contrastive Learning for Recommendation

Published:12/16/2021

Graph Contrastive Learning for RecommendationUser-Item Bipartite Graph AugmentationSelf-Supervised Signal ExtractionData Sparsity IssueEnhancement of Recommendation Accuracy

This study reveals that graph augmentations are unnecessary in contrastive learning for recommendations. The performance boost comes from the uniformity of representations. The proposed SimGCL uses uniform noise instead of complex augmentations, improving accuracy and efficiency.

501 - 520 / 980

Papers