Page 31 - Paper Library - AiPaper

SimpleVLA-RL: Scaling VLA Training via Reinforcement Learning

Published:9/12/2025

Vision-Language-Action ModelReinforcement Learning for Math ReasoningRL Training for Large Language ModelsMulti-Environment RenderingEfficient Reinforcement Learning Framework

SimpleVLARL is introduced to enhance the training of VisionLanguageAction models using reinforcement learning, addressing data scarcity and generalization issues. Results show stateoftheart performance on OpenVLAOFT, reducing reliance on labeled data.

WoW: Towards a World omniscient World model Through Embodied Interaction

Published:9/27/2025

Physical Intuition and Robot InteractionCausal Reasoning Benchmark WoWBenchLarge-Scale World Model TrainingVision-Language Model GuidanceRobot Motion Planning and Execution

The WoW model enhances understanding of physics through embodied interaction, trained on two million robot trajectories. It emphasizes the importance of physical intuition and employs the SOPHIA framework to ensure output's physical realism via visionlanguage model optimization.

Unleashing the Full Potential of Product Quantization for Large-Scale Image Retrieval

Published:11/2/2023

Deep Hashing Framework Based on Product QuantizationLarge-Scale Image RetrievalApproximate Nearest Neighbor SearchImageNet100 DatasetImageNet1K Dataset

This paper introduces a product quantizationbased deep hashing framework that addresses computational costs and accuracy issues in largescale datasets, efficiently learning classlevel codes through a differentiable PQ branch, demonstrating superior retrieval performance across

Lack of TYK2 signaling enhances host resistance to Candida albicans skin infection

Published:12/3/2024

TYK2 Signaling and Fungal InfectionsCandida albicans Skin InfectionNeutrophil Antifungal CapacityTranscriptomic Analysis and Immune ResponseRole of Interferon-gamma in Fungal Dissemination

This study reveals that the absence of TYK2 signaling enhances resistance to Candida albicans skin infections by limiting fungal spread and accelerating wound healing, while also affecting neutrophil antifungal capacity through regulation of interferoninducible genes.

Soil samples from sporotrichosis transmission belt area: Searching for fungal species and their antagonistic activity against Sporothrix brasiliensis

Published:12/1/2022

Soil Samples from Sporotrichosis Transmission BeltInteraction between Purpureocillium lilacinum and Sporothrix braStudy of Fungal Antagonistic ActivityIsolation and Identification of Sporothrix spp.Soil Fungi in Rio de Janeiro, Brazil

This study collected soil samples from sporotrichosis transmission areas in Rio de Janeiro and identified saprophytic fungi, notably Purpureocillium lilacinum, which inhibited the growth of Sporothrix brasiliensis, suggesting potential avenues for developing antifungal agents.

DartControl: A Diffusion-Based Autoregressive Motion Model for Real-Time Text-Driven Motion Control

Published:10/8/2024

Diffusion-based Autoregressive Motion GenerationText-Driven Human Motion GenerationReal-Time Motion ControlSpatially-Constrained Motion GenerationReinforcement Learning Motion Decision Making

DartControl (DART) is a diffusionbased autoregressive motion model enabling realtime textdriven motion control. It overcomes limitations of existing methods by generating complex, longduration movements that effectively integrate motion history and text inputs.

Medical image recognition and segmentation of pathological slices of gastric cancer based on Deeplab v3 + neural network

Published:5/29/2021

Gastric Cancer Pathological Slice Segmentation based on DeeplabGastric Cancer Image RecognitionMulti-Scale Input Neural NetworkMedical Image AnalysisPathological Slice Images

This study presents an automatic segmentation model for gastric cancer slices using a Deeplab v3 network. Tested on 1240 images, it outperforms existing models in key metrics while reducing parameter scale, showing strong clinical applicability.

Use of artificial intelligence and deep learning in fetal ultrasound imaging

Published:11/27/2022

Deep Learning Applications in Fetal Ultrasound ImagingAI Techniques in Medical ImagingUltrasound Imaging Diagnostic Support ToolsFetal Biometry and Anatomy RecognitionIntegration of Deep Learning and Ultrasound Imaging

This review explores deep learning's application in fetal ultrasound imaging, highlighting its potential to enhance diagnostic accuracy affected by operator experience, while covering areas like fetal anatomy identification and biometry measurement.

FlatQuant: Flatness Matters for LLM Quantization

Published:10/12/2024

LLM QuantizationPost-Training Quantization MethodsWeights and Activations FlatteningKronecker Product Matrix OptimizationLLaMA-3-70B Model Evaluation

FlatQuant introduces a new posttraining quantization method that optimizes the flatness of weights and activations, reducing quantization error significantly. It establishes a new benchmark for the LLaMA370B model, achieving less than 1% accuracy drop and up to 2.3x speed impr

A survey on physics informed reinforcement learning: Review and open problems

Published:1/1/2023

Physics-Informed Reinforcement Learning ReviewReinforcement Learning Architecture TaxonomyFusion of Physics Modeling and Reinforcement LearningMethods for Physics Information FusionPhysics-Driven Information Learning

This paper reviews the emerging field of PhysicsInformed Reinforcement Learning (PIRL), introducing a novel taxonomy based on the RL pipeline to better understand current methods and identify key challenges, highlighting the potential of enhancing RL algorithms' applicability an

IUP-BERT: Identification of Umami Peptides Based on BERT Features

Published:11/21/2022

Umami Peptide Prediction Based on BERTDeep Learning Feature ExtractionSupport Vector Machine ModelSynthetic Minority Over-Sampling TechniquePeptide Sequence Prediction

The study presents iUPBERT, a novel umami peptide predictor using BERT for feature extraction. Combined with SMOTE and SVM, it significantly improves the efficiency and accuracy of umami peptide identification, outperforming existing methods, and provides an openaccess web serv

In-depth discovery and taste presentation mechanism studies on umami peptides derived from fermented sea bass based on peptidomics and machine learning

Published:3/16/2024

Molecular Docking of Umami PeptidesFlavor Screening Based on PeptidomicsApplication of Machine Learning in Taste ResearchUmami Peptides from Fermented Sea BassT1R1/T1R3 Binding Site Analysis

This study identified 70 umami peptides from fermented sea bass using peptidomics and machine learning, exploring their binding mechanisms with the T1R1/T1R3 receptor. Key binding sites were identified, offering an efficient screening method for further flavor exploration.

IF-AIP: A machine learning method for the identification of anti-inflammatory peptides using multi-feature fusion strategy

Published:11/18/2023

Anti-Inflammatory Peptide IdentificationMulti-Feature Fusion StrategyVoting ClassifierMachine Learning MethodFeature Selection Algorithm

The study introduces IFAIP, a machine learning model using a voting classifier to identify antiinflammatory peptides (AIPs). This model integrates eight feature descriptors and five conventional classifiers, optimizing performance with feature selection. It significantly improv

PatchWiper: Leveraging Dynamic Patch-Wise Parameters for Real-World Visible Watermark Removal

Published:10/25/2025

Visible Watermark RemovalDynamic Patch ParametersWatermark Segmentation NetworkPixabay Real-world Watermark DatasetMulti-Task Framework

The PatchWiper framework combines an independent watermark segmentation network with a dynamic patch restoration network for effective visible watermark removal. It generates unique parameters for each image patch and leverages a diverse dataset for comprehensive evaluation, show

LLaDA-Rec: Discrete Diffusion for Parallel Semantic ID Generation in Generative Recommendation

Published:11/9/2025

Generative Recommendation SystemsDiscrete Diffusion FrameworkParallel Semantic ID GenerationBidirectional Attention MechanismAdaptive Sequence Generation

LLaDARec is a discrete diffusion framework for generative recommendation, addressing unidirectional constraints and error accumulation. By integrating bidirectional attention and adaptive generation order, it effectively models item dependencies, surpassing existing systems in r

012

Triton-distributed: Programming Overlapping Kernels on Distributed AI Systems with the Triton Compiler

Published:4/28/2025

Distributed AI Systems ProgrammingTriton Compiler ExtensionOverlapping Optimization TechniquesIntegration of Communication PrimitivesHigh-Level Python Programming Model

This paper presents Tritondistributed, an extension of the Triton compiler, addressing programming challenges in distributed AI systems. It supports native overlapping optimizations, integrates OpenSHMEM communication primitives, and achieves complex joint optimizations to enhan

Handling Heavy-tailed Input of Transformer Inference on GPUs

Published:6/16/2022

Efficient Inference of Transformer ModelsGPU Batch Processing OptimizationSelf-Attention Module Parallel StrategiesHandling Variable-Length SequencesPerformance Improvement for NLP Tasks

The study proposes a unified solution to enhance Transformer inference efficiency on GPUs dealing with heavytailed input, reducing redundancy through finegrained and wordaccumulation strategies. Results show a 63.9% latency reduction in the selfattention module and 28.1% in t

Look at the Sky: Sky-aware Efficient 3D Gaussian Splatting in the Wild

Published:3/7/2025

Sky-aware 3D Gaussian SplattingScene Reconstruction in Unconstrained EnvironmentsReal-Time 3D Reconstruction FrameworkNeural Radiance Fields RenderingPseudo Mask Generation and Utilization

This paper introduces a skyaware 3D Gaussian Splatting framework for efficient scene reconstruction from unconstrained photo collections. It leverages a greedy supervision strategy and pseudo masks from a pretrained segmentation network, improving efficiency and rendering quali

DreamText: High Fidelity Scene Text Synthesis

Published:5/23/2024

Scene Text SynthesisHigh Fidelity Text GenerationCharacter-Level Attention MechanismHybrid Optimization StrategyMultifont Learning

DreamText is proposed for highfidelity scene text synthesis, addressing issues in characterlevel guidance, text encoder generalization, and output quality. Utilizing a heuristic alternate optimization strategy and joint training, it enhances attention precision, outperforming s

Estimation and mapping of the missing heritability of human phenotypes

Published:11/12/2025

Estimation and Mapping of Missing Heritability in Human PhenotypAnalysis of Rare Non-Coding Variants ContributionWhole Genome Sequencing Data AnalysisGenetic Assessment of Complex Traits and DiseasesUK Biobank Study Dataset

This study analyzes wholegenome sequencing data from the UK Biobank to quantify the impact of rare noncoding variants on heritability of 34 complex traits. It finds that WGS captures about 88% of narrowsense heritability and identifies significant loci for lipid traits.

601 - 620 / 980

Papers