Zero Trust Security Architecture for AI agents in healthcare, protecting sensitive data from vulnerabilities.

Optimize 3D foundation models with S-VGGT's structure-aware subscene decomposition for enhanced scalability and efficiency.

Zipper-LoRA enhances multilingual speech recognition by dynamically optimizing language-specific and shared model parameters.

SafeLand is a vision-based system for safe autonomous landing of UAVs in dynamic environments without prior information.

MCoT-MVS enhances composed image retrieval by integrating multi-level vision features with multi-modal reasoning for improved semantic accuracy.

Loc3R-VLM enhances Vision-Language Models with advanced 3D understanding for improved spatial reasoning.

GMT is a multimodal transformer that generates realistic 6-DOF object manipulation trajectories for robots in complex 3D environments.

DynaEdit enables versatile, training-free video editing using pretrained models for complex scene interactions.

AHOY reconstructs animatable 3D avatars from occluded YouTube videos using advanced Gaussian splatting techniques.

C-TRAIL transforms dashcam video evidence into legal responsibility assessments using a multimodal approach.

A robust AI framework for legal age estimation using clavicle CT scans, enhancing forensic decision-making.

DexViTac is a portable data collection system that captures high-fidelity visuo-tactile-kinematic demonstrations for improving robot dexterity in contact-rich environments.

ProbeFlow accelerates action decoding in Vision-Language-Action models for responsive robotic control.

M2P enhances visual foundation models for dense point tracking using weakly-supervised learning with video object segmentation.

A fine-grained quantization strategy for large vision language models that enhances accuracy while reducing computational overhead.

TAPESTRY generates high-fidelity turntable videos from 3D models, enabling automated creation of production-ready assets.

VolumeDP enhances robotic manipulation through advanced 3D spatial reasoning for improved imitation learning.

MoBaNet is a parameter-efficient framework for multimodal remote sensing semantic segmentation that balances modality contributions while minimizing computational overhead.

SynRL is a post-training framework that enhances video understanding by teaching models fundamental temporal primitives through synthetic video generation.

A local LLM agent for Linux privilege escalation that achieves high success rates with verifiable rewards.

FINER addresses hallucinations in multimodal large language models through innovative fine-grained negative queries and tuning techniques.

A novel framework for collaborative ranking of scientific papers using LLMs to enhance evaluation accuracy.

PanoVGGT is a Transformer framework for accurate 3D reconstruction from panoramic imagery, leveraging a unique dataset and innovative training strategies.

KA2L is a framework that enhances LLMs' performance through targeted active learning by identifying knowledge gaps.

A novel object detection network that identifies potential objects without relying on external prompts, enhancing flexibility across various applications.

GLIDE is a generative recommender system that enhances podcast discovery by combining semantic reasoning with user context at Spotify.

NEO is a unified language model that enhances recommendation, search, and reasoning across large catalogs with language-steerable capabilities.

AirDDE leverages neural delay differential equations for improved air quality forecasting by integrating delay modeling with physical guidance.

Omni-I2C is a benchmark for evaluating Large Multimodal Models in generating executable code from complex digital graphics.

A targeted fine-tuning approach to reduce hallucinations in large language models by teaching epistemological humility.

A framework that enhances LLMs' critical thinking by teaching them to reconstruct arguments.

A novel approach to improve referring image segmentation for motion-related queries using multimodal learning.

A novel framework leveraging foundation models to enhance material classification accuracy through innovative dataset generation and prior incorporation.

SkeletonLLM translates human skeleton sequences into visual representations for enhanced multimodal reasoning.

EchoGen is a unified framework that enhances layout-to-image generation and image grounding through innovative training strategies.

A training-free framework for continuous and controllable image editing using text embeddings.

LoST revolutionizes 3D shape generation by introducing a semantic tokenization method that enhances autoregressive modeling efficiency.

AdaRadar offers an adaptive compression solution for radar data in autonomous driving, enhancing data transmission efficiency.

MUD optimizes transformer training by enhancing momentum updates for faster convergence.

A framework for optimizing action distributions in robotics models to ensure compliance with complex temporal specifications.

Robust-ComBat enhances diffusion MRI data harmonization by effectively mitigating outlier effects in patients with neurological disorders.

TransText enables high-fidelity, layer-aware text animation for dynamic visual design.

A training-free method for efficient multi-token prediction in large language models that enhances throughput and accuracy.

Aerial RGB-Thermal Segmentation tool enabling enhanced drone-based surveillance and monitoring.

CAT is a framework that enhances creativity in text-to-image models by generating reusable token templates for fuzzy prompts.

Scicode-lint automates the detection of methodology bugs in scientific Python code using AI-generated patterns.

A scalable framework for personalized audio-video generation that allows fine-grained control over identity attributes.

DebugLM enhances LLMs with data provenance for improved debugging and targeted remediation.

A domain-grounded retrieval system that enhances the reliability of LLMs by mitigating hallucinations through a structured verification process.

DiscoGen automates the creation of diverse algorithm discovery tasks to enhance machine learning optimization.

A new instance segmentation strategy for improving microscopy cell segmentation models.

Omni-3DEdit is a unified model for efficient and versatile 3D editing tasks in one pass.

A novel adaptive and robust control solution for robotics using generative models to optimize flow matching.

RPMS enhances LLM-based embodied planning by integrating rule-augmented memory to improve action feasibility and success rates.

FailureMem is a multimodal framework that enhances automated program repair by integrating visual reasoning and reusable knowledge from past failures.

EVA aligns video world models with executable robot actions to enhance robotic control through reinforcement learning.

RangeAD offers an efficient on-model approach for anomaly detection that reduces inference costs while improving performance.

A parameter-efficient fine-tuning approach for billion-parameter vision models to enhance behavior classification in precision livestock farming.

Introducing Knowledge Objects for efficient and accurate memory management in large language models.

CrowdGaussian reconstructs high-fidelity 3D models of human crowds from single images using advanced self-supervised learning techniques.

CoVerRL enhances label-free reinforcement learning by using a generator-verifier co-evolution framework to improve reasoning accuracy.

A training-free framework for detecting deepfakes using evidence-driven reasoning with large vision-language models.

A benchmark for detecting harmful humor across multiple languages and modalities.

PC-CrossDiff enhances 3D visual grounding by improving localization in complex scenes using dual-level differential attention.

A testbed for testing Connected and Autonomous Vehicles in mixed traffic environments using a human-in-the-loop approach.

SARE is a training-free framework that enhances Fine-Grained Visual Recognition by adapting reasoning based on sample difficulty.

AERR-Nav is an adaptive framework for zero-shot object navigation in complex multi-floor environments.

DancingBox is a lightweight motion capture system that enables intuitive character animation using everyday objects and a single webcam.

Anonymization-first framework for LLM trading agents that enhances trustworthiness and performance in portfolio optimization.

FMER is an efficient online RL framework that enhances exploration through principled maximum-entropy optimization.

A novel sampling strategy for YOLO detectors that optimizes training efficiency by selectively resampling images based on learning sufficiency.

Sensi is an LLM agent that learns efficiently in unknown environments through structured test-time learning.

A dual LSTM framework for enhancing the kinematic control of wearable soft robotic arms.

An instance-aware discretization framework that enhances the performance of diffusion models by adapting timestep allocations based on input-dependent priors.

A generative inbetweening method that synthesizes realistic intermediate frames with enhanced semantic and temporal guidance.

A novel framework for enhancing semantic and geometric representations in 3D affordance grounding.

VeriGrey is a grey-box testing framework that uncovers security risks in LLM agents through dynamic prompt mutation.

DSS-GAN introduces a novel conditioning mechanism for enhanced class-conditional image synthesis using a Mamba backbone.

ARES is a novel gradient inversion attack that reconstructs training samples in federated learning without architectural modifications.

A multi-agent framework for PPA-aware RTL code generation that integrates EDA tools for continuous optimization.

A novel framework for efficient open-vocabulary 3D perception and reasoning without scene-specific training.

Dynamic Unreliability-Driven Coreset Selection enhances medical image classification by focusing on informative unreliable samples.

A browser extension that provides contextual help for digital illiterate users by delivering on-demand definitions of technical terms.

Edit-As-Act enables precise 3D indoor scene editing from natural language instructions through goal-regressive planning.

SYRAN offers interpretable unsupervised anomaly detection through symbolic regression, providing human-readable equations for anomaly scoring.

HeiSD accelerates robot control through hybrid speculative decoding for improved inference speeds.

FoMo-X enhances outlier detection models with intrinsic explainability for safer decision-making.

A framework for face anonymization that preserves facial expressions and photometric realism for privacy-sensitive applications.

FrescoDiffusion enables high-quality, coherent video generation from complex images using a novel tiled diffusion approach.

ProGVC is a novel video compression framework that combines progressive transmission and efficient entropy coding for enhanced perceptual quality at low bitrates.

Rel-Zero offers a robust zero-watermarking solution that maintains visual fidelity against AI editing.

AdapTS is a lightweight Teacher-Student framework for efficient multi-class and continual visual anomaly detection optimized for edge deployment.

MM-OVSeg is a multimodal framework for resilient open-vocabulary segmentation in remote sensing, leveraging optical and SAR data.

PCA-Seg enhances open-vocabulary semantic and part segmentation by improving cost aggregation through a parallel structure.

XBridge enhances multilingual capabilities of LLMs by integrating pretrained encoder-decoder translation models for improved performance on low-resource languages.

A pipeline that enables robots to adapt their navigation based on context-aware human preferences using advanced language and reinforcement learning models.

QuantFL is a sustainable federated learning framework that reduces energy costs for IoT devices through efficient model quantization.

IMPACT is an interactive testing framework for vehicle-infrastructure cooperation systems that enhances corner-case generation through human interaction.

A novel dataset and network for robust UAV detection in complex backgrounds.

UniSAFE provides a comprehensive benchmark for evaluating the safety of unified multimodal models across various tasks.

A framework that enhances domain-adaptive learning through beneficial noise in cross-attention mechanisms.

VirPro enhances weakly-supervised monocular 3D detection by integrating visual-referred prompts for improved scene-aware representations.

GuidedSAC enhances reinforcement learning efficiency by integrating LLMs for action-level guidance in continuous control tasks.

AR-CoPO enhances autoregressive video generation by aligning it with contrastive policy optimization for improved quality and generalization.

P$^{3}$Nav is an end-to-end framework that enhances Vision-and-Language Navigation by integrating perception, prediction, and planning.

VLM2Rec enhances multimodal sequential recommendation by balancing modality utilization in Vision-Language Models.

AdaZoom-GUI enhances GUI grounding accuracy through adaptive zoom and instruction refinement for vision-language models.

Baguan-TS is a novel Transformer model that enhances time series forecasting through in-context learning and raw-sequence representation.

FloorPlan-VLN revolutionizes navigation by integrating concise instructions with structured floor plans for enhanced spatial reasoning.

TimeAPN is a novel framework that enhances time series forecasting by addressing non-stationarity through adaptive amplitude-phase normalization.

ECHO synthesizes lifelike avatar head videos with emotionally appropriate and contextually aware facial behaviors.

ASSR-EIC is a novel image compression framework that enables variable-rate extreme image compression with high fidelity and realism.

A lightweight video generation system that adapts temporal attention based on motion content using Stable Diffusion.

A two-stage framework for accurate 3D hand pose estimation using gesture-aware pretraining and token fusion.

VisionNVS revolutionizes Novel View Synthesis for autonomous driving by transforming it into a self-supervised inpainting task.

SCALE is a specialized model for predicting virtual cell responses to perturbations, enhancing both speed and biological accuracy.

An online learning algorithm that enhances data efficiency in reinforcement learning from human feedback by significantly reducing the number of required labels.

StereoWorld is a camera-conditioned stereo world model for efficient stereo video generation.

InfoShot is a shot-aware frame sampler that enhances long-video understanding by intelligently selecting keyframes to retain critical information.

Material Magic Wand simplifies the tedious process of material assignment in 3D modeling by automating part grouping based on material properties.

A novel safety alignment method for large reasoning models that enhances safety decision-making before chain-of-thought generation.

Profiler enhances citation recommendation systems by efficiently capturing human citation patterns without bias.

Reduce code regressions in AI coding agents with Test-Driven Agentic Development using graph-based impact analysis.

VideoAtlas offers a lossless, scalable environment for navigating long-form video using hierarchical grids.

IndicSafe is a benchmark for evaluating the safety of multilingual LLMs in culturally diverse South Asian contexts.

RAMP optimizes large language model inference on resource-constrained devices through adaptive mixed precision quantization.

EditSpilloverProbe leverages edit spillover in image editing models to assess their implicit understanding of world relations.

A novel approach to enhance robustness in iris presentation attack detection using human perceptual priors.

NEVU is a benchmark for actor-conditioned, event-centric human value recognition in news articles, enabling fine-grained evaluation of value cues.

CodeScout enhances developer productivity by using reinforcement learning to optimize code search.

A novel method for generating step-level labels to enhance multi-step reasoning in large language models.

A self-refining neural surrogate model that enhances the simulation of dynamical systems by ensuring long-time consistency.

A predictive framework for assessing long COVID severity in women by integrating clinical and wearable data.

Automate CT report generation using AI-enhanced visual semantic prompting for radiologists.

MALLES is a multi-agent economic simulation framework leveraging LLMs for consumer preference alignment.

A method to enhance local vision-language alignment in few-shot learning for better interpretability in medical diagnosis.

A multi-agent LLM system that fuses data to accurately determine urban building age for energy planning.

CLeAN is an adaptive normalization technique that enhances continual learning in dynamic environments by mitigating catastrophic forgetting.

KineVLA enhances robotic manipulation through a novel vision-language-action framework that integrates detailed kinematic attributes.

A novel framework for multimodal medical imaging that enhances disease recognition by leveraging early intervention techniques.

The Phasor Transformer introduces a novel approach to time-series forecasting by leveraging phase-native representations for efficient token mixing.

T1 is a generative retrieval model that enhances reasoning-intensive retrieval by dynamically generating intermediate reasoning trajectories.

A new dataset for benchmarking 3D reconstruction methods in asset inspection, addressing critical gaps in existing datasets.

Develop a framework that accumulates and reuses executable subagents for self-evolving AI systems.

Automated benchmark generator for scalable software vulnerability detection datasets.

ConGA provides a framework for gender annotation in machine translation to reduce bias and improve accuracy.

A noise-aware anomaly detection framework for securing collaborative DNN inference against misclassification attacks.

A method leveraging pretrained multilingual models to quantitatively measure language distance for linguistic analysis.

A framework for analyzing and optimizing privacy in AI agents using differential privacy techniques.

A study revealing how LLMs compute verbal confidence, enhancing our understanding of model uncertainty.

Automating stage-play layout generation from long-form narratives using advanced language models.

TINA is a novel attack method that uncovers erased concepts in text-to-image diffusion models by leveraging visual-only probes.

This research benchmarks dropout robustness in transformer models, providing insights for selecting models in uncertainty-aware applications.

Develop a universal, prompt-free AI for medical image segmentation.

A systematic approach to designing effective LLM-based embedding pipelines for tabular prediction.

Adaptive Retrieval-Augmented Masked Diffusion enhances QA performance by dynamically calibrating guidance based on retrieved context reliability.

AI system for analyzing coronary angiography videos with video-text context understanding.

Develop an agentic vision-and-language navigation tool for enhanced autonomous systems.

AURORA is a model that predicts tongue displacement in vowel sounds, serving as a didactic tool and biofeedback application.

A novel algorithm for generating informative semi-factual explanations in XAI that enhances user understanding of automated decisions.

A comprehensive benchmark for evaluating AI-generated text detectors across various models and conditions.

TRiMS optimizes reasoning efficiency in language models by minimizing token usage while maintaining accuracy.

A data-driven vehicle dynamics modeling method for distributed electric-drive trucks using Koopman operator theory.

PJB is a reasoning-aware benchmark designed to enhance person-job retrieval systems by diagnosing performance failures.

A novel shape-sensing method for steerable drilling robots using Optical Frequency Domain Reflectometry.

A novel framework for evaluating and mitigating gender bias in machine translation models.

An AI career coach that enhances goal progress through social accountability.

A novel framework for multi-label classification in video capsule endoscopy that tackles class imbalance using advanced attention mechanisms and optimization strategies.

RHYME-XT is a neural operator framework for efficient surrogate modeling of spatiotemporal control systems.

SymPINN leverages symmetry in tensegrity structures to enhance the efficiency and accuracy of physics-informed neural networks for dynamic predictions.

A decentralized algorithm for efficient shape assembly using minimalistic robots.

A toolkit for enhancing the software development lifecycle in RISC-V Trusted Execution Environments for IoT and automotive applications.

A unified multi-modal dataset and machine learning approach for effective network intrusion detection and synthetic data generation.

A comparative study of eye image segmentation using the Segment Anything Model 3 with visual and concept prompts.

WeatherReasonSeg benchmarks the performance of vision-language models under adverse weather conditions to enhance robust reasoning segmentation.

A novel approach to enhance contactless fingerprint spoof detection using paired flash-non-flash imaging.

This paper presents a novel approach to optimize multigrid methods using evolutionary algorithms and genetic programming.

rSDNet offers a robust neural learning framework to combat label noise and adversarial attacks in classification tasks.

Complementary RL enhances sample efficiency in LLM-based agents by optimizing experience extraction alongside policy learning.

A framework for monitoring evolving narratives in crisis information environments using semantic embeddings and clustering.

AdaMuS is a framework that addresses dimensional imbalances in multi-view learning through adaptive sparsity and self-supervised learning.

A data-driven framework for predicting urban airflow and pollutant dispersion to aid decision-makers in environmental management.

A method to recover latent actions and dynamics from offline trajectories using demonstrator diversity.

Emergent Trust Learning enhances AI agents' cooperation in competitive environments through a lightweight trust-based control algorithm.

A framework for estimating time-varying reproduction numbers in infectious disease surveillance using conditional inverse learning.

This research explores the impact of video-based fine-tuning on multimodal large language models, revealing trade-offs in visual performance.

This study benchmarks Neural Operators for modeling excitable cell dynamics in the FitzHugh-Nagumo model.

This research proposes adaptive network coding to enhance communication reliability in multi-robot systems operating over wireless channels.

A novel DDH-based technique for reducing key sizes in multi-party Function Secret Sharing schemes.

MSDN++ enhances zero-shot learning by distilling intrinsic semantic representations through mutually causal attention.

A framework for improving causal representation learning through enhanced dataset evaluation and reproducibility.

SafeTutors benchmarks the pedagogical safety of AI tutoring systems to enhance learning outcomes.

Introducing a novel token pruning technique for enhancing efficiency in vision-language models for video tasks.

Motion-MLLM enhances spatial reasoning in 3D scenes using egomotion data for improved efficiency.

LaDe enables the creation of fully editable layered design documents from natural language prompts.

A framework for rapid adaptation in reinforcement learning using shared low-dimensional goal embeddings.

CARE enhances multi-head latent attention conversion for improved inference efficiency.

A unified framework for optimizing text-guided locomotion in humanoid robots through physics-based motion generation.

A novel approach to compress large language models by focusing on the relative rank of weights.

This research identifies a Noise Sensitivity Exponent that links noise robustness and computational hardness in high-dimensional learning models.

A comprehensive analysis of hardware reverse engineering research to enhance security practices.

A machine learning approach to enhance the reliability of seismic and volcanic monitoring through improved signal interpretation.

ChopGrad introduces a memory-efficient method for fine-tuning video diffusion models using truncated backpropagation.

A novel architecture to enhance multi-agent workflows with governed memory.

A gastrointestinal video analysis pipeline using ResNet-50 for multi-label classification.

A novel algorithm for single-pass alignment of multi-talker recordings using finite-state automata.

A novel approach to monitor cognitive workload using facial movement dynamics from standard webcams.

A comprehensive review of synthetic data and simulation technologies for advancing autonomous driving.

A framework for identifying undervalued football players using market dynamics and news signals.

REAL is an advanced framework for reliable legged parkour that adapts to sensory corruption.

A rigorous benchmarking framework for evaluating Reinforcement Learning algorithms through controlled environments.

LoGSAM is a parameter-efficient framework for MRI tumor segmentation using radiologist dictation as prompts.

ECKConv introduces a novel kernel architecture for efficient SE(3) equivariant learning in point cloud tasks.

A physics-informed deep learning framework for predicting permeability tensors from porous media microstructure images.

L2A introduces a conditional memory access layer for long-context language models, optimizing attention usage and improving training efficiency.

An AutoML approach that optimizes wireless beamforming and waveforms using interpretable deep unfolding techniques.

ZipServ optimizes LLM inference through innovative lossless compression techniques tailored for GPU architectures.

A framework for proactive knowledge inquiry in doctor-patient dialogues to enhance EMR generation.

GMFlow is a physics-inspired framework for rapid generation of realistic ground-motion time histories for earthquake hazard analysis.

This paper explores a generative model for 3D avatar sign language motion using phonological attributes.

A novel framework for generative causal models that addresses global counterfactuals using sheaf theory.

This paper analyzes jailbreak behavior in vision-language models and proposes a defense method to enhance safety.

ShapleyLaw introduces a game-theoretic approach to optimize multilingual pretraining by quantifying cross-lingual transfer effects.

A theoretical framework for optimizing multi-armed sequential hypothesis testing through betting strategies.

SpiderCam is a low-power FPGA-based camera that captures real-time depth maps using differential defocus.

This paper presents a theoretical framework for optimizing Markov decision processes using linear operators.

This survey provides a comprehensive overview of video understanding models and their evolution.

This paper explores a method to enhance video generation quality using Massive Activations in video diffusion transformers.

A framework for discovering functional modules in Large Language Models to enhance interpretability.

A novel approach to federated reinforcement learning that enhances safety by preserving distributional information.

This paper explores the relationship between attention sinks and gradient sinks in Transformer models during training.

This paper presents a novel data obfuscation technique for securing classical values in quantum computation.

This research investigates how large language models encode semantic relationships through probing and interpretability techniques.

A novel framework for efficient sampling from Boltzmann distributions using a one-step neural generator.

This paper theoretically analyzes the structural benefits of graph transformers over traditional graph convolutional networks.

This paper explores the consistency of the k-nearest neighbor regressor in complex survey designs, addressing a gap in existing literature.

This paper explores a novel approach to learning Q-value functions for efficient planning in reinforcement learning.

This paper explores the limitations of PCA in estimating design parameters for CAD applications.

This paper presents a generalized Mirror Descent method for optimization on Riemannian manifolds.

UniSem enhances 3D reconstruction accuracy and semantic generalization from sparse images using innovative error-aware techniques.

A framework for proving authorship of AI-generated content using cryptographic methods.

This paper explores how transformer LMs learn linguistic categories, shedding light on language acquisition processes.

FACE-net enhances emotional video captioning by calibrating factual content and augmenting emotional cues.

A framework for token-level attribution in multi-agent language systems to enhance accountability.

SHIFT introduces a novel fine-tuning framework to enhance motion fidelity in video diffusion models.

This paper surveys the transition from digital twins to world models for enhancing edge general intelligence in 6G networks.

A theoretical exploration of self-bootstrapping coding agents based on specifications.

A theoretical exploration of the limits of causal interventions in generative models.

A theoretical framework for understanding and designing internal noise mechanisms in deep learning.