Adapt existing visual geometric transformers for enhanced RGB-Thermal 3D reconstruction.

Optimize video question answering efficiency with HORNet's advanced frame selection for vision-language models.

EdgeCrafter offers a compact ViT framework for efficient dense prediction on edge devices, outperforming traditional models.

MeInTime is a diffusion-based face restoration method that enhances identity fidelity and age consistency using cross-age references.

CubiD is a novel discrete generation model that enhances visual generation using high-dimensional representation tokens.

A mechanistic study of Vision-Language-Action models revealing insights into multimodal input translation for robotics.

SAMA revolutionizes instruction-guided video editing by factorizing semantic anchoring and motion alignment for superior performance.

F2LLM-v2 offers efficient multilingual embeddings with state-of-the-art performance for over 200 languages.

Nemotron-Cascade 2 is a compact yet powerful open-weight LLM that excels in reasoning and agentic tasks, backed by a robust training methodology.

DreamPartGen enables semantically grounded, part-aware text-to-3D generation for coherent and interpretable 3D object synthesis.

Splat2BEV enhances autonomous driving perception by integrating explicit 3D reconstruction for superior BEV representation.

A mechanistic interpretability approach to enhance generalization in Vision-Language-Action models for robot manipulation.

ARIADNE is a novel framework for reliable coronary angiography analysis that enhances vessel segmentation through preference-aligned perception and reasoning.

JAMA optimizes multimodal jailbreaks for spoken language models, enhancing security against adversarial prompts.

TAU-R1 is a vision-language model designed to enhance traffic anomaly understanding using a specialized dataset and innovative training strategies.

A benchmark dataset and network for accurate building change detection using multi-modal RGB-NIR imagery.

CAMO is a conditional neural solver that optimizes multi-agent coordination for complex multi-objective traveling salesman problems.

TerraScope revolutionizes earth observation with pixel-grounded visual reasoning for enhanced geospatial analysis.

ATG-MoE revolutionizes robotic assembly by integrating multi-modal inputs for efficient trajectory generation and skill learning.

A novel segmentation method leveraging a single segmentation token in Multi-modal Large Language Models for enhanced object-level segmentation.

MultiTempBench is a multilingual benchmark for evaluating temporal reasoning in large language models across various languages and calendar systems.

Hypothesis-Conditioned Query Rewriting enhances decision-making in retrieval-augmented generation by refining query strategies for better evidence retrieval.

CRAFT is a lightweight fine-tuning paradigm for diffusion models that drastically reduces data requirements while enhancing efficiency.

PRIOR is an efficient framework for humanoid locomotion that achieves robust terrain traversal with human-like gaits using a simple design.

PromptHub enhances visual in-context learning through locality-aware fusion and alignment techniques.

DriftGuard is a federated continual learning framework that efficiently mitigates asynchronous data drift in resource-constrained environments.

RewardFlow enhances agentic reasoning in large language models through topology-aware reward propagation on state graphs.

A novel denoising network that enhances high-resolution transmission electron microscopy imaging by effectively reducing noise while preserving atomic details.

Agent Control Protocol provides a robust governance framework for autonomous agents in B2B environments through cryptographic admission control.

V-Dreamer automates the creation of diverse robotic simulation environments and trajectories from natural language, enhancing robot training efficiency.

Perceptio enhances vision language models with spatial reasoning through innovative token generation.

Mi:dm K 2.5 Pro is a flagship LLM optimized for enterprise-grade complexity and Korean-language understanding.

ProCal is a novel probability calibration method for enhancing source-free domain adaptation by balancing knowledge retention and local noise mitigation.

NeuroGame Transformer redefines attention mechanisms using game theory and statistical physics for improved performance in NLP tasks.

CausalRM offers a scalable solution for reward modeling in RLHF using observational user feedback to improve alignment of language models.

MemMA is a multi-agent framework that enhances memory management in LLMs through coordinated reasoning and self-evolution.

SCNP enhances image segmentation by improving topology accuracy through neighbor pixel penalization.

A framework that enhances geometric reasoning in MLLMs through strategic visual aid construction and optimization.

Switch is a novel semi-supervised learning framework for robust ultrasound image segmentation that outperforms existing methods with limited labeled data.

A benchmarking framework for evaluating PDF parsers using LLM-based semantic evaluation to enhance table extraction accuracy.

Click-to-Ask is an AI assistant that enhances live streaming commerce by providing real-time responses and generating promotional copy.

PhysVideo generates physically plausible videos by leveraging cross-view geometry guidance for enhanced realism.

MOSAIC is a multi-objective framework for optimizing dataset curation to enhance AI safety and instruction following.

AFS-Search revolutionizes Text-to-Image generation with a training-free, closed-loop framework that enhances performance and speed.

Matryoshka Gaussian Splatting enables continuous level of detail rendering for 3D scenes without sacrificing quality.

MonoArt offers a unified framework for efficient and accurate 3D reconstruction of articulated objects from single images.

A novel framework that enhances motion generation by integrating semantic and kinematic conditions through a diffusion-based discrete motion tokenizer.

NavTrust is a benchmark for evaluating and enhancing the trustworthiness of embodied navigation systems under realistic conditions.

EffectErase offers a novel solution for high-quality video object removal and insertion using a comprehensive dataset and advanced learning techniques.

RPiAE enhances image generation and editing by improving reconstruction fidelity through a novel representation-based tokenizer.

A lightweight prompt-tuning method to enhance attention allocation in Vision-Language Models for improved visual reasoning.

FASTER enhances real-time responsiveness in Vision-Language-Action models by optimizing action sampling for immediate reactions.

A challenge evaluating the privacy resilience of synthetic tabular data generated by diffusion models against membership inference attacks.

FLAC is a few-shot acoustic synthesis method that generates room impulse responses using a probabilistic flow-matching approach.

A decentralized MPC framework for safe trajectory planning in multi-robot quadrupedal systems.

MAPG enables robots to convert natural language commands into actionable decisions in 3D environments through probabilistic grounding.

A unified framework for stable and target-faithful image generation in low-density regions using adaptive prompt blending.

ADAPT is a training-free framework that enhances text-to-image synthesis by providing deterministic control over rare concept generation.

D5P4 enhances discrete diffusion decoding by improving candidate diversity through a novel beam-search framework.

A novel approach to enhance continual representation learning using a data-guided Random Projection Layer.

An adaptive stock price prediction framework that utilizes autoencoders and dual node transformers to enhance accuracy during volatile market conditions.

M is an open-source social robot platform that simplifies the reproduction and modification of robots for research and real-world applications.

A novel autoregressive model for generative image classification that outperforms diffusion models in efficiency and accuracy.

CustomTex enables high-fidelity, customizable 3D indoor scene texturing using reference images for precise instance-level control.

FedTrident enhances road condition classification in federated learning by effectively countering targeted label-flipping attacks.

DaPT is a novel multilingual framework that enhances retrieval-augmented generation for multi-hop question answering across languages.

A graph neural network that enhances robot learning by incorporating dynamics properties for improved policy efficiency.

DROID-SLAM optimizes real-time 3D mapping for autonomous drones navigating complex environments.

Fire as a Service enhances robot simulators with accurate fire dynamics for safer firefighting training.

SignAgent leverages LLMs for efficient and linguistically-grounded sign language annotation and dataset curation.

Em-Garde is a framework that enhances proactive video understanding by efficiently matching user queries to visual proposals.

SwiftTailor revolutionizes 3D garment generation with a fast and efficient geometry image representation.

FUMO leverages prior modulation in diffusion models for effective single image reflection removal.

A novel framework for debiasing vision-language models using sparse embedding modulation.

Stability Monitor provides a black-box solution for monitoring the behavioral consistency of AI model endpoints.

GenHOI offers a novel framework for accurate hand-object pose estimation under occlusion using hierarchical semantic knowledge.

AgentDS benchmarks AI agents against human performance in domain-specific data science tasks.

A GAN-based framework to analyze and mitigate racial bias in predictive policing systems using multi-city crime data.

A supervised ML framework for reliable anomaly detection in power plant monitoring that balances performance, interpretability, and fairness.

Context Bootstrapped Reinforcement Learning enhances exploration efficiency in RL by integrating few-shot demonstrations into training.

A novel unsupervised contrastive learning approach for efficient and robust 3D shape matching.

Lightweight model predictive control solutions for real-time spacecraft attitude synchronization.

Create realistic hand-object interaction simulations using RGB videos with GHOST's Gaussian Splatting technique.

A Semantic-Augmented DRL framework that enhances UAV deployment strategies in VANETs by integrating LLMs for improved connectivity.

FEDB revolutionizes confidential databases by eliminating cryptographic overhead for secure queries in cloud environments.

A multi-stage classification framework for detecting human values in noisy Russian social media text using LLMs.

dTRPO enhances policy optimization in diffusion large language models for improved efficiency and performance.

A robust watermarking framework for large language models that ensures ownership protection without compromising performance.

Weaver is a greybox fuzzing framework that uncovers vulnerabilities at the JavaScript-WebAssembly boundary.

Points-to-3D leverages point cloud priors for enhanced 3D asset and scene generation.

AutoPipe optimizes LLM post-training configurations efficiently using a budget-aware framework.

A novel framework for concept unlearning in text-to-image diffusion models that enhances robustness and precision.

This research identifies implicit grading biases in LLMs based on writing style, highlighting the need for bias auditing in educational AI systems.

DA-Mamba enhances domain adaptive object detection by integrating CNNs with State Space Models for improved global-local feature alignment.

A simplified training method for LLMs that enhances reasoning without complex loss functions.

A comparative framework for detecting AI-generated texts using advanced neural models.

A mixed-precision quantization framework that optimizes video diffusion models for efficient inference.

A study revealing how confirmation bias in LLMs affects security code reviews and proposing debiasing methods.

Ontology-Guided Diffusion (OGD) enables interpretable and efficient zero-shot visual sim2real transfer using structured knowledge.

Li-Net is a novel architecture for efficient multi-channel time series forecasting using sparse attention mechanisms.

OCP optimizes embedding representation for industrial commodity recommendation systems, enhancing scalability and performance.

SCALe enhances vision-language model training by optimizing reasoning segment supervision for improved accuracy and efficiency.

SVOO is a training-free sparse attention framework that enhances video generation speed without compromising quality.

D-Mem is a dual-process memory system designed to enhance long-horizon reasoning in autonomous agents by combining lightweight vector retrieval with high-fidelity deliberation.

GEAR is a framework for efficiently retrieving geological analogs in extreme environments using advanced topographic analysis.

REST is a training-free framework for efficient zero-shot object-goal navigation using a tree of paths for decision-making.

OpenT2M provides a large-scale, high-quality motion dataset and a novel motion model for realistic human movement generation from text.

i-SDT is an intelligent self-defending digital twin that enhances cyber-physical system resilience against attacks while maintaining operational efficiency.

DriveTok is a 3D driving scene tokenizer that enhances multi-view reconstruction and understanding for autonomous driving systems.

OS-Themis is a scalable multi-agent critic framework designed to enhance GUI agent performance through improved reward functions.

GSMem enhances embodied exploration by using 3D Gaussian Splatting for persistent spatial memory and high-fidelity reasoning.

A customizable, low-cost tendon-actuated continuum robot design for versatile inspection and manipulation tasks.

HEALIX provides a novel dataset for identifying patient health literacy from clinical notes, enhancing automated detection methods.

Developing AI models for improving human-robot interaction through advanced vision-language reasoning and grounding capabilities.

VGGT-360 is a training-free framework for zero-shot panoramic depth estimation that leverages 3D consistency for improved accuracy.

Develop a tool to translate MRI scans to PET-like images using conditional diffusion models for improved pathology detection.

A framework for enhancing mathematical reasoning in LLMs through improved training data and methodologies.

ProRL Agent offers a scalable API service for efficient RL training of multi-turn LLM agents.

An agent that generates unique and valuable mathematical research problems using LLMs.

VesselTok offers a novel framework for encoding and generating complex 3D biomedical graph representations of anatomical structures.

CSSDF-Net provides a differentiable distance query mechanism for safe motion planning in robotics.

A framework for enhancing speech spoofing detection through invariant domain feature extraction.

GenVideoLens is a fine-grained benchmark for evaluating LVLMs in detecting AI-generated videos, revealing critical performance gaps.

A multimodal classification framework that enhances explainability in humanitarian crisis classification using cross-modal rationale transfer.

FinTradeBench is a benchmark designed to evaluate financial reasoning in LLMs by integrating company fundamentals and trading signals.

Evaluating state space models as competitive vision encoders for vision-language models.

A cost-aware framework for enhancing the robustness of phishing detection systems against feature manipulation attacks.

OmniVTA aids robots in efficiently manipulating objects using integrated visual and tactile feedback.

Box Maze is a novel architecture designed to enhance the reliability of LLM reasoning through explicit cognitive control layers.

A novel method for optimizing language model training by effectively allocating compute resources between pretraining and specialization.

UGID is a framework for debiasing large language models by modeling their internal representations as structured computational graphs.

SHAPCA provides interpretable explanations for machine learning models applied to spectroscopy data, enhancing trust in clinical settings.

SAVeS introduces a framework to steer safety judgments in vision-language models using semantic cues.

A dynamic ensemble framework that enhances loan default prediction accuracy through optimized model weighting.

A safety-guaranteed imitation learning framework for spacecraft close proximity control using Control Barrier Functions.

A privacy-preserving framework for cross-model alignment of language models enabling secure inference without direct data sharing.

Enable advanced video reasoning through trajectory-grounded analysis for educational and security applications.

ClawTrap is a MITM-based framework designed for evaluating the security of OpenClaw agents against real-world network threats.

WeNLEX generates weakly supervised natural language explanations for multilabel chest X-ray classification, enhancing interpretability and classification performance.

A study exploring bias in LLM outputs based on dialect and proposing mitigation strategies through multi-agent architectures.

A novel off-policy learning method for optimizing item allocation in constrained environments like e-commerce.

HISR enhances multi-turn decision-making in reinforcement learning by improving reward assignment through hindsight information.

APUN-Bench is a benchmark for evaluating audio pun understanding in large audio language models.

A novel approach to policy optimization in blackjack using a dynamic programming oracle for improved sample efficiency.

A study on mitigating catastrophic forgetting in continual intent classification for NLP systems.

Build a tool to extract implicit 3D priors from 2D scenes for enhanced scene understanding in AR applications.

MultiGP is a generative inverse rendering method that disentangles reflectance, texture, and illumination from a single image.

Developing algorithms for online learning using ranking feedback to improve decision-making in adversarial environments.

A novel approach to generative segmentation using vector field learning to enhance performance over traditional methods.

LVOmniBench is a benchmark for evaluating long-form audio-video comprehension in omnimodal large language models.

A novel framework for enhancing multidimensional path representation with exponentially weighted signatures for improved memory dynamics.

SOL-ExecBench provides a novel benchmarking framework for optimizing GPU kernels against hardware limits.

VEPO optimizes low-resource language models through a novel reinforcement learning framework that enhances tokenization and translation quality.

LuMamba enhances EEG modeling by providing electrode topology-invariant and efficient data processing.

CoMFed is a framework for efficient multi-modal federated learning that enhances model training while preserving privacy.

Introducing a novel metric for evaluating 3D spatial geometric consistency in dynamically generated videos.

A neural network approach for fast and interpretable autoregressive parameter estimation in time series analysis.

A minimalist approach to fingerprint enhancement that outperforms complex methods for clearer and more accurate images.

A novel intrinsic reward mechanism for reinforcement learning that enhances exploration efficiency.

BVSIMC enhances drug discovery by improving predictive accuracy and interpretability through Bayesian variable selection.

A verification tool for ensuring the security of masked hardware generated through High Level Synthesis.

PASTE enhances LLM agent performance by reducing latency through speculative tool execution.

A framework for evaluating human-AI decision-making readiness to enhance collaboration safety and effectiveness.

MultihopSpatial is a benchmark for enhancing multi-hop spatial reasoning in Vision-Language Models.

A passive mechanism for stackable airdrop sensors that enhances environmental monitoring efficiency.

Dual-IFM is an interpretable foundation model for retinal fundus images, enhancing decision-making in medical imaging.

A post-processing framework that enhances fairness in predictions across various machine learning tasks.

A system that predicts user performance in interactive tasks using physiological signals.

A comprehensive study on uncertainty quantification in medical image segmentation to improve interpretability and performance.

A visual-tactile imitation learning method for tracing deformable objects to enhance manipulation tasks.

A comprehensive framework for releasing differentially private graphs to enhance privacy without sacrificing utility.

A deep learning model combining VAE and GAN for improved parameterization in petroleum reservoir simulation.

A new polynomial kernel for Gaussian Splatting enhances performance while maintaining dataset compatibility.

STEP is a framework for unified representation learning of scientific time series through cross-domain distillation.

A novel defense mechanism against label inference attacks in vertical federated learning using task reassignment and layer adjustment.

MANAR is a memory-augmented attention mechanism that enhances contextualization by implementing principles of Global Workspace Theory.

A benchmarking study comparing CNN and transformer models for abdominal multi-organ segmentation in CT scans.

ZebraArena is a diagnostic simulation environment designed to study the coupling of reasoning and action in tool-augmented LLMs.

DiscoPhon is a multilingual benchmark for unsupervised phoneme discovery from discrete speech units.

AutORAN is an LLM-driven framework that automates xApp development for Open Radio Access Networks, enabling rapid deployment from user intents.

DyMoE optimizes MoE inference for edge devices through dynamic mixed-precision quantization.

cuGenOpt provides a general-purpose GPU-accelerated framework for combinatorial optimization.

This research explores how LLMs generate analogies better than humans, challenging traditional models of analogy.

A novel adaptive algorithm for nonlinear data assimilation using P-spline triangular measure transport.

MoRI aims to integrate motivation-grounded reasoning in scientific ideation within large language models.

A lightweight cryptographic proof framework for verifying AI inference correctness in cloud-based services.

RADIUS is an alignment suite for improving the evaluation of survey simulations using LLMs.

A comparative study of OmniAnomaly and PCA for multivariate time series anomaly detection.

A novel platform for conducting a distributed Turing Test with LLMs and humans.

A framework for structured intent representation to enhance human-AI interaction.

A framework for analyzing complex ideologies beyond traditional left/right paradigms using NLP.

A novel training pipeline for bilingual dialogue systems that aims to eliminate hallucination through citation grounding.

BeamAgent optimizes MIMO beamforming by decoupling intent parsing from numerical optimization using LLMs.

A framework for generating instruction-aware gestures in humanoid educational robots using a reasoning-guided approach.

A speaker-centric Emotion AI approach predicting audience engagement and vocal attractiveness from speaker expressiveness in video learning.

A deep learning method for robust visual feature tracking in monocular camera images for SLAM and Odometry.

A novel approach for analyzing vulnerabilities in Python applications through cross-ecosystem dependency analysis.

A new framework for improved side-channel leakage detection in neural networks using the Anderson-Darling test.

A philosophical framework for governing the development and integration of synthetic minds in society.

A reinforcement learning framework that enables large language models to iteratively refine their contexts for improved performance.

This paper proposes a new method for designing noise schedules in diffusion models to enhance image and video generation quality.

CALM improves treatment effect estimation by aligning covariate representations from RCTs and observational studies.

This paper evaluates LLMs' strategic reasoning in game-theoretic contexts, revealing limitations in their performance.

This paper addresses the generalization error in physics-informed neural networks for solving partial differential equations.

PPI integrates machine learning predictions with statistical inference, drawing from survey sampling methods.

This paper evaluates the performance of ChaCha20-Poly1305 encryption for IoT and ICS devices, highlighting its low impact on latency.

HOLMES is a computational framework for hierarchical latent structure learning through online inference.

This paper presents a study on implicit patterns in LLM-based binary vulnerability analysis.

This paper critiques existing efficiency metrics for Vision-Language-Action models and proposes a focus on embodied efficiency.

This paper explores uncertainty estimation in reasoning models but lacks a clear product application.

This paper critiques the theoretical foundations of Spectral Graph Neural Networks and their effectiveness in node classification.

This research explores the effects of cross-domain mappings on creativity in humans and LLMs.

This paper analyzes the effectiveness of the UK NIS Regulations in reporting cybersecurity incidents affecting critical infrastructure.

This paper presents theoretical lower bounds for linear classification problems.

Exploring the integration of AI in judicial decision-making to enhance transparency and accountability.

This paper analyzes privacy and convergence in differential privacy for wireless federated learning.

This paper explores the limitations of LLM agents in recognizing secure environments for privacy-preserving negotiations.

This paper presents a theoretical analysis of online resource allocation with endogenous costs and interaction effects.

A study on evaluating game difficulty in Tetris using a novel planning agent.

This paper develops the mathematical foundations of Schrödinger bridges for generative modeling.

A novel algorithm for multi-dueling bandits that optimally adapts to both stochastic and adversarial environments.

This paper explores the use of structural causal models to understand the intentions of goal-directed agents through intentional interventions.

A comprehensive taxonomy for multivariate time series anomaly detection using deep learning.

This study explores how the shape of uncertainty dynamics in LLM reasoning can predict accuracy.

A theoretical exploration of kernelized algorithms for contextual bandits with single-index models.

A manifesto proposing a new paradigm for Business Process Management focused on autonomous agents.

This paper reviews the evolving regulatory landscape for AI, focusing on security and privacy in the context of agentic AI.

A framework for approximating transition probability density functions using Neural Galerkin Normalizing Flows.

This paper presents a theoretical error analysis of the Adam optimizer for stochastic optimization problems.

This research evaluates the integrity of AI agents in governance roles, highlighting the need for pre-deployment safeguards.

This research explores using numeric self-reports in LLMs to track internal emotive states during conversations.

This paper proposes Authority-Level Priors as constraints in hierarchical predictive processing to explain stress reactivity and behavioral control.

Exploring how generative AI represents and reasons about geography through theoretical vignettes.

A framework integrating human feedback into LLM-based text simplification for cognitive accessibility.

This study identifies gaps in language learning apps like Duolingo and suggests personalized lesson generation for professional fluency.

This research explores how AI-mediated video communication affects trust and judgment in interpersonal interactions.

A new method for optimal multi-agent path finding that handles asynchronous actions.

This paper analyzes the failure of cross-lingual alignment to improve transfer performance in NLP tasks.

This paper explores student perspectives on the ethical and societal impacts of AI, highlighting gender differences in awareness and concerns.

This paper presents a theoretical framework for improving the generalization of generative models using discriminator guidance.

Maple is a socially assistive robot designed to enhance language and cultural learning for young newcomers through interactive storytelling.

SRRM enhances the efficiency of surrogates for the Wasserstein distance but lacks practical implementation details.

This paper analyzes the security vulnerabilities and deployment challenges of the new IEEE 802.11az and 802.11bk standards for secure Wi-Fi ranging.

This paper presents a framework for evaluating the balance between cognitive amplification and delegation in human-AI systems.

A comprehensive review of challenges and advancements in multimodal computational pathology.

This paper theoretically analyzes the convergence properties of No-U-Turn Sampler variants without providing a clear path to practical application.

SwiftGS offers a novel approach to rapid 3D reconstruction from satellite imagery using meta-learning techniques.

This paper classifies the complexity of shuffles in card-based cryptography protocols.

An anthology of research papers on Theory of Mind in AI.