7458 papers
The paper proposes ST-Lite, a training-free KV cache compression framework that leverages the uniform high-sparsity of GUI attention patterns through a dual-branch scoring policy of spatial saliency and trajectory-aware semantic gating, achieving significant decoding acceleration with minimal performance loss in long-horizon GUI agents.
This paper proposes LoDA, a task-driven subspace decomposition method for LoRA-based continual learning that enhances knowledge sharing and isolation by decoupling general and task-specific directions through energy-based objectives and gradient-aligned optimization, thereby outperforming existing approaches.
SKeDA is a generative watermarking framework for text-to-video diffusion models that enhances robustness against frame reordering and temporal distortions through Shuffle-Key-based Distribution-preserving Sampling and Differential Attention, while maintaining high video generation quality.
This case study demonstrates that a concept induction-based framework for interpreting hidden neurons in CNNs, previously validated on ADE20K, successfully generalizes to the SUN2012 scene recognition dataset, confirming its broader applicability.
This paper proposes a stateful, progressive token reduction framework with a unified language-aware scoring mechanism for hybrid video VLMs, achieving significant prefilling speedups while maintaining near-baseline accuracy by addressing the layerwise instability of token importance in architectures combining attention and state-space blocks.
This paper proposes AdURA-Net, a geometry-driven deep learning framework that combines adaptive dilated convolutions with a dual-head loss function to effectively handle diagnostic uncertainty and improve reliability in multilabel thoracic disease classification.
This research enhances MR image reconstruction by optimizing SOUP-GAN and CSR-GAN models through architectural modifications and hyperparameter tuning, demonstrating that CSR-GAN excels in preserving high-frequency details while SOUP-GAN produces superior structural clarity with reduced noise.
This paper proposes FMCT and its efficient variant EFMCT, which leverage the deterministic nature of Flow Matching and a velocity field reuse strategy to achieve high-quality, computationally efficient sparse-view CT reconstruction with bounded error and significantly fewer neural network function evaluations compared to diffusion-based methods.
The TACIT Benchmark introduces a programmatic visual reasoning evaluation framework featuring 10 tasks across 6 domains with dual-track generative and discriminative assessments, utilizing deterministic computer-vision verification and structurally rigorous distractors to overcome the limitations of existing language-dependent and subjective benchmarks.
The paper proposes VisRef, a computationally efficient test-time scaling framework that improves multi-modal reasoning performance by dynamically re-injecting a diverse, semantically relevant coreset of visual tokens to prevent models from losing focus on image content during extended textual reasoning.
This paper evaluates the effectiveness of Naturalistic Adversarial Patches in physically disrupting traffic-sign detection for autonomous vehicles by introducing the customized CompGTSRB dataset and validating the attack's impact on a YOLOv5 model through systematic experiments on a Quanser QCar testbed.
This paper introduces a quantum-inspired multi-class classifier based on the Pretty Good Measurement (PGM) that reformulates classification as quantum state discrimination, demonstrating competitive and often superior performance in radiomics tasks for lung cancer subtyping and prostate cancer risk stratification compared to established classical baselines.
This paper presents a novel, end-to-end quantum Wasserstein GAN framework that overcomes previous scaling limitations by utilizing advanced image loading techniques and tailored variational circuit architectures to generate high-resolution, diverse images from full MNIST, Fashion-MNIST, and Street View House Numbers datasets without relying on dimensionality reduction or patch-based tricks.
This paper proposes AP-PCO, a joint position-color optimization framework that generates cross-spectral adversarial patches to effectively attack visual-infrared dense prediction systems by simultaneously perturbing both modalities while maintaining stealth through color adaptation.
This paper introduces quadspectral and hyperspectral passive LWIR ranging methods that utilize ozone absorption features to estimate and mitigate reflected downwelling radiance, significantly improving distance measurement accuracy from over 100 meters to as low as 1.2 meters.
This paper proposes a novel multimodal learning framework that decomposes representations into invariant and specific components to derive tractable Probability of Necessity and Sufficiency (PNS) objectives, thereby enhancing predictive performance and robustness to missing modalities in medical data analysis.
Proof-of-Perception (PoP) is a tool-using multimodal reasoning framework that generates calibrated, stepwise uncertainty via conformal sets to dynamically allocate computational resources, thereby reducing hallucinations and improving accuracy-efficiency trade-offs compared to existing baselines.
This paper proposes a novel conditional diffusion framework for low-light image enhancement that utilizes a Structured Control Embedding Module (SCEM) to decompose input images into physical priors, achieving state-of-the-art performance and strong generalization across multiple benchmarks without fine-tuning.
This paper presents a percept-aware surgical planning framework that optimizes electrode placement for cortical visual prostheses by formulating it as a differentiable constrained optimization problem, which simultaneously maximizes perceptual reconstruction fidelity and adheres to critical vascular safety and anatomical feasibility constraints.
This study presents a deep learning framework for meat freshness detection that combines U-Net-based segmentation with OOD-aware classification, demonstrating that EfficientNet-B0 achieves the highest accuracy (98.10%) on RGB images while supporting practical on-device deployment via TensorFlow Lite.