10075 papers
This paper introduces FlashCache, a frequency-domain-guided KV cache compression framework that identifies and preserves critical "Outlier KVs" while leveraging low-pass filtering and dynamic budget allocation to achieve significant inference speedups and memory reduction in multimodal large language models without compromising performance.
The paper introduces Observer-Actor (ObAct), a novel active vision imitation learning framework for dual-arm robots that dynamically assigns one arm to construct a 3D Gaussian Splatting representation and identify optimal viewing angles for the other arm, thereby significantly enhancing policy robustness and performance by reducing occlusions compared to static-camera setups.
This paper introduces CycleChemist, a dual-pronged machine learning framework that leverages a new large-scale dataset (OPV2D) to simultaneously predict organic photovoltaic performance and generate synthetically accessible high-efficiency donor-acceptor materials through an integrated predictive and generative approach.
This paper proposes a novel Physics-Informed U-Net-LSTM framework that integrates physical laws with deep learning to overcome the computational limitations of traditional Finite Element Methods and the generalization issues of purely data-driven models, thereby enabling accurate and efficient prediction of nonlinear structural seismic responses.
This paper introduces DPAC, a diffusion guidance method that projects adversarial gradients onto the tangent space of iso-density surfaces to minimize path-space KL divergence and control energy, thereby theoretically and empirically achieving higher sample quality (lower FID) while maintaining target classification success.
This study presents an AI implementation science case study at Shriners Childrens that modernizes its research data infrastructure to the OMOP CDM standard, introduces a Python-based tool extending data quality assessment with Trustworthy AI principles, and evaluates hybrid implementation strategies for clinical applications like Craniofacial Microsomia to accelerate trustworthy AI adoption in healthcare.
The paper proposes GRAND, a hybrid hierarchical algorithm that combines reinforcement learning-based graph guidance with minimum-cost flow rebalancing and local assignment to achieve up to 10% higher throughput than state-of-the-art schedulers for large-scale, lifelong multi-agent pickup-and-delivery tasks while maintaining real-time execution.
This paper introduces fairness-aware Low-Rank Adaptation methods, specifically FR-LoRA, GR-LoRA, and Hybrid-LoRA, which utilize a differentiable MaxAccGap loss and inverse frequency weighting to significantly reduce diagnostic accuracy disparities in glaucoma detection across demographic groups while maintaining high overall accuracy with minimal trainable parameters.
The paper introduces Guided Flow Policy (GFP), a novel offline reinforcement learning method that couples a multi-step flow-matching policy with a distilled one-step actor to selectively focus on high-value actions, achieving state-of-the-art performance across diverse benchmarks by overcoming the limitations of indiscriminate behavior regularization.
This paper demonstrates that injecting a canonical action ordering signal into the reward function during RL post-training significantly improves Transformer performance on Zebra puzzles compared to optimizing for task success alone, even when the model is fine-tuned on randomized solution sequences.
NeuralRemaster introduces Phase-Preserving Diffusion (Ï-PD), a model-agnostic method that replaces standard Gaussian noise with phase-preserving, magnitude-randomized noise to enable structure-aligned image and video generation without architectural changes or inference-time costs.
This paper introduces a post-training method that induces extreme sparsity in transformer attention (reducing connectivity to ~0.4%) without sacrificing performance, thereby revealing simplified, interpretable task-specific circuits and unifying feature-based and circuit-based perspectives on model behavior.
The paper introduces SACP, a novel method that aggregates nonconformity scores from multiple predictive models into e-values using symmetric functions to generate valid and more efficient uncertainty sets within the conformal prediction framework.
This paper proposes TAP, a novel training-free framework that leverages Large Language Models and evolutionary search to automatically discover superior mixed-precision quantization proxies, eliminating the need for human expertise or costly optimization while achieving state-of-the-art performance.
ReFusion introduces a novel masked diffusion model that integrates sequence reorganization with a hybrid parallel-autoregressive decoding strategy to simultaneously achieve full KV cache efficiency, reduce learning complexity, and significantly outperform existing diffusion models while narrowing the performance gap with autoregressive models.
This paper presents non-asymptotic, distribution-free methods for estimating the calibration error of binary classifiers, including a theoretical upper bound for bounded-variation functions and a practical modification technique that ensures efficient error bounding without significantly compromising performance.
This paper introduces RePo, a novel mechanism that leverages a differentiable module to dynamically re-position tokens based on contextual dependencies rather than fixed linear indices, thereby reducing extraneous cognitive load and enhancing LLM performance on tasks involving noisy contexts, structured data, and long-range dependencies.
The paper introduces AMPEND-LS, an agentic multi-persona framework that synergizes Large Language Models and Small Language Models with evidence-grounded reasoning to achieve state-of-the-art, explainable, and robust multimodal fake news detection.
This paper introduces Parallel Token Prediction (PTP), a framework that accelerates language model inference by predicting multiple tokens in a single forward pass through deterministic functions of random input variables, achieving a 2.4x speedup over autoregressive decoding.
This paper introduces an SVGP-KAN framework that reconstructs time-resolved flow fields from sparse velocimetry data with accuracy comparable to classical methods while providing well-calibrated epistemic uncertainty estimates to guide experimental design.