4112 papers
This paper introduces a parameter-efficient Diffusion Transformer (DiT) with a 2D CNN encoder that generates high-quality, cell-type-specific synthetic regulatory DNA sequences with significantly faster convergence, reduced memorization, and enhanced regulatory activity compared to existing U-Net-based models.
This paper proposes novel kernel-based equivalence tests using kernel Stein discrepancy and Maximum Mean Discrepancy to rigorously assess the absence of statistically meaningful differences between distributions, addressing the limitations of traditional goodness-of-fit tests and existing parametric methods through asymptotic and bootstrap critical value approximations.
This paper introduces Dynamics-Predictive Sampling (DPS), a method that models prompt solving progress as a dynamical system to predict and select informative training samples via online Bayesian inference, thereby significantly reducing the computational overhead of extensive rollouts while accelerating and improving the reinforcement learning finetuning of large reasoning models.
This paper argues that the standard expected value objective in reinforcement learning is inadequate for non-ergodic environments where individual agent performance matters, and it explores the relationship between ergodic reward processes and Markov chains while presenting solutions to optimize long-term performance for single trajectories.
LookaheadKV is a lightweight KV cache eviction framework that achieves fast and accurate long-context inference by using parameter-efficient modules to predict future token importance without the computational overhead of explicit draft generation, thereby outperforming existing methods in both accuracy and speed.
This paper demonstrates that applying Combinatorial Fusion Analysis (CFA) to generate team rankings for the 2024 NCAA tournament yields a 74.60% prediction accuracy, outperforming the best of ten popular public ranking systems and validating CFA as an effective paradigm for enhancing sports prediction precision.
The paper introduces ECoLAD, a deployment-oriented evaluation protocol that reveals how throughput constraints on automotive hardware often render deep learning anomaly detectors infeasible before they lose accuracy, whereas lightweight classical methods maintain both coverage and detection performance.
This paper introduces Historical Consensus Training, an iterative method that eliminates posterior collapse in Variational Autoencoders by progressively refining Gaussian Mixture Model priors to create a stable parameter barrier that prevents the degeneration of latent variables, achieving robust representations without relying on specific architectural constraints or hyperparameter tuning.
This paper proposes a practical Kernel Density Estimator-based method to quantify membership disclosure risk in tabular synthetic data by modeling nearest-neighbor distances, demonstrating through empirical evaluation that it outperforms existing baselines in accuracy and efficiency without requiring computationally expensive shadow models.
This paper proposes Risk-sensitive Alignment via Dominance (RAD), a novel Safe RLHF framework that replaces traditional expected cost constraints with First-Order Stochastic Dominance constraints within an Optimal Transport framework to universally control spectral risk measures, thereby achieving superior robustness against tail risks and out-of-distribution failures while maintaining helpfulness.
This paper introduces a selective prediction framework for molecular structure retrieval from mass spectra that leverages retrieval-level uncertainty and distribution-free risk control to allow models to abstain from low-confidence predictions, thereby ensuring annotations meet specified error rate constraints in high-stakes applications.
This paper introduces Scorio, an open-source library that formalizes and implements statistical methods for reliably ranking reasoning LLMs under test-time scaling, demonstrating that most approaches achieve high agreement with a Bayesian gold standard across multiple Olympiad-style math benchmarks.
This paper introduces a bio-inspired self-supervised learning framework for wrist-worn IMU signals that leverages motor control submovement theory to tokenize motion into segments, enabling a Transformer encoder to outperform existing baselines in human activity recognition through masked segment reconstruction on a large-scale NHANES dataset.
The paper introduces Pointy, a lightweight transformer-based point cloud architecture that, despite being trained on a small dataset of only 39k samples, outperforms larger foundation models and approaches state-of-the-art results through a carefully curated training setup and rigorous, standardized evaluation framework.
This paper introduces TOSSS, a CVE-based benchmark designed to evaluate the ability of Large Language Models to distinguish between secure and vulnerable code snippets in C/C++ and Java, revealing that current models achieve security scores ranging from 0.48 to 0.89.
This paper proposes a privacy-preserving Federated Learning framework that jointly optimizes multi-RIS configuration and eavesdropper detection in B5G cell-free mmWave networks, achieving a 30% improvement in secrecy rate while maintaining high detection accuracy through collaborative DCNN training on local Channel State Information.
This paper proposes a Federated Learning framework for LEO 6G Non-Terrestrial Networks that leverages High-Altitude Platform Stations to distribute beam selection tasks, demonstrating that a Graph Neural Network model outperforms a Multi-Layer Perceptron in prediction accuracy and stability, especially at low elevation angles.
This paper demonstrates that MLP layers in transformer models function as binary routing switches that direct continuous signals through distinct computational paths based on consensus and exception-handling neuron architectures, a mechanism that explains the limitations of smooth polynomial approximations and is validated by significant causal performance differences.
This paper introduces an MCMC-informed neural emulator framework that decouples uncertainty quantification from network architecture by incorporating model-parameter distributions as training inputs, thereby enabling computationally efficient and accurate surrogate modeling for dynamical systems while avoiding exhaustive sampling and unphysical parameter evaluations.
This paper proposes "ForwardFlow," a frequentist deep learning framework that utilizes a branched neural network trained on simulated data to directly estimate statistical parameters, demonstrating advantages such as finite sample exactness, robustness to contamination, and the ability to automatically approximate complex algorithms like the EM-algorithm.