This paper shows that in the large depth and width asymptotics, Dropout and Random Gradient Masking (RaM) converge to the same limiting dynamics for ResNets.

The paper introduces MDM-VGB, a reward-guided sampler for Masked Diffusion Models, which extends the Jerrum-Sinclair backtracking Markov chain to a masked-state graph for effective high-reward generat…

This paper proposes TileLens, a system to mitigate read amplification in Large-Granularity Memory Systems (LGMS) for Large Language Model (LLM) inference by adopting a tile-major layout.

The paper introduces DSL-LLaDA, a method that lightly adapts a pre-trained masked diffusion language model to perform continuous denoising in embedding space, significantly improving text generation q…

The paper shows that using random cropping, a standard data augmentation technique, can naturally amplify differential privacy guarantees for machine learning models without requiring any changes to t…

This paper introduces Curvature-Weighted Gradient Diversity (CWGD), a geometry-aware measure for optimization noise that reduces the asymptotic optimization error floor by up to a factor of two compar…

The paper systematically characterizes column-level activation sparsity across various diffusion model architectures, demonstrating that element-level sparsity metrics significantly overestimate the a…

MaskForge is a novel, adaptive, black-box attack framework that significantly improves jailbreaking diffusion large language models (dLLMs) by treating red-teaming as an optimized search over reusable…

The paper introduces Token-Aware Gradient Optimization (TAGO), demonstrating that sparse optimization focusing only on high-gradient audio tokens is sufficient for effective jailbreaking of audio lang…

The paper presents a highly optimized, low-stack implementation of the HAETAE signature scheme, reducing peak stack usage significantly to enable its use on severely memory-constrained microcontroller…

This paper develops provably undetectable and robust watermarking schemes for LLM outputs even when the per-token entropy is only constant, removing previous dependencies on high entropy rates or larg…

This paper characterizes per-bit-position fault sensitivity in machine learning inference across various workloads and floating-point formats, identifying a sharp bit-sensitivity transition and derivi…

VideoMLA introduces a novel Multi-Head Latent Attention (MLA) mechanism that replaces per-head KV caches with a shared low-rank content latent, significantly reducing memory and improving throughput f…

This paper identifies prediction bias, a failure mode of entropy minimization in test-time adaptation, and proposes Distribution Shift Bias Reduction (DSBR) to stabilize adaptation and prevent model c…

The paper introduces 'infilling extraction' to accurately model training data memorization in Diffusion Language Models (DLMs), finding that bidirectional masking significantly increases the extractab…

The paper introduces ImageProtector, a user-side method that embeds an imperceptible perturbation into images to prevent Multi-modal Large Language Models (MLLMs) from analyzing and extracting sensiti…

This paper explores the convergence of AdamW optimizer under heavy-tailed assumptions in large language models and proposes a corridor lower-bound mechanism.

This paper proposes using color statistics, specifically through novel color transformations, to detect AI-generated synthetic images by exploiting the color-imitation weaknesses of current generative…