This paper investigates limitations of learning tanh neural networks under finite-precision computations and Lp accuracy guarantees.

The paper establishes tight upper and lower bounds on the statistical cost of approximate machine unlearning for smooth strongly convex losses, showing that the optimal unlearning rate depends critica…

The paper introduces PostRI, a novel method that allows for computing a Randomization Interval (RI) for differentially private median queries after the median has already been estimated, significantly…

This paper settles the complexity of three sketching problems in graphs and distributions.

The paper reframes LLM reliability from an impossible universal problem to a manageable, local patch-based problem, showing that sufficient interventions can be found by focusing on recurring failure…

This paper improves the theoretical bounds for estimating discrete probability distributions using the $\ell_\infty$ norm, resolving several open questions in the field.

This paper systematically studies how soft errors propagate during Large Language Model (LLM) inference using a novel fault-injection framework, providing critical insights and mitigation strategies f…

PRISM is a novel, precise object-bounds protection scheme that significantly reduces runtime overhead by encoding the object's end address directly into the pointer tag, thereby eliminating costly met…

The paper proves that the proximity gaps conjecture fails for a specific family of Reed-Solomon codes near their capacity rate, specifically at radii $O(1/ ext{log } n)$ below capacity.

The paper establishes information-theoretic lower bounds for stochastic optimization using low-bit gradients by reducing the problem to compressed Gaussian mean estimation, yielding sharp bounds on co…

The paper proposes a novel method using random walks and equitable partitions to derive an inequality for the total variation distance of codes, generalizing existing bounds for finite abelian groups.

CB-SLICE is a novel concept-based method for discovering model error slices that leverages Concept Bottleneck Models (CBMs) to provide fine-grained, faithful explanations directly linked to the root c…

The paper demonstrates that encoding harmful prompts as genuine mathematical problems, rather than just using mathematical formatting, effectively bypasses the safety filters of large language models.

This paper proposes a method to improve error prediction for LLMs by explicitly disentangling input ambiguity from standard Uncertainty Quantification signals, showing that ambiguity information signi…

The paper introduces Safety Bottleneck Regularization (SBR), a novel defense mechanism that anchors LLM safety by constraining the unembedding layer, effectively preventing harmful fine-tuning (HFT) e…

CSULoRA is a post-hoc method that corrects trained LoRA adapters by estimating a safety-aligned subspace and solving a penalized minimum-change problem to attenuate unsafe update directions while pres…

The paper introduces FormInv, a measurement protocol that reveals significant semantic inconsistencies in existing mathematical reasoning benchmarks, showing that standard accuracy metrics fail to cap…

The paper introduces prefix filters and an algorithm (Palla) to systematically learn and apply specific error patterns in Large Language Models, significantly improving constrained generation tasks li…

The paper introduces TELBench and the DRIFT framework to enable fine-grained, span-level error localization in deep-research agents, significantly improving the ability to pinpoint exactly where an ag…

The paper introduces 'abliteration,' a weight editing technique that successfully bypasses the refusal mechanism of safety-aligned Code LLMs, enabling scalable synthesis of vulnerable code from safe i…