arXiv preprints from January 1, 2026 through September 23, 2026 — 09:43:11 EST
Large Language Models (LLMs) offer a promising solution to complement traditional teaching and address global teacher shortages that affect hundreds of millions of children, but they fail to provide grade-appropriate responses for students at different educational levels. We introduce a framework for finetuning LLMs to generate...
Recent end-to-end autonomous driving approaches have leveraged Vision-Language Models (VLMs) to enhance planning capabilities in complex driving scenarios. However, VLMs are inherently trained as generalist models, lacking specialized understanding of driving-specific reasoning in 3D space and time. When applied to autonomous driving,...
Deep learning models for image compression often face practical limitations in hardware-constrained applications. Although these models achieve high-quality reconstructions, they are typically complex, heavyweight, and require substantial training data and computational resources. We propose a methodology to significantly reduce...
A new approach for analyzing waveguide junctions containing conductive cylindrical objects is proposed. The algorithm is based on mode matching technique using local projection functions, which improves the numerical conditioning of the problem. Moreover, the approach enhances computational efficiency by reducing the boundary where...
As generative models become ubiquitous, there is a critical need for fine-grained control over the generation process. Yet, while controlled generation methods from prompting to fine-tuning proliferate, a fundamental question remains unanswered: are these models truly controllable in the first place? In this work, we provide a...
We study deletion-correcting codes in the space of length-$n$ multisets over a $q$-ary alphabet. We present an explicit cyclic Sidon-type construction for arbitrary alphabet size $q$ and deletion radius $t$, defined by a single congruence modulo $t(t+1)^{q-2}+1$. The construction has redundancy at most $\log_q(t(t+1)^{q-2}+1)$ and...
Preserving privacy in sensitive data while pretraining large language models on small, domain-specific corpora presents a significant challenge. In this work, we take an exploratory step toward privacy-preserving continual pretraining by proposing an entity-based framework that synthesizes encrypted training data to protect personally...
We show a correspondence between the compact exceptional curves and divisors on $G-{\rm Hilb}(\mathbf{C}^3)$ and some non-trivial irreducible representations of $G \subset GL(n,C)$ which are special (or essential). Moreover, we provide an explicit construction of the small resolution of $G-{\rm Hilb}(\mathbf{C}^3)$ and, using this...
Recent LLMs excel at formal tasks such as mathematical reasoning and code generation, but still struggle with broader abilities such as planning, creativity, and social intelligence. Inspired by human learning, where formal instruction and informal experience jointly shape intelligence, we introduce informal learning into LLM training...
Current model structural discovery methods for power system dynamics impose rigid priors on the basis functions and variable sets of dynamic models while often neglecting algebraic constraints, thereby limiting the formulation of high-fidelity models required for precise simulation and analysis. This letter presents a novel large...
The concept of keplerian shear was introduced by Damien Thomine recently. It is useful for non ergodic systems, and can be seen as strong mixing conditionally on invariant fibers. The notion is particularly interesting when a.e. fiber is not strongly mixing. We develop here an approach appropriate for systems such that a.e. fiber is...
Since decades, the CEA has been involved in the development of 4th generation reactors cooled by sodium. The dedicated experimental platform PLATEAU was erected to study the different issues and built a results database for the code validation. As experiments with sodium are complex, the tests are led in mock-ups using water as a...
Interaction between energy-abundant (sub-)mesoscale eddies and internal waves can lead to turbulence generation and may prove important for replenishment of nutrients for deep-sea life and circulation. However, observational evidence of such interaction is scarce and precise energy transfer is unknown. In this paper, an extensive...
Conventional Knowledge Graph Completion (KGC) methods aim to infer missing information in incomplete Knowledge Graphs (KGs) by leveraging existing information, which struggle to perform effectively in scenarios involving emerging entities. Inductive KGC methods can handle the emerging entities and relations in KGs, offering greater...
Recent studies in \cite{Park:2020jio,Paul:2025gpk} have calculated various holographic information theoretic quantities of the four-dimensional FLRW universe for different matter-dominated eras using the braneworld model of cosmology. These studies are done for a single matter component, which is a good toy model for understanding the...
We investigate thermalization and the quantum-classical correspondence in the collective Bose-Hubbard model, focusing on the four-site case. Our analysis of the classical phase-space structure and its excited-state quantum phase transitions leads us to three dynamical regimes: symmetry-breaking low-energy states, an intermediate...
Understanding how lipid bilayers respond to pressure is essential for interpreting the coupling between membrane proteins and their native environments. Here, we use all-atom molecular dynamics to examine the pressure-temperature behavior of model membranes composed of DMPC or $Δ$9-cis-PC. Within the studied range (288-308 K, 1-2000...
Geometric function theory increasingly draws on $q$-calculus to model discrete and quantum-inspired phenomena. Motivated by this, the present paper introduces new subclasses of analytic functions: the class $\mathcal{S}^{*}_{ξ_q}$ of $q$-starlike functions associated with the Ma-Minda function $ξ_q(z)$, and its limiting classical...
Toxic language is one of the major barrier to safe online participation, yet robust mitigation tools are scarce for African languages. This study addresses this critical gap by investigating automatic text detoxification (toxic to neutral rewriting) for two low-resource African languages, isiXhosa and Yorùbá. The work contributes a...
Existing research on continual learning (CL) of a sequence of tasks focuses mainly on dealing with catastrophic forgetting (CF) to balance the learning plasticity of new tasks and the memory stability of old tasks. However, an ideal CL agent should not only be able to overcome CF, but also encourage positive forward and backward...
A Software Bill of Materials (SBOM) is a machine-readable artifact that systematically organizes software information, enhancing supply chain transparency and security. To facilitate the exchange and utilization of SBOMs, organizations such as the Linux Foundation and OWASP have proposed SBOM standards. Following standards,...
Interfacial interactions in two parallel-stacked hexagonal boron-nitride (hBN) layers facilitate sliding ferroelectricity, enabling novel device functionalities. Additionally, when Bernal or twisted bilayer graphene is aligned with an hBN layer, unconventional ferroelectric behavior was observed, though its precise origin remains...
Island models offer a compelling resolution of the black hole information paradox, but they also raise persistent questions about causal consistency in effective descriptions. In particular, effective theories arising in double holography can exhibit apparent violations of micro-causality, despite the underlying bulk dynamics being...
Precipitate size evolution during room temperature high-pressure torsion (HPT) of a Mg-1.35wt.%Mn alloy was studied using scanning transmission electron microscopy (STEM) and Small-/Wide-angle X-ray scattering (SAXS/WAXS). The volume fraction of the nm-scale $α$-Mn particles increased with applied strain, however small angle X-ray...
The Hilbert transform is a multiplier operator and is widely used in the theory of Fourier transforms. The Hilbert transform was the motivation for the development of modern harmonic analysis. Its discrete version is also widely used in many areas of science and technology and plays an important role in digital signal processing. The...