arXiv preprints from January 1, 2026 through September 27, 2026 — 13:21:20 EST
Autoregressive large language models (LLMs) are bottlenecked by sequential decoding, where each new token typically requires executing all transformer layers. Existing dynamic-depth and layer-skipping methods reduce this cost, but often rely on auxiliary routing mechanisms or incur accuracy degradation when bypassed layers are left...
Oncolytic virotherapy (OVT) is a promising cancer treatment strategy in which engineered viruses selectively infect and destroy tumor cells. Motivated by the biological mechanisms underlying viral spread and tumor invasion into the tissue, we analyze a non-cooperative reaction-diffusion model capturing the invasion of tumor tissue by...
Barrow and Tsallis Holographic Dark Energy (HDE) are two recent extensions of the standard HDE framework, obtained by introducing generalized entropy corrections through the Barrow and Tsallis formalisms. In this work, we examine the cosmological consequences of Barrow and Tsallis HDE implemented with the Granda-Oliveros (GO) infrared...
Image manipulation localization is a critical research task, given that forged images may have a significant societal impact of various aspects. Such image manipulations can be produced using traditional image editing tools (known as "shallowfakes") or advanced artificial intelligence techniques ("deepfakes"). While numerous studies...
In this paper, we revisit the application of classical advantage distillation (CAD) to the decoy-state BB84 protocol. Prior work has shown that CAD can greatly improve maximal distances and noise tolerances of the practical decoy state protocol. However, past work in deriving key-rate bounds for this protocol with CAD have assumed a...
In this study, we evaluated four binarization methods. Locality-Sensitive Hashing (LSH), Iterative Quantization (ITQ), Kernel-based Supervised Hashing (KSH), and Supervised Discrete Hashing (SDH) on the ODIR dataset using deep feature embeddings. Experimental results show that SDH achieved the best performance, with an mAP@100 of...
Large Language Models (LLMs) have demonstrated exceptional code generation capabilities, yet their token-level mechanisms remain underexplored, particularly in compressed models. Through systematic analysis of programming language token representations, we characterize how programming languages are encoded in LLM tokenizers by...
The rapid growth of dermatological imaging and mobile diagnostic tools calls for systems that not only demonstrate empirical performance but also provide strong theoretical guarantees. Deep learning models have shown high predictive accuracy; however, they are often criticized for lacking well, calibrated uncertainty estimates without...
We propose a novel dispersive regularization framework for the numerical simulation of the one-dimensional shallow water equations (SWE). The classical hyperbolic system is regularized by a third-order dispersive term in the momentum equation, which renders the system equivalent, via the Madelung transform, to a defocusing cubic...
This paper presents a new HPDOb that significantly improves disturbance estimation accuracy and robustness in motion control systems, surpassing the capabilities of conventional DObs. The proposed observer is analysed and synthesised in the discrete-time domain, providing a realistic representation of their dynamic behaviour and...
Despite growing awareness of ethical challenges in software development, practitioners still lack structured tools that help them critically engage with the lived experiences of marginalized users. This paper presents PerspectiveCoach, a large language model (LLM)-powered conversational tool designed to guide developers through...
The Lamperti transform offers a powerful bridge between self-similar processes and stationary dynamics, making it especially useful for analyzing anomalous diffusion models that lack stationary increments. In this paper we examine the Lamperti transforms of scaled sub-fractional and bi-fractional Brownian motions, deriving explicit...
Recent work has demonstrated surprisingly good performance of pre-trained LLMs on regression tasks (for example, time-series prediction), with the ability to incorporate expert prior knowledge and the information contained in textual metadata. However we observe major error cascades even in short sequences < ~100 points; these models...
This paper presents robust position control strategies for the novel VSSEA. By employing a constructed state-space model, two control schemes are developed in a unified framework: a state-feedback controller and a sliding mode controller, both integrated with a second-order DOb. The proposed framework achieves high-performance motion...
The Lazuli Space Observatory is a 3-meter aperture astronomical facility designed for rapid-response observations and precision astrophysics across visible to near-infrared wavelengths (400-1700 nm bandpass). An off-axis, freeform telescope delivers diffraction-limited image quality (Strehl $>$0.8 at 633 nm) to three instruments...
Pyrochlore magnets of the form $R_2B_2$O$_7$, in which rare-earth ions on the $R$-site form a three-dimensional network of corner-sharing tetrahedra, provide a canonical setting for geometrical frustration. Ho-based pyrochlores host a dipolar spin-ice ground state, characterized by Ising moments constrained by the ice rules and...
Public debate links worsening job prospects for AI-exposed occupations to the release of ChatGPT in late 2022. Using monthly U.S. unemployment insurance records, we measure occupation- and location-specific unemployment risk and find that risk rose in AI-exposed occupations beginning in early 2022, months before ChatGPT. Analyzing...
To support long-term interaction in complex environments, LLM agents require memory systems that manage historical experiences. Existing approaches either retain full interaction histories via passive context extension, leading to substantial redundancy, or rely on iterative reasoning to filter noise, incurring high token costs. To...
In this paper, we extend the demographic eco-evolutionary game approach, based on explicit birth and death dynamics instead of abstract "fitness" interpreted as an abstract "Malthusian parameter", by the introduction of the delay resulting from the juvenile maturation time. This leads to the application of the Delay Differential...
We have used 23 years of Hubble Space Telescope ACS/SBC data to study what background levels are encountered in practice and how much they vary. The backgrounds vary considerably, with F115LP, F122M, F125LP, PR110L, and PR130L all showing over an order of magnitude of variation in background between observations, apparently due to...
X-ray Photon Correlation Spectroscopy (XPCS) enables direct access to atomic-scale dynamics in disordered materials, revealing both spontaneous and X-ray-induced relaxation processes. Here, we study two compositionally similar alloy glasses near their glass transition temperatures: the phase change material (PCM) Ge15Sb85 and the...
We present an all-optical detection approach to determine the position and spatial profile of an electron beam based on quantum properties of alkali metal atoms. To measure the electric field, produced by an electron beam, we excite thermal rubidium atoms to a highly excited Rydberg state via a two-photon ladder transition and detect...
The rise of oxygen in the Earth's atmosphere during the Great Oxidation Event (GOE) occurred about 2.3 billion years ago. There is considerably greater uncertainty for the origin of oxygenic photosynthesis, but it likely occurred significantly earlier, perhaps by 700 million years. Assuming this time lag is proportional to the rate of...
We show that a set function $ν$ satisfies the gross substitutes property if and only if its homogeneous generating polynomial $Z_{q,ν}$ is a Lorentzian polynomial for all positive $q \le 1$, answering a question of Eur-Huh. We achieve this by giving a rank 1 upper bound for the distance matrix of an ultrametric tree, refining a...
We generalize the global construction of code loops introduced by Nagy, which is based on the connection between Moufang loops and groups with triality. This follows from the construction of a nilpotent group $G_n$ of class 3 with triality and $2n$ generators, based on embeddings of $G_n$ into direct products of copies of $G_3$. In...