Tootfinder

Scaling Laws and Spectra of Shallow Neural Networks in the Feature Learning Regime
Leonardo Defilippis, Yizhou Xu, Julius Girardin, Emanuele Troiani, Vittorio Erba, Lenka Zdeborov\'a, Bruno Loureiro, Florent Krzakala
https://arxiv.org/abs/2509.24882

Scaling Laws and Spectra of Shallow Neural Networks in the Feature Learning Regime
Neural scaling laws underlie many of the recent advances in deep learning, yet their theoretical understanding remains largely confined to linear models. In this work, we present a systematic analysis of scaling laws for quadratic and diagonal neural networks in the feature learning regime. Leveraging connections with matrix compressed sensing and LASSO, we derive a detailed phase diagram for the scaling exponents of the excess risk as a function of sample complexity and weight decay. This anal…

The evolution of influence operations
from crude Russian troll farms to sophisticated AI systems using large language models;
the discovery of GoLaxy documents revealing a "Smart Propaganda System" that collects millions of data points daily, builds psychological profiles, and generates resilient personas;
the fundamental challenges of measuring effectiveness;
GoLaxy's ties to Chinese intelligence agencies;
operations targeting Hong Kong's…

Go with Your Gut: Scaling Confidence for Autoregressive Image Generation
Harold Haodong Chen, Xianfeng Wu, Wen-Jie Shu, Rongjin Guo, Disen Lan, Harry Yang, Ying-Cong Chen
https://arxiv.org/abs/2509.26376

Go with Your Gut: Scaling Confidence for Autoregressive Image Generation
Test-time scaling (TTS) has demonstrated remarkable success in enhancing large language models, yet its application to next-token prediction (NTP) autoregressive (AR) image generation remains largely uncharted. Existing TTS approaches for visual AR (VAR), which rely on frequent partial decoding and external reward models, are ill-suited for NTP-based image generation due to the inherent incompleteness of intermediate decoding results. To bridge this gap, we introduce ScalingAR, the first TTS fr…

LatentEvolve: Self-Evolving Test-Time Scaling in Latent Space
Guibin Zhang, Fanci Meng, Guancheng Wan, Zherui Li, Kun Wang, Zhenfei Yin, Lei Bai, Shuicheng Yan
https://arxiv.org/abs/2509.24771

LatentEvolve: Self-Evolving Test-Time Scaling in Latent Space
Test-time Scaling (TTS) has been demonstrated to significantly enhance the reasoning capabilities of Large Language Models (LLMs) during the inference phase without altering model parameters. However, existing TTS methods are largely independent, implying that LLMs have not yet evolved to progressively learn how to scale more effectively. With the objective of evolving LLMs to learn ``how to scale test-time computation,'' we propose LatentEvolve, a self-evolving latent TTS framework inspired by…

From my LinkedIn post: “Telling your dev team to use AI coding tools is like telling your 2010 ops team to use AWS. They didn’t know how to code, they were ticket and click-it VMware people… developers who don’t have product management mindset or have never managed a dev team will fail by trying to micromanage the output of the tool rather than specifying the outcome of the product and managing the agent team to deliver that outcome.”

Back in 2010 when we first started using AWS to run Netflix, the AWS services were very incomplete, somewhat unreliable, had scaling issues, but were super fast to provision compared to filing… | Adrian Cockcroft
Back in 2010 when we first started using AWS to run Netflix, the AWS services were very incomplete, somewhat unreliable, had scaling issues, but were super fast to provision compared to filing tickets and waiting for data center ops teams to do what we need. We worked around the issues, built a new cloud native developer driven architecture and scaled Netflix at speed. At the time many people said we were crazy to run a large enterprise on cloud and would be back in datacenters when it failed. …

SecInfer: Preventing Prompt Injection via Inference-time Scaling
Yupei Liu, Yanting Wang, Yuqi Jia, Jinyuan Jia, Neil Zhenqiang Gong
https://arxiv.org/abs/2509.24967 https://

SecInfer: Preventing Prompt Injection via Inference-time Scaling
Prompt injection attacks pose a pervasive threat to the security of Large Language Models (LLMs). State-of-the-art prevention-based defenses typically rely on fine-tuning an LLM to enhance its security, but they achieve limited effectiveness against strong attacks. In this work, we propose \emph{SecInfer}, a novel defense against prompt injection attacks built on \emph{inference-time scaling}, an emerging paradigm that boosts LLM capability by allocating more compute resources for reasoning dur…

Metadata-Guided Adaptable Frequency Scaling across Heterogeneous Applications and Devices
Jinqi Yan, Fang He, Qianlong Sang, Bifeng Tong, Peng Sun, Yili Gong, Chuang Hu, Dazhao Cheng
https://arxiv.org/abs/2509.22707

Metadata-Guided Adaptable Frequency Scaling across Heterogeneous Applications and Devices
Dynamic Voltage and Frequency Scaling is essential for enhancing energy efficiency in mobile platforms. However, traditional heuristic-based governors are increasingly inadequate for managing the complexity of heterogeneous System-on-Chip designs and diverse application workloads. Although reinforcement learning approaches offer improved performance, their poor generalization capability and reliance on extensive retraining for each hardware and application combination leads to significant deplo…

Are neural scaling laws leading quantum chemistry astray?
Siwoo Lee, Adji Bousso Dieng
https://arxiv.org/abs/2509.26397 https://arxiv.org/pdf/2509.26397

Are neural scaling laws leading quantum chemistry astray?
Neural scaling laws are driving the machine learning community toward training ever-larger foundation models across domains, assuring high accuracy and transferable representations for extrapolative tasks. We test this promise in quantum chemistry by scaling model capacity and training data from quantum chemical calculations. As a generalization task, we evaluate the resulting models' predictions of the bond dissociation energy of neutral H$_2$, the simplest possible molecule. We find that, reg…

Scaling rules for multiphase flow in vertical Venturis
Meng-Ke Zhan, Cheng-Gang Xie, Jian-Jun Shu
https://arxiv.org/abs/2509.23786 https://arxiv.org/pdf/25…

Scaling rules for multiphase flow in vertical Venturis
Employing a suitable scaling rule in gas-liquid flow can produce dynamically comparable results, which helps in the development of flow models applicable to a wide range of flow conditions and reduces the carbon footprint; however, matching all dimensionless numbers in gas-liquid flow is a challenge. This study uses a computational fluid dynamics approach to identify key dimensionless numbers that can produce dimensionally equivalent results under a variety of flow conditions in the vertical Ve…

Scaling Accessibility Education: Reflections from a Workshop Targeting CS Educators and Software Professionals
P D Parthasarathy, Anshu M Mittal, Swaroop Joshi
https://arxiv.org/abs/2509.22759

Scaling Accessibility Education: Reflections from a Workshop Targeting CS Educators and Software Professionals
Despite growing global attention to digital accessibility, research from India highlights a significant gap in accessibility training for both computing educators and software professionals. To address this need, we designed and conducted an experiential workshop aimed at building foundational capacity in accessibility practices among 77 participants, including computer science (CS) faculty and industry practitioners. The one-day workshop combined hands-on activities, tool demonstrations, and c…

Cambridge, UK-based Paragraf, which makes graphene-based electronics using standard semiconductor processes, raised a $55M Series C to scale its manufacturing (Tamara Djurickovic/Tech.eu)
https://tech.eu/2025/08/25/paragraf-closes-55m-series-c-funding-round/…

Paragraf closes $55M Series C funding round
Paragraf, the UK-based company leading the way in mass-producing graphene-based electronics using industry-standard semiconductor processes, has completed a $55 million Series C funding round. The funding will speed up the scaling of Paragraf’s manufacturing capabilities and increase production capacity, paving the way for graphene electronics to enter mass markets.Paragraf’s technology fits smoothly into the existing semiconductor ecosystem, offering customers ready-to-use solutions…

Geometry-induced criticality in $p$-adic scaling limits of random walks
Rahul Rajkumar, David Weisbart
https://arxiv.org/abs/2509.24234 https://arxiv.org/p…

Geometry-induced criticality in $p$-adic scaling limits of random walks
An anisotropy parameter $h$ in $(0,1]$ induces on $\mathbb{Q}_p^2$ a duality-compatible, two-scale filtration that collapses to one scale at the right endpoint. This filtration defines shell-uniform transition laws for hierarchical random walks on a discrete group whose scaling limits are Lévy processes on $\mathbb{Q}_p^2$. The diffusion constants of the coordinate processes jump at the right endpoint, even though the radial jump law depends continuously on $h$. This instance of geometry-induc…

SIRI: Scaling Iterative Reinforcement Learning with Interleaved Compression
Haoming Wen, Yushi Bai, Juanzi Li, Jie Tang
https://arxiv.org/abs/2509.25176 https://

SIRI: Scaling Iterative Reinforcement Learning with Interleaved Compression
We introduce SIRI, Scaling Iterative Reinforcement Learning with Interleaved Compression, a simple yet effective RL approach for Large Reasoning Models (LRMs) that enables more efficient and accurate reasoning. Existing studies have observed repetitive thinking patterns in LRMs, and attempts to reduce them often come at the cost of performance. In this paper, we show that this trade-off can be overcome through a training regime that iteratively alternates between compressing and expanding the r…

MGM-Omni: Scaling Omni LLMs to Personalized Long-Horizon Speech
Chengyao Wang, Zhisheng Zhong, Bohao Peng, Senqiao Yang, Yuqi Liu, Haokun Gui, Bin Xia, Jingyao Li, Bei Yu, Jiaya Jia
https://arxiv.org/abs/2509.25131

MGM-Omni: Scaling Omni LLMs to Personalized Long-Horizon Speech
We present MGM-Omni, a unified Omni LLM for omni-modal understanding and expressive, long-horizon speech generation. Unlike cascaded pipelines that isolate speech synthesis, MGM-Omni adopts a "brain-mouth" design with a dual-track, token-based architecture that cleanly decouples multimodal reasoning from real-time speech generation. This design enables efficient cross-modal interaction and low-latency, streaming speech generation. For understanding, a unified training strategy coupled with a du…

Scaling Equilibrium Propagation to Deeper Neural Network Architectures
Sankar Vinayak. E. P, Gopalakrishnan Srinivasan
https://arxiv.org/abs/2509.26003 https://

Scaling Equilibrium Propagation to Deeper Neural Network Architectures
Equilibrium propagation has been proposed as a biologically plausible alternative to the backpropagation algorithm. The local nature of gradient computations, combined with the use of convergent RNNs to reach equilibrium states, make this approach well-suited for implementation on neuromorphic hardware. However, previous studies on equilibrium propagation have been restricted to networks containing only dense layers or relatively small architectures with a few convolutional layers followed by a…

"Chart: Check out how fast China is shifting to renewables"
#China #Energy #Renewables

Chart: Check out how fast China is shifting to renewables
China still emits more planet-warming pollution than any other country, but its furious build-out of wind and solar means the nation is scaling back on…

Scaling of the Electrical Conductivity Spectra Reveals Distinct Transport Responses in A2SmTaO6 [A = Ba, Sr, Ca]
Saswata Halder, Binita Ghosh, T. P. Sinha
https://arxiv.org/abs/2508.21621

Scaling of the Electrical Conductivity Spectra Reveals Distinct Transport Responses in A2SmTaO6 [A = Ba, Sr, Ca]
Disorder plays an important role in materials science, influencing material behavior across different length scales. Imperfections like vacancies, atomic substitutions, lattice distortions, and microstructural inhomogeneities, disrupt ideal periodicity thereby altering physical properties. Analogous to spin-glass systems, electrical 'glassiness' arises when charge carriers confront disordered energy landscapes, leading to a broad range of relaxation times, especially in polycrystalline material…

Cosmic domain walls on a lattice: illusive effects of initial conditions
I. Dankovsky, S. Ramazanov, E. Babichev, D. Gorbunov, A. Vikman
https://arxiv.org/abs/2509.25367 https:/…

Cosmic domain walls on a lattice: illusive effects of initial conditions
Evolution of cosmic domain walls (DWs) settles to the scaling solution, which is often assumed to be independent of initial conditions. However, lattice simulations performed in this work reveal a clear dependence of the scaling DW area on the initial configuration of the sourcing scalar field, specifically, its infrared (IR) properties. Namely, the DW area grows as one suppresses IR modes in the initial scalar field spectrum. This growth is saturated, when the area parameter $ξ$ commonly used…

Scaling LLM Test-Time Compute with Mobile NPU on Smartphones
Zixu Hao, Jianyu Wei, Tuowei Wang, Minxing Huang, Huiqiang Jiang, Shiqi Jiang, Ting Cao, Ju Ren
https://arxiv.org/abs/2509.23324

Scaling LLM Test-Time Compute with Mobile NPU on Smartphones
Deploying Large Language Models (LLMs) on mobile devices faces the challenge of insufficient performance in smaller models and excessive resource consumption in larger ones. This paper highlights that mobile Neural Processing Units (NPUs) have underutilized computational resources, particularly their matrix multiplication units, during typical LLM inference. To leverage this wasted compute capacity, we propose applying parallel test-time scaling techniques on mobile NPUs to enhance the performa…

Computing Linear Combinations of $\varphi$-Function Actions for Exponential Integrators
Awad H. Al-Mohy
https://arxiv.org/abs/2509.26475 https://arxiv.org/…

Computing Linear Combinations of $φ$-Function Actions for Exponential Integrators
We propose a matrix-free algorithm for evaluating linear combinations of $φ$-function actions, $w_i := \sum_{j=0}^{p} α_i^{\,j}\,φ_j(t_i A)v_j$ for $i=1\colon r$, arising in exponential integrators. The method combines the scaling and recovering method with a truncated Taylor series, choosing a spectral shift and a scaling parameter by minimizing a power-based objective of the shifted operator. Accuracy is user-controlled and ultimately limited by the working precision. The algorithm decoupl…

Power-Law Spectra and Asymptotic $\omega/T$ Scaling in the Orbital-Selective Mott Phase of a Three-Orbital Hubbard Model
Fabian Eickhoff
https://arxiv.org/abs/2509.23758 https:/…

Power-Law Spectra and Asymptotic $ω/T$ Scaling in the Orbital-Selective Mott Phase of a Three-Orbital Hubbard Model
Quantum materials whose properties lie beyond the celebrated Landau Fermi-liquid paradigm have been observed for decades across diverse material platforms. Finding microscopic lattice models for metallic states that exhibit such peculiar behavior remains a major theoretical challenge, as these features often originate from strong quantum fluctuations in strongly interacting electron systems. Here we investigate a three-orbital Hubbard model at a high-symmetry point that hosts a transition from …

Scaling Generalist Data-Analytic Agents
Shuofei Qiao, Yanqiu Zhao, Zhisong Qiu, Xiaobin Wang, Jintian Zhang, Zhao Bin, Ningyu Zhang, Yong Jiang, Pengjun Xie, Fei Huang, Huajun Chen
https://arxiv.org/abs/2509.25084

Scaling Generalist Data-Analytic Agents
Data-analytic agents are emerging as a key catalyst for automated scientific discovery and for the vision of Innovating AI. Current approaches, however, rely heavily on prompt engineering over proprietary models, while open-source models struggle to face diverse-format, large-scale data files and long-horizon, multi-step reasoning that real-world analytics demands. This paper introduces DataMind, a scalable data synthesis and agent training recipe designed to build generalist data-analytic agen…

Scaling with Collapse: Efficient and Predictable Training of LLM Families
Shane Bergsma, Bin Claire Zhang, Nolan Dey, Shaheer Muhammad, Gurpreet Gosal, Joel Hestness
https://arxiv.org/abs/2509.25087

Scaling with Collapse: Efficient and Predictable Training of LLM Families
Effective LLM training relies on *consistency*, meaning that key quantities -- such as final losses and optimal hyperparameters -- scale predictably across model sizes. Qiu et al. (2025) recently showed that this consistency extends beyond scalars: whole training loss curves can *collapse* onto a universal trajectory after a simple normalization. What remains unclear is whether this phenomenon holds for LLM families trained under *practical scaling recipes*, where width, depth, learning rate, b…

🚀 AI's Energy Crisis?
#AI is exploding, but scaling compute with more data & power is too slow and costly—data centers are building nuclear plants! We need smarter ways to turn energy into intelligence.
🧠 Nature-Inspired Innovation At #Extropic rethinks hardware: Probabilistic…

Transport Model Investigation of NCQ Scaling and the Hadronic-Partonic Transition in Au Au Collisions
Li-Ke Liu, Shusu Shi
https://arxiv.org/abs/2508.21577 https://

Transport Model Investigation of NCQ Scaling and the Hadronic-Partonic Transition in Au+Au Collisions
We investigate elliptic flow ($v_2$) in Au+Au collisions at $\sqrt{s_{\text{NN}}} = 3.0$ and 4.5 GeV using hadronic and partonic transport models, including JAM, SMASH, AMPT-Hadronic Cascade, and AMPT-String Melting. At 3.0 GeV, hadronic models reproduce the centrality dependence of $v_2$ and the particle ordering ($K^0_S > p > π^+$), while violating number-of-constituent-quark (NCQ) scaling, indicating that collectivity is primarily driven by hadronic interactions and influenced by spectator …

Scaling Fabric-Based Piezoresistive Sensor Arrays for Whole-Body Tactile Sensing
Curtis C. Johnson, Daniel Webb, David Hill, Marc D. Killpack
https://arxiv.org/abs/2508.20959 ht…

Scaling Fabric-Based Piezoresistive Sensor Arrays for Whole-Body Tactile Sensing
Scaling tactile sensing for robust whole-body manipulation is a significant challenge, often limited by wiring complexity, data throughput, and system reliability. This paper presents a complete architecture designed to overcome these barriers. Our approach pairs open-source, fabric-based sensors with custom readout electronics that reduce signal crosstalk to less than 3.3% through hardware-based mitigation. Critically, we introduce a novel, daisy-chained SPI bus topology that avoids the practi…

Replaced article(s) found for cs.AI. https://arxiv.org/list/cs.AI/new
[7/9]:
- Scaling RL to Long Videos
Chen, Huang, Shi, Hu, Ye, Zhu, Liu, Molchanov, Kautz, Qi, Liu, Yin, Lu, Han

Homogeneity
Martin Himmel
https://arxiv.org/abs/2509.25227 https://arxiv.org/pdf/2509.25227

Homogeneity
The four types of homogeneity - additive, multiplicative, exponential, and logarithmic - are generalized as transformations describing how a function $f$ changes under scaling or shifting of its arguments. These generalized homogeneity functions capture different scaling behaviors and establish fundamental properties. Such properties include how homogeneity is preserved under function operations and how it determines the transformation behavior of related constructions like quotient functions…

One of my professors asked me today how I would balance the principle of least privilege against efficiency and ease of use in a large organization with thousands of employees.
The best answer I could come up with is that any single organization that size is an unholy abomination, and that I would move mountains to not be in that position.
Seriously, I'm sick of the question "How can this be scaled?" The better question is, "Would scaling <thing> be a net…

Towards Repository-Level Program Verification with Large Language Models
Si Cheng Zhong, Xujie Si
https://arxiv.org/abs/2509.25197 https://arxiv.org/pdf/25…

Towards Repository-Level Program Verification with Large Language Models
Recent advancements in large language models (LLMs) suggest great promises in code and proof generations. However, scaling automated formal verification to real-world projects requires resolving cross-module dependencies and global contexts, which are crucial challenges overlooked by existing LLM-based methods with a special focus on targeting isolated, function-level verification tasks. To systematically explore and address the significant challenges of verifying entire software repositories, …

MUSS-TI: Multi-level Shuttle Scheduling for Large-Scale Entanglement Module Linked Trapped-Ion
Xian Wu, Chenghong Zhu, Jingbo Wang, Xin Wang
https://arxiv.org/abs/2509.25988 htt…

MUSS-TI: Multi-level Shuttle Scheduling for Large-Scale Entanglement Module Linked Trapped-Ion
Trapped-ion computing is a leading architecture in the pursuit of scalable and high fidelity quantum systems. Modular quantum architectures based on photonic interconnects offer a promising path for scaling trapped ion devices. In this design, multiple Quantum Charge Coupled Device (QCCD) units are interconnected through entanglement module. Each unit features a multi-zone layout that separates functionalities into distinct areas, enabling more efficient and flexible quantum operations. However…

A theoretical guarantee for SyncRank
Yang Rao
https://arxiv.org/abs/2509.22766 https://arxiv.org/pdf/2509.22766

A theoretical guarantee for SyncRank
We present a theoretical and empirical analysis of the SyncRank algorithm for recovering a global ranking from noisy pairwise comparisons. By adopting a complex-valued data model where the true ranking is encoded in the phases of a unit-modulus vector, we establish a sharp non-asymptotic recovery guarantee for the associated semidefinite programming (SDP) relaxation. Our main theorem characterizes a critical noise threshold - scaling as sigma = O(sqrt(n / log n)) - below which SyncRank achieves…

🧵 Scaling asyncio on Free-Threaded Python
#python

Scaling asyncio on Free-Threaded Python
A recap on the work done in Python 3.14 to enable asyncio to scale on the free-threaded build of CPython.

Scaling Spoken Language Models with Syllabic Speech Tokenization
Nicholas Lee, Cheol Jun Cho, Alan W Black, Gopala K. Anumanchipalli
https://arxiv.org/abs/2509.26634 https://

Scaling Spoken Language Models with Syllabic Speech Tokenization
Spoken language models (SLMs) typically discretize speech into high-frame-rate tokens extracted from SSL speech models. As the most successful LMs are based on the Transformer architecture, processing these long token streams with self-attention is expensive, as attention scales quadratically with sequence length. A recent SSL work introduces acoustic tokenization of speech at the syllable level, which is more interpretable and potentially more scalable with significant compression in token len…

Breaking Universality in the Lower Order Terms in the 1-level and 2-level Density of Holomorphic Cusp Newforms
Lawrence Dillon, Xiaoyao Huang, Say-Yeon Kwon, Meiling Laurence, Steven J. Miller, Vishal Muthuvel, Luke Rowen, Pramana Saldin, Steven Zanetti
https://arxiv.org/abs/2508.21691

Breaking Universality in the Lower Order Terms in the 1-level and 2-level Density of Holomorphic Cusp Newforms
The Katz-Sarnak density conjecture states that, as the analytic conductor $R \to \infty$, the distribution of the normalized low-lying zeros (those near the central point $s = 1/2$) converges to the scaling limits of eigenvalues clustered near 1 of subgroups of $U(N)$. There is extensive evidence supporting this conjecture for many families, including the family of holomorphic cusp newforms. Interestingly, there are very few choices for the main term of the limiting behavior. In 2009, S. J. Mil…

Recursive Self-Aggregation Unlocks Deep Thinking in Large Language Models
Siddarth Venkatraman, Vineet Jain, Sarthak Mittal, Vedant Shah, Johan Obando-Ceron, Yoshua Bengio, Brian R. Bartoldson, Bhavya Kailkhura, Guillaume Lajoie, Glen Berseth, Nikolay Malkin, Moksh Jain
https://arxiv.org/abs/2509.26626…

Recursive Self-Aggregation Unlocks Deep Thinking in Large Language Models
Test-time scaling methods improve the capabilities of large language models (LLMs) by increasing the amount of compute used during inference to make a prediction. Inference-time compute can be scaled in parallel by choosing among multiple independent solutions or sequentially through self-refinement. We propose Recursive Self-Aggregation (RSA), a test-time scaling method inspired by evolutionary methods that combines the benefits of both parallel and sequential scaling. Each step of RSA refines…

Long-range minimal models
Connor Behan, Dario Benedetti, Fanny Eustachon, Edoardo Lauria
https://arxiv.org/abs/2509.26372 https://arxiv.org/pdf/2509.26372

Long-range minimal models
We study a class of nonlocal conformal field theories in two dimensions which are obtained as deformations of the Virasoro minimal models. The construction proceeds by coupling a relevant primary operator $ϕ_{r,s}$ of the $m$-th minimal model to a generalized free field, in such a way that the interaction term has scaling dimension $2-δ$. Flowing to the infrared, we reach a new class of CFTs that we call long-range minimal models. In the case $r=s=2$, the resulting line of fixed points, param…

If AI chatbot companies truly had what they claim they have (arbitrary scaling human-level intelligence)—they would use it exclusively themselves, prompting it to come up with schemes to make money and execute them.
In reality these companies all lose money (in historically unprecedented amounts), to fuel a drug dealer-like approach by giving it away for free and hoping enough people get addicted to sycophantic chatbots; with the goal to charge exorbitant fees for it in the future.

Addressing Methodological Uncertainty in MCDM with a Systematic Pipeline Approach to Data Transformation Sensitivity Analysis
Juan B. Cabral, Alvaro Roy Schachner
https://arxiv.org/abs/2509.24996

Addressing Methodological Uncertainty in MCDM with a Systematic Pipeline Approach to Data Transformation Sensitivity Analysis
Multicriteria decision-making methods exhibit critical dependence on the choice of normalization techniques, where different selections can alter 20-40% of the final rankings. Current practice is characterized by the ad-hoc selection of methods without systematic robustness evaluation. We present a framework that addresses this methodological uncertainty through automated exploration of the scaling transformation space. The implementation leverages the existing Scikit-Criteria infrastructure to…

Characteristic tensors for almost Finsler manifolds
James F. Davis, Benjamin R. Edwards, Alan Kostelecky
https://arxiv.org/abs/2508.21744 https://arxiv.org…

Characteristic tensors for almost Finsler manifolds
Almost Finsler manifolds and partial Finsler manifolds are introduced, extending the standard definition of a Finsler manifold to allow for a nontrivial slit containing points fixed under homogeneous scaling and for metrics where the fundamental tensor has nonpositive eigenvalues. The bipartite spaces offer examples of comparatively simple almost Finsler manifolds and partial Finsler manifolds with physics applications. Special cases are the $\bf{a}$ and $\bf{b}$ spaces, which have almost Finsl…

VerilogMonkey: Exploring Parallel Scaling for Automated Verilog Code Generation with LLMs
Juxin Niu, Yuxin Du, Dan Niu, Xi Wang, Zhe Jiang, Nan Guan
https://arxiv.org/abs/2509.16246

VerilogMonkey: Exploring Parallel Scaling for Automated Verilog Code Generation with LLMs
We present VerilogMonkey, an empirical study of parallel scaling for the under-explored task of automated Verilog generation. Parallel scaling improves LLM performance by sampling many outputs in parallel. Across multiple benchmarks and mainstream LLMs, we find that scaling to hundreds of samples is cost-effective in both time and money and, even without any additional enhancements such as post-training or agentic methods, surpasses prior results on LLM-based Verilog generation. We further diss…

Scalable Boltzmann Generators for equilibrium sampling of large-scale materials
Maximilian Schebek, Jutta Rogal
https://arxiv.org/abs/2509.25486 https://ar…

Scalable Boltzmann Generators for equilibrium sampling of large-scale materials
The use of generative models to sample equilibrium distributions of many-body systems, as first demonstrated by Boltzmann Generators, has attracted substantial interest due to their ability to produce unbiased and uncorrelated samples in `one shot'. Despite their promise and impressive results across the natural sciences, scaling these models to large systems remains a major challenge. In this work, we introduce a Boltzmann Generator architecture that addresses this scalability bottleneck with …

Probing the Star Formation Main Sequence down to 10$^{7} M_\odot$ at $1 < z < 9$
Rosa M. M\'erida, Marcin Sawicki, Kartheik G. Iyer, Ga\"el Noirot, Chris J. Willott, Maru\v{s}a Brada\v{c}, Guillaume Desprez, Nicholas S. Martis, Adam Muzzin, Gregor Rihtar\v{s}i\v{c}, Ghassan T. E. Sarrouh, Jeremy Favaro, Gaia Gaspar, Anishya Harshan, Jon Jude\v{z}

Probing the Star Formation Main Sequence down to 10$^{7} M_\odot$ at $1 < z < 9$
The Main Sequence of Star-Forming Galaxies (SFGMS or MS) is a fundamental scaling relation that provides a global framework for studying galaxy formation and evolution, as well as insight into the complex star formation histories (SFHs) of individual galaxies. In this work, we combine large-area pre-JWST surveys (COSMOS2020, CANDELS), which probe high-$M_\star$ sources (${>10^9\,M_\odot}$), with SHARDS/CANDELS FAINT and JWST data from CANUCS, CEERS, JADES, and UNCOVER, to obtain a high-$z$, sta…

Quantifying non-equilibrium pressure-gradient turbulent boundary layers through a symmetry-based framework
Wei-Tao Bi, Ke-Xin Zheng, Jun Chen, Zhen-Su She
https://arxiv.org/abs/2508.21447

Quantifying non-equilibrium pressure-gradient turbulent boundary layers through a symmetry-based framework
This study establishes a symmetry-based framework to quantify non-equilibrium processes in complex pressure gradient (PG) turbulent boundary layers (TBLs), using a Lie-group-informed dilation-symmetry-breaking formalism. We derive a universal multilayer defect scaling law for the evolution of total shear stress (TSS). The law shows that gradually varying adverse pressure gradients (APGs) break the dilation symmetry in the two-layer defect scaling of equilibrium TSS, leading to three-layer TSS s…

Crosslisted article(s) found for cs.MM. https://arxiv.org/list/cs.MM/new
[2/2]:
- MGM-Omni: Scaling Omni LLMs to Personalized Long-Horizon Speech
Wang, Zhong, Peng, Yang, Liu, Gui, Xia, Li, Yu, Jia

Anti-hyperuniform Critical States of Active Topological Defects
Simon Guldager Andersen, Tianxiang Ma, Makito F. Katsume, Kexin Li, Xiao Liu, Martin Cramer Pedersen, Amin Doostmohammadi
https://arxiv.org/abs/2509.22911

Anti-hyperuniform Critical States of Active Topological Defects
Topological defects are fundamental to the collective dynamics of non-equilibrium systems and in active matter, mediating spontaneous flows, dynamic self-organization, and emergent pattern formation. Here, we reveal critical states in active nematics, marked by slowed defect density relaxation, amplified fluctuations, and heightened sensitivity to activity. Near criticality, defect interactions become long-ranged, scaling with system size, and the system enters an anti-hyperuniform regime with …

Are We Scaling the Right Thing? A System Perspective on Test-Time Scaling
Youpeng Zhao, Jinpeng LV, Di Wu, Jun Wang, Christopher Gooley
https://arxiv.org/abs/2509.19645 https://…

Are We Scaling the Right Thing? A System Perspective on Test-Time Scaling
Test-time scaling (TTS) has recently emerged as a promising direction to exploit the hidden reasoning capabilities of pre-trained large language models (LLMs). However, existing scaling methods narrowly focus on the compute-optimal Pareto-frontier, ignoring the simple fact that compute-optimal is not always system-optimal. In this work, we propose a system-driven perspective on TTS, analyzing how reasoning models scale against practical metrics, such as latency and cost-per-token. By evaluating…

A critical stochastic heat equation with long-range noise
Alexander Dunlap, Martin Hairer, Xue-Mei Li
https://arxiv.org/abs/2509.23790 https://arxiv.org/pd…

A critical stochastic heat equation with long-range noise
We consider a semilinear stochastic heat equation in spatial dimension at least $3$, forced by a noise that is white in time with a covariance kernel that decays like $\lvert x\rvert^{-2}$ as $\lvert x\rvert\to\infty$. We show that in an appropriate diffusive scaling limit with a logarithmic attenuation of the noise, the pointwise statistics of the solution can be approximated by the solution to a forward-backward stochastic differential equation (FBSDE). The scaling and structure of the proble…

Replaced article(s) found for cs.ET. https://arxiv.org/list/cs.ET/new
[1/1]:
- SOT-MRAM Bitcell Scaling with BEOL Read Selectors: A DTCO Study
Yang Xiang, et al.
…

Autoproof: Automated Segmentation Proofreading for Connectomics
Gary B Huang, William M Katz, Stuart Berg, Louis Scheffer
https://arxiv.org/abs/2509.26585 https://

Autoproof: Automated Segmentation Proofreading for Connectomics
Producing connectomes from electron microscopy (EM) images has historically required a great deal of human proofreading effort. This manual annotation cost is the current bottleneck in scaling EM connectomics, for example, in making larger connectome reconstructions feasible, or in enabling comparative connectomics where multiple related reconstructions are produced. In this work, we propose using the available ground-truth data generated by this manual annotation effort to learn a machine lear…

Observation of universal non-Gaussian statistics of the order parameter across a continuous phase transition
Maxime Allemand, G\'eraud Dupuy, Paul Paquiez, Nicolas Dupuis, Adam Ran\c{c}on, Tommaso Roscilde, Thomas Chalopin, David Cl\'ement
https://arxiv.org/abs/2508.21623

Observation of universal non-Gaussian statistics of the order parameter across a continuous phase transition
Second-order phase transitions are characterised by critical scaling and universality. The singular behaviour of thermodynamic quantities at the transition, in particular, is determined by critical exponents of the universality class of the transition. However, critical properties are also characterised by the probability distribution of order parameter across the transition, where non-Gaussian statistics are expected, but remain largely unexplored. Here, making use of single-atom-resolved dete…

Large Language Models for 3D IC Space Planning
Hung-Ying Chu, Guan-Wei Chen, Shao-Yu Wei, Yu-Cheng Lin
https://arxiv.org/abs/2509.22716 https://arxiv.org/p…

Large Language Models for 3D IC Space Planning
Three-dimensional integrated circuits (3D ICs) have emerged as a promising solution to the scaling limits of two-dimensional designs, offering higher integration density, shorter interconnects, and improved performance. As design complexity increases, effective space planning becomes essential to reduce dead space and ensure layout quality. This study investigates the use of large language models (LLMs) for 3D IC space planning through a post-order slicing tree representation, which guarantees …

Understanding Embedding Scaling in Collaborative Filtering
Zhuangzhuang He, Zhou Kaiyu, Haoyue Bai, Fengbin Zhu, Yonghui Yang
https://arxiv.org/abs/2509.15709 https://

Understanding Embedding Scaling in Collaborative Filtering
Scaling recommendation models into large recommendation models has become one of the most widely discussed topics. Recent efforts focus on components beyond the scaling embedding dimension, as it is believed that scaling embedding may lead to performance degradation. Although there have been some initial observations on embedding, the root cause of their non-scalability remains unclear. Moreover, whether performance degradation occurs across different types of models and datasets is still an un…

Advantage of utilizing nonlocal magic resource in Haar-random circuits
Xiao Huang, Guanhua Chen, Yao Yao
https://arxiv.org/abs/2509.26342 https://arxiv.org…

Advantage of utilizing nonlocal magic resource in Haar-random circuits
In design and simulation of quantum circuits with multiple units, the computational ability is greatly limited by quickly increasing entanglement, and the ordinary sampling methods normally exhibit low efficiency. Herein, we uncover an intrinsic scaling law of the nonlocal magic resource and the bond dimension of matrix product states in Haar-random quantum circuits, that is, the nonlocal magic resource is converged on a bond dimension in logarithmic scale with the system size. It means, in the…

Redesigning GROMACS Halo Exchange: Improving Strong Scaling with GPU-initiated NVSHMEM
Mahesh Doijade, Andrey Alekseenko, Ania Brown, Alan Gray, Szil\'ard P\'all
https://arxiv.org/abs/2509.21527

Redesigning GROMACS Halo Exchange: Improving Strong Scaling with GPU-initiated NVSHMEM
Improving time-to-solution in molecular dynamics simulations often requires strong scaling due to fixed-sized problems. GROMACS is highly latency-sensitive, with peak iteration rates in the sub-millisecond, making scalability on heterogeneous supercomputers challenging. MPI's CPU-centric nature introduces additional latencies on GPU-resident applications' critical path, hindering GPU utilization and scalability. To address these limitations, we present an NVSHMEM-based GPU kernel-initiated rede…

U-SWIFT: A Unified Surface Wave Inversion Framework with Transformer via Normalization of Dispersion Curves
Tianjian Cheng, Hongrui Xu, Jiayu Feng, Xiongyu Hu, Chaofan Yao
https://arxiv.org/abs/2509.24872

U-SWIFT: A Unified Surface Wave Inversion Framework with Transformer via Normalization of Dispersion Curves
Deep learning is an increasingly popular approach for inverting surface wave dispersion curves to obtain Vs profiles. However, its generalizability is constrained by the depth and velocity scales of training data. We propose a unified deep learning framework that overcomes this limitation via normalization of dispersion curves. By leveraging the scaling properties of dispersion curves, our approach enables a single, pre-trained model to predict Vs profiles across diverse scales, from shallow su…

Asymptotic Scaling Law Analysis of Multicast Satellite Communications with Massive MIMO
Seyong Kim, Jeonghun Park
https://arxiv.org/abs/2509.16921 https://…

Asymptotic Scaling Law Analysis of Multicast Satellite Communications with Massive MIMO
In this paper, we consider a geostationary orbit (GEO) satellite communication system that employs massive multiple-input multiple-output (MIMO) for multicast transmission. By modeling the spatial distribution of ground users using a Poisson point process (PPP) and assuming a fixed-beam precoding is adopted, we find a closed-form expression for the asymptotical rate scaling law as a function of the number of antennas and the scaling factors of user density and multicast users. From the derived …

Scaling Digital Twin Models
Deniz Karanfil, Bahram Ravani
https://arxiv.org/abs/2509.16209 https://arxiv.org/pdf/2509.16209

Scaling Digital Twin Models
In many industries, the scale and complexity of systems can present significant barriers to the development of accurate digital twin models. This paper introduces a novel methodology and a modular computational tool utilizing machine learning and dimensional analysis to establish a framework for scaling digital twin models. Scaling techniques have not yet been applied to digital twin technology, but they can eliminate the need for repetitive physical calibration of such models in industries whe…

Linear-scaling calculation of experimental observables for molecular augmented dynamics simulations
Tigany Zarrouk, Miguel A. Caro
https://arxiv.org/abs/2509.22388 https://

Linear-scaling calculation of experimental observables for molecular augmented dynamics simulations
Aligning theoretical atomistic structural models of materials with available experimental data presents a significant challenge for disordered systems. The configurational space to navigate is vast, and faithful realizations require large system sizes with quantum-mechanical accuracy in order to capture the distribution of structural motifs present in experiment. Traditional equilibrium sampling approaches offer no guarantee of generating structures that coincide with experimental data for such…

Crosslisted article(s) found for cs.CV. https://arxiv.org/list/cs.CV/new
[2/2]:
- Scaling Up Temporal Domain Generalization via Temporal Experts Averaging
Liu, Miller, Saligrama, Saenko, Gong, Lim, Plummer

Breakdown of Kolmogorov Scaling and Modified Energy Transfer in Bubble-Laden Turbulence
Andrea Montessori, Marco Lauricella, Aritra Mukherjee, Luca Brandt
https://arxiv.org/abs/2509.22324

Breakdown of Kolmogorov Scaling and Modified Energy Transfer in Bubble-Laden Turbulence
We investigate the effect of a dispersed bubble phase on forced homogeneous and isotropic turbulence using high-resolution high-performance simulations based on the lattice Boltzmann method. While the classical Kolmogorov energy cascade is largely preserved when considering the system as a whole, a phase-specific analysis reveals striking deviations from the classical turbulence scaling. In particular, the gas phase exhibits significant departures from Kolmogorov's predictions, whereas the cont…

High Energy Particle Production from Proton Synchrotron Radiation in Strong Magnetic Fields in Relativistic Quantum Field Theory
Tomoyuki Maruyama, A. Baha Balantekin, Myung-Ki Cheoun, Akira Dohi, Ryo Higuch, Toshitaka Kajino, Grant J. Mathews
https://arxiv.org/abs/2509.24366

High Energy Particle Production from Proton Synchrotron Radiation in Strong Magnetic Fields in Relativistic Quantum Field Theory
We investigate photon, pion, and rho-meson production from proton synchrotron radiation in the presence of strong magnetic fields. The proton decay widths and the luminosities of the emitted particles are calculated within a relativistic quantum framework that incorporates Landau quantization. A scaling rule is derived for the transition probability between different Landau levels. This allows an evaluation of transitions for extremely high Landau numbers exceeding $10^{15}$. Furthermore, we ca…

Can Structured Templates Facilitate LLMs in Tackling Harder Tasks? : An Exploration of Scaling Laws by Difficulty
Zhichao Yang, Zhaoxin Fan, Gen Li, Yuanze Hu, Xinyu Wang, Ye Qiu, Xin Wang, Yifan Sun, Wenjun Wu
https://arxiv.org/abs/2508.19069

Can Structured Templates Facilitate LLMs in Tackling Harder Tasks? : An Exploration of Scaling Laws by Difficulty
Structured, procedural reasoning is essential for Large Language Models (LLMs), especially in mathematics. While post-training methods have improved LLM performance, they still fall short in capturing deep procedural logic on complex tasks. To tackle the issue, in this paper, we first investigate this limitation and uncover a novel finding: a Scaling Law by Difficulty, which reveals that model performance follows a U-shaped curve with respect to training data complexity -- excessive low-difficu…

Symmetry of concentration and scaling for self-bounding functions
George Crowley, I\~naki Esnaola
https://arxiv.org/abs/2509.22375 https://arxiv.org/pdf/25…

Symmetry of concentration and scaling for self-bounding functions
We prove generalised concentration inequalities for a class of scaled self-bounding functions of independent random variables, referred to as ${(M,a,b)}$ self-bounding. The scaling refers to the fact that the component-wise difference is upper bounded by an arbitrary positive real number $M$ instead of the case $M=1$ previously considered in the literature. Using the entropy method, we derive symmetric bounds for both the upper and lower tails, and study the tightness of the proposed bounds. Ou…

Conditional Multidimensional Scaling with Incomplete Conditioning Data
Anh Tuan Bui
https://arxiv.org/abs/2509.16627 https://arxiv.org/pdf/2509.16627

Conditional Multidimensional Scaling with Incomplete Conditioning Data
Conditional multidimensional scaling seeks for a low-dimensional configuration from pairwise dissimilarities, in the presence of other known features. By taking advantage of available data of the known features, conditional multidimensional scaling improves the estimation quality of the low-dimensional configuration and simplifies knowledge discovery tasks. However, existing conditional multidimensional scaling methods require full data of the known features, which may not be always attainable …

VLA-Reasoner: Empowering Vision-Language-Action Models with Reasoning via Online Monte Carlo Tree Search
Wenkai Guo, Guanxing Lu, Haoyuan Deng, Zhenyu Wu, Yansong Tang, Ziwei Wang
https://arxiv.org/abs/2509.22643

VLA-Reasoner: Empowering Vision-Language-Action Models with Reasoning via Online Monte Carlo Tree Search
Vision-Language-Action models (VLAs) achieve strong performance in general robotic manipulation tasks by scaling imitation learning. However, existing VLAs are limited to predicting short-sighted next-action, which struggle with long-horizon trajectory tasks due to incremental deviations. To address this problem, we propose a plug-in framework named VLA-Reasoner that effectively empowers off-the-shelf VLAs with the capability of foreseeing future states via test-time scaling. Specifically, VLA-…

A Quantum Computer Based on Donor-Cluster Arrays in Silicon
Shihang Zhang, Chunhui Zhang, Guanyong Wang, Tao Xin, Guangchong Hu, Yu He, Peihao Huang
https://arxiv.org/abs/2509.24749

A Quantum Computer Based on Donor-Cluster Arrays in Silicon
Significant advances in silicon spin qubits highlight the potential of silicon quantum dots for scalable quantum computing, given their compatibility with industrial fabrication and long coherence times. In particular, phosphorus (P)-doped spin qubits possess excellent coherence and have demonstrated high-fidelity two-qubit gates exceeding 99.9%. However, scaling P-donor systems is challenging due to crosstalk caused by the uniformity of individual P donors and the low tolerance for imprecise a…

Monte Carlo Diagonalization for Hubbard Model
B. Bernard, M. Charlebois
https://arxiv.org/abs/2509.22975 https://arxiv.org/pdf/2509.22975

Monte Carlo Diagonalization for Hubbard Model
The Hubbard model has often been studied with exact diagonalization (ED). This impurity solver is fundamentally limited by the exponential scaling of the Fock space. To address this problem, we introduce Monte Carlo diagonalization. Using a truncated Fock space constructed with a Monte Carlo approach, we reduce the size of the basis required to represent the Hamiltonian. We can then apply the Lanczos and band Lanczos algorithms in this truncated basis to find the ground state and the Green func…

Training Matryoshka Mixture-of-Experts for Elastic Inference-Time Expert Utilization
Yaoxiang Wang, Qingguo Hu, Yucheng Ding, Ruizhe Wang, Yeyun Gong, Jian Jiao, Yelong Shen, Peng Cheng, Jinsong Su
https://arxiv.org/abs/2509.26520

Training Matryoshka Mixture-of-Experts for Elastic Inference-Time Expert Utilization
Mixture-of-Experts (MoE) has emerged as a promising paradigm for efficiently scaling large language models without a proportional increase in computational cost. However, the standard training strategy of Top-K router prevents MoE models from realizing their full potential for elastic inference. When the number of activated experts is altered at inference time, these models exhibit precipitous performance degradation. In this work, we introduce Matryoshka MoE (M-MoE), a training framework that …

Exact infrared scaling behavior of Randers-Finsler scalar field theories
M. S. Mendes, J. F. S. Neto, R. F. Silva, H. A. S. Costa, P. R. S. Carvalho
https://arxiv.org/abs/2508.17533

Exact infrared scaling behavior of Randers-Finsler scalar field theories
We study the scaling behavior of Randers-Finsler massless scalar field theories in the infrared regime. For that, we compute analytically the radiative corrections to the corresponding anomalous dimensions, related to the critical exponents of the theory, first up to next-to-leading loop order and later for all-loop levels. We consider the effect of the Randers-Finsler space-time properties on the critical exponents by considering the parameter characterizing those space-times in its exact form…

Despite hype and optimistic projections, the humanoid robot industry faces hurdles, from battery life and design to limited demand for large-scale deployments (Evan Ackerman/IEEE Spectrum)
https://spectrum.ieee.org/humanoid-robot-scaling

Humanoid Robots: The Scaling Challenge
​It takes more than building a humanoid robot to build a humanoid robot product.

Universal critical dynamics near the chiral phase transition and the QCD critical point
Yunxin Ye, Johannes V. Roth, S\"oren Schlichting, Lorenz von Smekal
https://arxiv.org/abs/2509.26355

Universal critical dynamics near the chiral phase transition and the QCD critical point
We use a novel real-time formulation of the functional renormalization group (FRG) for dynamical systems with reversible mode couplings to study Model G and H, which are the conjectured dynamic universality classes of the two-flavor chiral phase transition and the QCD critical point, respectively. We compute the dynamic critical exponent in both models in spatial dimensions $2

FLAME: A Serving System Optimized for Large-Scale Generative Recommendation with Efficiency
Xianwen Guo, Bin Huang, Xiaomeng Wu, Guanlin Wu, Fangjian Li, Shijia Wang, Qiang Xiao, Chuanjiang Luo, Yong Li
https://arxiv.org/abs/2509.22681

FLAME: A Serving System Optimized for Large-Scale Generative Recommendation with Efficiency
Generative recommendation (GR) models possess greater scaling power compared to traditional deep learning recommendation models (DLRMs), yet they also impose a tremendous increase in computational burden. Measured in FLOPs, a typical GR model's workload sits in $10^9 \sim 10^{11}$ range, roughly four orders of magnitude higher than traditional DLRMs. Delivering accurate results in a few tens of milliseconds while processing billions of such requests per day puts extreme demands on the performan…

Crosslisted article(s) found for cs.LG. https://arxiv.org/list/cs.LG/new
[3/11]:
- HEART: Emotionally-driven test-time scaling of Language Models
Pinto, Goyal, Song, Chakraborty, Wang, Pfister, Palangi

Replaced article(s) found for cs.CL. https://arxiv.org/list/cs.CL/new
[2/2]:
- Trust but Verify! A Survey on Verification Design for Test-time Scaling
V Venktesh, Mandeep Rathee, Avishek Anand

SOT-MRAM Bitcell Scaling with BEOL Read Selectors: A DTCO Study
Yang Xiang, Fernando Garc\'ia-Redondo, Arvind Sharma, Van Dai Nguyen, Andrea Fantini, Philippe Matagne, Siddharth Rao, Subhali Subhechha, Lynn Verschueren, Mohammed Aftab Baig, Marie Garcia Bardon, Geert Hellings
https://arxiv.org/abs/2508.18250

SOT-MRAM Bitcell Scaling with BEOL Read Selectors: A DTCO Study
This work explores the cross-node scaling potential of SOT-MRAM for last-level caches (LLCs) under heterogeneous system scaling paradigm. We perform extensive Design-Technology Co-Optimization (DTCO) exercises to evaluate the bitcell footprint for different cell configurations at a representative 7 nm technology and to assess their implications on read and write power-performance. We crucially identify the MTJ routing struggle in conventional two-transistor one-resistor (2T1R) SOT-MRAMs as the …

Dynamic Experts Search: Enhancing Reasoning in Mixture-of-Experts LLMs at Test Time
Yixuan Han, Fan Ma, Ruijie Quan, Yi Yang
https://arxiv.org/abs/2509.22572 https://

Dynamic Experts Search: Enhancing Reasoning in Mixture-of-Experts LLMs at Test Time
Test-Time Scaling (TTS) enhances the reasoning ability of large language models (LLMs) by allocating additional computation during inference. However, existing approaches primarily rely on output-level sampling while overlooking the role of model architecture. In mainstream Mixture-of-Experts (MoE) LLMs, we observe that varying the number of activated experts yields complementary solution sets with stable accuracy, revealing a new and underexplored source of diversity. Motivated by this observa…

VideoChat-R1.5: Visual Test-Time Scaling to Reinforce Multimodal Reasoning by Iterative Perception
Ziang Yan, Xinhao Li, Yinan He, Zhengrong Yue, Xiangyu Zeng, Yali Wang, Yu Qiao, Limin Wang, Yi Wang
https://arxiv.org/abs/2509.21100

VideoChat-R1.5: Visual Test-Time Scaling to Reinforce Multimodal Reasoning by Iterative Perception
Inducing reasoning in multimodal large language models (MLLMs) is critical for achieving human-level perception and understanding. Existing methods mainly leverage LLM reasoning to analyze parsed visuals, often limited by static perception stages. This paper introduces Visual Test-Time Scaling (VTTS), a novel approach to enhance MLLMs' reasoning via iterative perception during inference. VTTS mimics humans' hierarchical attention by progressively refining focus on high-confidence spatio-tempora…

ZKProphet: Understanding Performance of Zero-Knowledge Proofs on GPUs
Tarunesh Verma (Computer Science and Engineering, University of Michigan, USA), Yichao Yuan (Computer Science and Engineering, University of Michigan, USA), Nishil Talati (Computer Science and Engineering, University of Michigan, USA), Todd Austin (Computer Science and Engineering, University of Michigan, USA)

ZKProphet: Understanding Performance of Zero-Knowledge Proofs on GPUs
Zero-Knowledge Proofs (ZKP) are protocols which construct cryptographic proofs to demonstrate knowledge of a secret input in a computation without revealing any information about the secret. ZKPs enable novel applications in private and verifiable computing such as anonymized cryptocurrencies and blockchain scaling and have seen adoption in several real-world systems. Prior work has accelerated ZKPs on GPUs by leveraging the inherent parallelism in core computation kernels like Multi-Scalar Mul…

Near-the-Axis Universality in Last Passage Percolation
Sam McKeown, Xinyi Zhang
https://arxiv.org/abs/2509.25059 https://arxiv.org/pdf/2509.25059

Near-the-Axis Universality in Last Passage Percolation
This note establishes a universal directed landscape limit for last passage percolation models in an intermediate scaling regime. We find as a quick consequence the transversal fluctuations for geodesics taken near the axis. We extend the technique of Bodineau and Martin, who in arXiv:math/0410042 have already shown universal one-point fluctuations in this regime.

Mitigating Strategy-Selection Bias in Reasoning for More Effective Test-Time Scaling
Zongqian Wu, Baoduo Xu, Tianyu Li, Zhu Sun, Xiaofeng Zhu, Lei Feng
https://arxiv.org/abs/2509.17905

Mitigating Strategy-Selection Bias in Reasoning for More Effective Test-Time Scaling
Test-time scaling (TTS) has been shown to improve the performance of large language models (LLMs) by sampling and aggregating diverse reasoning paths. However, existing research has overlooked a critical issue: selection bias of reasoning strategies during scaling. Specifically, when generating reasoning processes, LLMs tend to follow certain strategies (e.g., algebraic solutions for math problems) while neglecting other valid alternatives (e.g., geometric solutions), resulting in insufficient …

Logical Reasoning with Outcome Reward Models for Test-Time Scaling
Ramya Keerthy Thatikonda, Wray Buntine, Ehsan Shareghi
https://arxiv.org/abs/2508.19903 https://

Logical Reasoning with Outcome Reward Models for Test-Time Scaling
Logical reasoning is a critical benchmark for evaluating the capabilities of large language models (LLMs), as it reflects their ability to derive valid conclusions from given premises. While the combination of test-time scaling with dedicated outcome or process reward models has opened up new avenues to enhance LLMs performance in complex reasoning tasks, this space is under-explored in deductive logical reasoning. We present a set of Outcome Reward Models (ORMs) for deductive reasoning. To tra…

Mind the crosscap: $\tau$-scaling in non-orientable gravity and time-reversal-invariant systems
Gabriele Di Ubaldo, Altay Etkin, Felix M. Haehl, Moshe Rozali
https://arxiv.org/abs/2509.20448

Mind the crosscap: $τ$-scaling in non-orientable gravity and time-reversal-invariant systems
Spectral statistics of quantum chaotic systems are governed by random matrix universality. In many cases of interest, time-reversal symmetry selects the Gaussian Orthogonal Ensemble (GOE) as the relevant universality class. In holographic CFTs, this is mirrored by the presence of non-orientable geometries in the dual gravitational path integral. In this work, we analyze general properties of these matrix models and their gravitational counterparts. First, we develop a formalism to express the u…

Closed-form $\ell_r$ norm scaling with data for overparameterized linear regression and diagonal linear networks under $\ell_p$ bias
Shuofeng Zhang, Ard Louis
https://arxiv.org/abs/2509.21181

Closed-form $\ell_r$ norm scaling with data for overparameterized linear regression and diagonal linear networks under $\ell_p$ bias
For overparameterized linear regression with isotropic Gaussian design and minimum-$\ell_p$ interpolator $p\in(1,2]$, we give a unified, high-probability characterization for the scaling of the family of parameter norms $ \\{ \lVert \widehat{w_p} \rVert_r \\}_{r \in [1,p]} $ with sample size. We solve this basic, but unresolved question through a simple dual-ray analysis, which reveals a competition between a signal *spike* and a *bulk* of null coordinates in $X^\top Y$, yielding closed-form …

Local limit of focusing discrete NLS
Kesav Krishnan, Gourab Ray
https://arxiv.org/abs/2509.25759 https://arxiv.org/pdf/2509.25759

Local limit of focusing discrete NLS
We examine the behavior of a function sampled from the invariant measure associated to the focusing discrete Non Linear Schrödinger equation, defined on a discrete torus of dimension $d \geq 3$, and nonlinearity parameter $p>4$, in the infinite volume limit. The Gibbs measure has two parameters, the inverse temperature and the strength of the non linearity, and the scaling is such that the non linear and linear parts of the Hamiltonian contribute on the same scale. It was shown by Dey, Kirkpat…

ScaleDiff: Scaling Difficult Problems for Advanced Mathematical Reasoning
Qizhi Pei, Zhuoshi Pan, Honglin Lin, Xin Gao, Yu Li, Zinan Tang, Conghui He, Rui Yan, Lijun Wu
https://arxiv.org/abs/2509.21070

ScaleDiff: Scaling Difficult Problems for Advanced Mathematical Reasoning
Large Reasoning Models (LRMs) have shown impressive capabilities in complex problem-solving, often benefiting from training on difficult mathematical problems that stimulate intricate reasoning. Recent efforts have explored automated synthesis of mathematical problems by prompting proprietary models or large-scale open-source models from seed data or inherent mathematical concepts. However, scaling up these methods remains challenging due to their high computational/API cost, complexity of prom…

Investigating Test-Time Scaling with Reranking for Machine Translation
Shaomu Tan, Ryosuke Mitani, Ritvik Choudhary, Toshiyuki Sekiya
https://arxiv.org/abs/2509.19020 https://…

The point is the mask: scaling coral reef segmentation with weak supervision
Matteo Contini, Victor Illien, Sylvain Poulain, Serge Bernard, Julien Barde, Sylvain Bonhommeau, Alexis Joly
https://arxiv.org/abs/2508.18958

The point is the mask: scaling coral reef segmentation with weak supervision
Monitoring coral reefs at large spatial scales remains an open challenge, essential for assessing ecosystem health and informing conservation efforts. While drone-based aerial imagery offers broad spatial coverage, its limited resolution makes it difficult to reliably distinguish fine-scale classes, such as coral morphotypes. At the same time, obtaining pixel-level annotations over large spatial extents is costly and labor-intensive, limiting the scalability of deep learning-based segmentation …

Birth of a gap: Critical phenomena in 2D Coulomb gas
Matthias Allard, Sampad Lahiry
https://arxiv.org/abs/2509.24529 https://arxiv.org/pdf/2509.24529

Birth of a gap: Critical phenomena in 2D Coulomb gas
We investigate a family of radially symmetric Coulomb gas systems at inverse temperature $β= 2$. The family is characterised by the property that the density of the equilibrium measure vanishes on a ring at radius $r_*$, which lies strictly inside the droplet. The large $n$ expansion of the logarithm of the partition function is obtained up to a novel $n^{1/4}$ term. We perform a double scaling limit of the correlation kernel at the $n^{1/4}$ scale and obtain a new limiting kernel in the bulk,…

Input Time Scaling
Rapheal Huang (Yuming), Weilong Guo
https://arxiv.org/abs/2508.13654 https://arxiv.org/pdf/2508.13654

Input Time Scaling
Current Large Language Models (LLMs) are usually post-trained on large-scale carefully curated datasets (data & training scaling) and doing reasoning in test time (inference time scaling). In this work, we present a new scaling paradigm, Input Time Scaling, to complement previous scaling methods by putting resources on queries (input time). During training and testing, we combine meta-knowledge from LLMs to refine inputs with different strategies. We also find a new phenomenon, training-testing…

MIRAGE: Scaling Test-Time Inference with Parallel Graph-Retrieval-Augmented Reasoning Chains
Kaiwen Wei, Rui Shan, Dongsheng Zou, Jianzhong Yang, Bi Zhao, Junnan Zhu, Jiang Zhong
https://arxiv.org/abs/2508.18260

MIRAGE: Scaling Test-Time Inference with Parallel Graph-Retrieval-Augmented Reasoning Chains
Large reasoning models (LRMs) have shown significant progress in test-time scaling through chain-of-thought prompting. Current approaches like search-o1 integrate retrieval augmented generation (RAG) into multi-step reasoning processes but rely on a single, linear reasoning chain while incorporating unstructured textual information in a flat, context-agnostic manner. As a result, these approaches can lead to error accumulation throughout the reasoning chain, which significantly limits its effec…

Eigenvalue distribution of the Neural Tangent Kernel in the quadratic scaling
Lucas Benigni, Elliot Paquette
https://arxiv.org/abs/2508.20036 https://arxiv…

Eigenvalue distribution of the Neural Tangent Kernel in the quadratic scaling
We compute the asymptotic eigenvalue distribution of the neural tangent kernel of a two-layer neural network under a specific scaling of dimension. Namely, if $X\in\mathbb{R}^{n\times d}$ is an i.i.d random matrix, $W\in\mathbb{R}^{d\times p}$ is an i.i.d $\mathcal{N}(0,1)$ matrix and $D\in\mathbb{R}^{p\times p}$ is a diagonal matrix with i.i.d bounded entries, we consider the matrix \[ \mathrm{NTK} = \frac{1}{d}XX^\top \odot \frac{1}{p} σ'\left( \frac{1}{\sqrt{d}}XW \right)D…

HyperFlexis: Joint Design of Algorithms and Systems for Multi-SLO Serving and Fast Scaling
Zahra Yousefijamarani, Xinglu Wang, Qian Wang, Morgan Lindsay Heisler, Taha Shabani, Niloofar Gholipour, Parham Yassini, Hong Chang, Kan Chen, Qiantao Zhang, Xiaolong Bai, Jiannan Wang, Ying Xiong, Yong Zhang, Zhenan Fan
https://arxiv.org/abs/2508.15…

HyperFlexis: Joint Design of Algorithms and Systems for Multi-SLO Serving and Fast Scaling
Modern large language model (LLM) serving systems face challenges from highly variable requests with diverse lengths, priorities, and stage-specific service-level objectives (SLOs). Meeting these requires real-time scheduling, rapid and cost-effective scaling, and support for both collocated and disaggregated Prefill/Decode (P/D) architectures. We present \textbf{HyperFlexis}, a unified LLM serving system that integrates algorithmic and system-level innovations to jointly optimize scheduling …

Better Language Model-Based Judging Reward Modeling through Scaling Comprehension Boundaries
Meiling Ning, Zhongbao Zhang, Junda Ye, Jiabao Guo, Qingyuan Guan
https://arxiv.org/abs/2508.18212

Better Language Model-Based Judging Reward Modeling through Scaling Comprehension Boundaries
The emergence of LM-based judging reward modeling, represented by generative reward models, has successfully made reinforcement learning from AI feedback (RLAIF) efficient and scalable. To further advance this paradigm, we propose a core insight: this form of reward modeling shares fundamental formal consistency with natural language inference (NLI), a core task in natural language understanding. This reframed perspective points to a key path for building superior reward models: scaling the mod…

UltraMemV2: Memory Networks Scaling to 120B Parameters with Superior Long-Context Learning
Zihao Huang, Yu Bao, Qiyang Min, Siyan Chen, Ran Guo, Hongzhi Huang, Defa Zhu, Yutao Zeng, Banggu Wu, Xun Zhou, Siyuan Qiao
https://arxiv.org/abs/2508.18756

UltraMemV2: Memory Networks Scaling to 120B Parameters with Superior Long-Context Learning
While Mixture of Experts (MoE) models achieve remarkable efficiency by activating only subsets of parameters, they suffer from high memory access costs during inference. Memory-layer architectures offer an appealing alternative with very few memory access, but previous attempts like UltraMem have only matched the performance of 2-expert MoE models, falling significantly short of state-of-the-art 8-expert configurations. We present UltraMemV2, a redesigned memory-layer architecture that closes t…

DATS: Distance-Aware Temperature Scaling for Calibrated Class-Incremental Learning
Giuseppe Serra, Florian Buettner
https://arxiv.org/abs/2509.21161 https://

DATS: Distance-Aware Temperature Scaling for Calibrated Class-Incremental Learning
Continual Learning (CL) is recently gaining increasing attention for its ability to enable a single model to learn incrementally from a sequence of new classes. In this scenario, it is important to keep consistent predictive performance across all the classes and prevent the so-called Catastrophic Forgetting (CF). However, in safety-critical applications, predictive performance alone is insufficient. Predictive models should also be able to reliably communicate their uncertainty in a calibrated…

Thinking Before You Speak: A Proactive Test-time Scaling Approach
Cong Li, Wenchang Chai, Hejun Wu, Yan Pan, Pengxu Wei, Liang Lin
https://arxiv.org/abs/2508.18648 https://

Thinking Before You Speak: A Proactive Test-time Scaling Approach
Large Language Models (LLMs) often exhibit deficiencies with complex reasoning tasks, such as maths, which we attribute to the discrepancy between human reasoning patterns and those presented in the LLMs' training data. When dealing with complex problems, humans tend to think carefully before expressing solutions. However, they often do not articulate their inner thoughts, including their intentions and chosen methodologies. Consequently, critical insights essential for bridging reasoning steps…

Compute-Optimal Scaling for Value-Based Deep RL
Preston Fu, Oleh Rybkin, Zhiyuan Zhou, Michal Nauman, Pieter Abbeel, Sergey Levine, Aviral Kumar
https://arxiv.org/abs/2508.14881

Compute-Optimal Scaling for Value-Based Deep RL
As models grow larger and training them becomes expensive, it becomes increasingly important to scale training recipes not just to larger models and more data, but to do so in a compute-optimal manner that extracts maximal performance per unit of compute. While such scaling has been well studied for language modeling, reinforcement learning (RL) has received less attention in this regard. In this paper, we investigate compute scaling for online, value-based deep RL. These methods present two pr…

BoN Appetit Team at LeWiDi-2025: Best-of-N Test-time Scaling Can Not Stomach Annotation Disagreements (Yet)
Tomas Ruiz, Siyao Peng, Barbara Plank, Carsten Schwemmer
https://arxiv.org/abs/2510.12516

BoN Appetit Team at LeWiDi-2025: Best-of-N Test-time Scaling Can Not Stomach Annotation Disagreements (Yet)
Test-time scaling is a family of techniques to improve LLM outputs at inference time by performing extra computation. To the best of our knowledge, test-time scaling has been limited to domains with verifiably correct answers, like mathematics and coding. We transfer test-time scaling to the LeWiDi-2025 tasks to evaluate annotation disagreements. We experiment with three test-time scaling methods: two benchmark algorithms (Model Averaging and Majority Voting), and a Best-of-N sampling method. T…

A1: Asynchronous Test-Time Scaling via Conformal Prediction
Jing Xiong, Qiujiang Chen, Fanghua Ye, Zhongwei Wan, Chuanyang Zheng, Chenyang Zhao, Hui Shen, Alexander Hanbo Li, Chaofan Tao, Haochen Tan, Haoli Bai, Lifeng Shang, Lingpeng Kong, Ngai Wong
https://arxiv.org/abs/2509.15148

A1: Asynchronous Test-Time Scaling via Conformal Prediction
Large language models (LLMs) benefit from test-time scaling, but existing methods face significant challenges, including severe synchronization overhead, memory bottlenecks, and latency, especially during speculative decoding with long reasoning chains. We introduce A1 (Asynchronous Test-Time Scaling), a statistically guaranteed adaptive inference framework that addresses these challenges. A1 refines arithmetic intensity to identify synchronization as the dominant bottleneck, proposes an online…