High Energy Physics (Phenomenology)
Dark Matter Stability from residual flavour symmetries
Giacomo Landini, Simone Marciano and Federico Mescia
We propose a model-independent framework in which Dark Matter stability emerges naturally from residual discrete $\mathbb{Z}_N$
symmetries left unbroken by the spontaneous breaking of continuous flavour
symmetries, both Abelian and non-Abelian. No ad hoc stabilizing
symmetry is required, as Dark Matter stability is an inherent consequence of the underlying flavour dynamics. As an explicit example, we construct a prototype model based on a $\text{SU}(2)_\text{F} \otimes \text{U}(1)_\text{F}$ Froggatt-Nielsen
symmetry, which successfully reproduces the observed charged lepton mass spectrum and the PMNS mixing
matrix, while identifying the lightest sterile
neutrino as a stable Dark Matter candidate. Furthermore, we study the Dark Matter phenomenology within our model in the freeze-in regime, including production channels mediated by higher-dimensional operators suppressed by the cutoff scale $\Lambda$. Our analysis shows that continuous flavour
symmetries can provide a unified, self-consistent framework addressing both the flavour puzzle and the origin, stability, and production of Dark Matter.
Read the full abstract ▾
Strongly Correlated Electrons
Berry curvature from plane-wave mixing: A three-band model with emergent symmetries
Avi Vadali and Tomohiro Soejima
Two-dimensional materials can host topological phases stabilized either by an intrinsic moiré potential or topological crystallization. Crucial to the
topology of these phases is the interplay between the (self-consistently generated) potential and the spinor structure, which imbues the band-projected potential with momentum-dependence. Truncating this potential at linear-order in momentum, we obtain a three-band model that captures the low-energy band structure and Berry
curvature of such topological bands. We analytically find that the Berry
curvature strongly depends on momentum-dependence of the potential. Furthermore, we identify that the model can be tuned to have emergent
symmetries, including $U(1)$ and anti-
unitary symmetries. Emergent-
symmetry breaking perturbations drive the system to regimes hosting topological flat bands, quadratic band touchings, and multiple Dirac cones. We recover many of these features when studying an un-truncated model of a $\lambda$-jellium spinor inside of a periodic potential. Our results reveal a rich interplay of spinor structure and a potential in determining the
topology of the ground state.
Read the full abstract ▾
Robotics
DexNinja: Learning Robust Dexterous Cutting Policy with a Real-to-Sim-to-Real Data Engine
Haozhe Lou, Runyi Yang, Wanqi Zhong et al. (+10 more)
Cutting a piece of food looks effortless in human hands. For a
robot, however, every cut is a high-contact interaction with a deformable, elasto-plastic object whose behavior changes with pose, contact mode, and cutting stage. Collecting real-world data is expensive and wasteful: each trial typically contaminates or damages the food, so obtaining scalable real world data for dexterous cutting is expensive and difficult. Simulation is an appealing alternative, but a large sim-to-real gap exists in both physics and perception. This gap is especially severe for
topology-changing interactions like cutting, where small errors in contact, friction, or material and elasticity responses quickly compound. We propose DexNinja, a real2sim2real framework that turns a handful of real demonstrations into a dense and realistic training distribution. DexNinja (i) reconstructs object instances from real captures, (ii) randomizes physically meaningful parameters under category-level constraints, and (iii) produces augmented dexterous manipulation episodes using a custom differentiable simulator that couples
robot dynamics, tactile contact, and deformable cutting. We evaluate DexNinja on a dexterous food-slicing task and show that (1) with tens of real demonstrations and hundreds of augmented simulator data, the success rate could reach more than 60%. (2) the augmented data consistently improves success rates when only using real data with more than 20% via sim-to-real, and (3) the resulting policies generalize to out-of-distribution objects with varying shapes and sizes.
Read the full abstract ▾
All authors Haozhe Lou, Runyi Yang, Wanqi Zhong, Caiyun Liu, Yicheng Liu, Wenhao Liu, Wenlong Ma, Jiawei Xia, Kanzhi Wu, Danda Pani Paudel, Luc Van Gool, Hang Zhao, Yiming Li
High Energy Physics (Phenomenology)
Implications for new physics solutions to the LSND and MiniBooNE anomalies after MicroBooNE results
Waleed Abdallah, Raj Gandhi and Samiran Roy
Over the past couple of decades, the
Liquid Scintillator
Neutrino Detector (LSND) and MiniBooNE (MB)
neutrino experiments have provided statistically significant evidence of persistent and as yet unexplained electron-like event excesses. The MicroBooNE experiment, with sensitive particle identification capabilities designed to shed light on these anomalous results, has conducted a number of analyses to test whether the excess is due to
photons, electrons or $e^+e^-$ pair productions, all of which give almost identical signals in LSND and MB. Recent MicroBooNE searches, in addition to checking Standard Model (SM) processes, have focused on proposed new physics scenarios involving new mediators and dark portals beyond the SM. We study the consequences and effects of these analyses for the many theoretical proposals put forth to generate the three types of signals above in order to understand the observed excesses. We discuss which classes of solutions are allowed, disallowed or require further scrutiny involving theoretical and experimental efforts. We give an example of a new solution that survives the restrictions imposed by MicroBooNE results and other existing bounds. We point out that a light ($m\leq 30$ MeV) dark neutral
spin zero mediator with dominantly pseudoscalar couplings, decaying to an $e^+e^-$ pair, can account for the observations. We indicate that if such a model is realised, it may point to the possibility that
neutrinos are Dirac particles, and that there are dark Heavy Neutral Leptons (HNLs) of mass ${\cal O}(100)$ MeV. We discuss issues related to CP violation and ultra-violet completion of these models. Finally we stress that a search for the excess in incoherent scattering channels in MicroBooNE and SBND could help clarify the physics behind these persistent anomalies.
Read the full abstract ▾
Artificial Intelligence
AutoMARS: Automatic and Scalable Multi-Agent Large Language Model Framework for Supply Chain Risk Assessment
Yinzhu Quan, Zefang Liu, Frederick Benaben et al. (+1 more)
We propose AutoMARS (Automatic Multi-
Agent Risk Assessment), an agentic framework for automatic and scalable comparative risk assessment in supply chain networks. The framework enables end-to-end execution of analytical workflows by automatically planning tasks, coordinating specialized
agents, invoking external analytical tools, and synthesizing domain-specific results into integrated risk evaluations. AutoMARS further supports scalable risk comparison across multiple locations by parallelizing independent analytical tasks. The framework adopts a multi-domain, multi-
agent, and multi-tool architecture that integrates heterogeneous risk factors from socioeconomic, disaster, environmental, and transportation domains. Through coordinated
agent interactions and tool integration, AutoMARS enables structured, systematic, and interpretable comparative analysis to support supply chain risk-aware location evaluation.
Read the full abstract ▾
All authors Yinzhu Quan, Zefang Liu, Frederick Benaben, Benoit Montreuil
Artificial Intelligence
Releasing the Inner Jev: A Simple Recipe for Open Decision Models
Jonathan Geuter, Andrew Cheng and David Alvarez-Melis
LLM-based decision-making systems, also called System One models, have recently gained popularity, most notably through Jev (Almeida, 2026). Given a typed input state, a question, and a set of options, they return a probability for each option in a single forward pass, without text generation. We show that existing open instruction-tuned
language models can achieve comparable performance without any additional training, by reading out answers from letters (A, B, C, etc.) in the model's output logits at the last
token position. Furthermore, we show how a simple supervised fine-tuning recipe can improve their performance further. We evaluate four open instruction-tuned
language models: Gemma 4 31B, Gemma 4 12B, Qwen3.6-27B, and Qwen3.5-4B. We fine-tune each using LoRA and a decision head initialized from a slice of the
language model head. Furthermore, we introduce OpenDecisions, a novel 2,352-question validation
dataset designed to reflect the task composition of JevBench (JevBench main- tainers, 2026) and which computes a Capability score based on accuracy and calibration metrics. Without fine-tuning, Gemma 4 31B scores 79.6 on OpenDecisions, compared to Jev's 80.1. After undergoing our fine-tuning recipe, it scores 84.4 on OpenDecisions and 81.5 on JevBench v1.7, outperforming Jev by 4.3 and 4.4 Capability points respectively and setting a new state-of-the-art on OpenDecisions. On a suite of public decision
datasets, our models achieve accuracy comparable to Jev, with lower calibration error. Our results show that open
language models already possess System One capabilities, and with simple fine-tuning they outperform Jev on a wide range of decision tasks. We release our models, serving code, and OpenDecisions.
Read the full abstract ▾
Numerical Analysis
Quantum integration in Besov spaces with mixed smoothness
Bernd Käßemodel and Tino Ullrich
We study the complexity of high-dimensional integration of periodic functions from Besov spaces $S^r_{p,q}B(\mathbb{T}^d)$ with mixed smoothness in the
quantum model of computation. We construct new
quantum algorithms based on the multivariate Faber-Schauder decomposition of such functions and analyze their convergence rates. We show a gain of one convergence order compared to the well-known deterministic rates of convergence. We additionally obtain lower bounds showing the optimality of our results up to logarithms.
Read the full abstract ▾
Machine Learning
Online LLM Personalization from Pairwise Preferences Across Users
Romain Puech, Georgy Salakhutdinov and Dimitris Bertsimas
Large
language models can answer the same question in different styles, but users' preferred styles are hidden and vary across users and questions. We study inference-time personalization through online selection from a fixed set of
prompts, each representing a persona. Users arrive sequentially, each asks a short sequence of questions, and the system displays two responses per question and observes one pairwise choice. We formulate this interaction as a variant of the contextual dueling bandit problem in which the learner starts without preference data, adapts within each session, and exploits information shared across users. The objective is to minimize cumulative strong regret, summing the utility gaps of both displayed responses relative to the user's best response among the available options. We make two contributions. First, we introduce the problem and release a reproducible preference
benchmark and
dataset built around a tutoring case study, containing educational questions, LLM-generated answers in six teaching styles, and pairwise judgments from LLM-simulated students with hidden preference profiles. We fit Bradley-Terry models to these judgments to obtain reference utilities for regret evaluation. Second, we propose a baseline that learns a mixture prior over users from completed sessions and adapts to each new user within the session. On students and topics never used in development, it removes 44-48% of random pairs' excess regret and achieves over half the performance of a per-user linear oracle fitted in hindsight. We release code,
prompts, and the
benchmark.
Read the full abstract ▾
Machine Learning
MLAfication: Efficiently Enabling High-Fidelity MLA for Modern LLMs
Zile Wang, Zhide Xie and Tao Ji
Multi-Head Latent
Attention (MLA) compresses the KV cache into a latent representation, reducing inference
memory and bandwidth. Existing MHA2MLA/TransMLA methods do not support QKNorm and require costly recovery on modern LLMs, especially MoE. We present MLAfication, a QKNorm-compatible, parameter-efficient extension of MHA2MLA. It combines calibrated KNorm, parameter-efficient
distillation, contribution-aware Partial-RoPE, and activation-aware joint KV initialization. Across MHA and GQA models, including QKNorm- and MoE-equipped variants, MLAfication preserves performance under substantial KV-cache compression, surpasses the current SOTA on commonsense and long-context
benchmarks, and extends MLA conversion to modern architectures. On Qwen3-30B-A3B, MLAfication delivers up to $5.35\times$ the serving throughput of BF16 GQA and, combined with 4-bit KV quantization, reduces the KV-cache footprint by $94.73\%$. Our code is available at https://github.com/JT-Ushio/MLAfication.
Read the full abstract ▾
Computer Vision
Multics: Generative Multi-Player World Model with Hierarchical Communication
Kaiyi Huang, Wangbo Yu, Yukun Huang et al. (+6 more)
Multi-player interactive world models aim to simulate shared environments from multiple players' perspectives, but achieving robust generation while scaling to more players remains challenging. Existing approaches that rely on fully-connected communication (e.g., Solaris) incur quadratic
attention costs as the number of players grows, while centralized approaches (e.g., $\gamma$-World) suffer from interference between unrelated events in a shared global representation. To address these challenges, we present Multics, a multi-player
video generation framework with hierarchical communication that structures information exchange across player, group, and global levels. This hierarchy introduces group
tokens into self-
attention as intermediate group-level representations that aggregate locally relevant player information. These
tokens connect to global
tokens for cross-group communication, while the hierarchical
topology limits direct mixing between unrelated local events. Group-
token parameters are shared across group slots, allowing the same communication module to accommodate varying numbers of players without introducing new group-specific parameters. We further propose budget-constrained dynamic grouping based on recent camera trajectories, which adaptively groups relevant players while limiting unnecessary communication. Experiments demonstrate that Multics achieves state-of-the-art performance in 4- and 6-player settings, with improved cross-player consistency, player-specific action and camera controllability, and interference robustness. Moreover, Multics generalizes from 4-player training to 6-, 8-, and 10-player inference without retraining, while reducing self-
attention computation compared with fully-connected communication.
Read the full abstract ▾
All authors Kaiyi Huang, Wangbo Yu, Yukun Huang, Jiwen Yu, Kunhao Liu, Donglin Yang, Wenbo Hu, Ying Shan, Xihui Liu
Cosmology and Nongalactic Astrophysics
Interpreting the stacked kinetic SZ signal II: the degeneracy between satellite galaxies and baryonic feedback
Lurdes Ondaro-Mallea, Boryana Hadzhiyska and Raul E. Angulo
The kinetic Sunyaev-Zel'dovich (kSZ) signal stacked around a
galaxy sample traces the gas around
halos and is a powerful probe of baryonic feedback. Its interpretation, however, depends on the satellite
galaxies in the sample. Using the FLAMINGO hydrodynamical simulations and DESI-like samples of luminous red
galaxies (LRGs), bright
galaxies survey (BGS), and emission-line
galaxies (ELGs), we show that satellites and feedback have similar effects: both suppress the small-scale signal. This degeneracy arises because stacking on satellites, which are offset from the centre of their host
haloes, mimics the extended gas distribution produced by feedback. Although satellites make up only 20-30% of each sample, they account for 50-80% of the signal because they occupy more massive
halos than centrals; thus their imprint on the stacked profiles can be as large as that of feedback. At the precision of current DESI$\times$ACT data, neglecting satellites in the modelling biases the inferred gas fractions low, by up to a factor of two, and overestimates the feedback strength, particularly for the LRG and BGS samples. Remaining agnostic about the satellite population avoids this bias but reduces the constraining power, so robust constraints would require knowing the satellite fraction to better than 10%, for instance from
lensing and clustering data of the same
galaxy samples. Finally, a further contribution from satellites will become statistically significant with upcoming surveys: satellites trace the anisotropic gas distribution around their host
haloes, an effect that contributes 10% of the signal, and will likely require simulation-based models to capture.
Read the full abstract ▾
Signal Processing
On the CSIT-Limited Delay Robustness of IR-HARQ in Finite Blocklength RSMA Networks
Shaima Abidrabbu
This letter studies the throughput and reliability of power-constrained rate-splitting multiple access (RSMA) with variable-length incremental-redundancy hybrid automatic repeat request (IR-HARQ) under finite blocklength (FBL) transmission and imperfect channel state information at the transmitter (CSIT). While perfect-CSIT feedback-delay tolerance approaches a natural ceiling as signal-to-noise ratio (SNR) grows, we show imperfect CSIT imposes a strictly lower, finite ceiling set solely by CSIT quality, one that transmit power cannot overcome. We derive analytical expressions and tight approximations for common- and private-stream decoding failure probabilities, capturing FBL and CSIT uncertainty via a unified CSIT-degradation factor. Also, we show that residual interference saturates effective stream power at high SNR, limiting reliability and producing this delay-tolerance ceiling. Simulations validate the analysis and quantify the ceiling, offering design insights for delay-robust RSMA under imperfect CSIT.
Read the full abstract ▾
Mesoscale and Nanoscale Physics
Revealing topological protection through unprotected operations in a Majorana parity qubit
Sebastian Miles, Juan Daniel Torres Luna, Francesco Zatelli et al. (+7 more)
A recent experiment reported coherent parity oscillations between minimal Kitaev chains, demonstrating a Majorana parity
qubit (arXiv:2607.09511). Operations on this
qubit are controlled by coupling two Kitaev chains and are intrinsically unprotected against noise in the coupling strength. Here we show how the exponential protection in the underlying chains can nevertheless be revealed through unprotected
qubit operations. To this end, we develop an analytical theory of quasistatic dephasing for two coupled $N$-site Kitaev chains at the sweet spot. In Rabi experiments, protection can be resolved by adapting the operating point with chain length: depending on the coupling architecture, it appears as an exponentially increasing optimized number of coherent oscillations or as a longer coherence time. Ramsey experiments show an exponentially increasing coherence time, provided the chains can be sufficiently decoupled. Our analytical theory takes into account the non-Gaussian splitting statistics of the Kitaev chains, allowing us to quantitatively predict the scaling with chain length. Moreover, we determine how residual noise and imperfections limit the range of chain lengths over which this scaling can be observed. Our results show how
qubit dynamics can be used to probe Majorana protection in experimentally realistic devices.
Read the full abstract ▾
All authors Sebastian Miles, Juan Daniel Torres Luna, Francesco Zatelli, Nick van Loo, Antonio Lombardi, Vincent P. M. Sietses, Florian J. Bennebroek Evertsz', Bart Roovers, Leo P. Kouwenhoven, Michael Wimmer
Software Engineering
Deleting Code Without Losing Knowledge: Regenerative Software for Python Libraries
Thomas Bartz-Beielstein
In the age of "Software 3.0", when a complete program can be rewritten in minutes by an
agent, the traditional approach of relying on code as the sole source of truth becomes questionable. Fowler advocated for a set of seven primitives, which lay the framework for regenerative software, so that programs can be deleted and rebuilt without losing their knowledge. The knowledge lives in machine-readable records in a tracked directory, from which a *regeneration bundle* is built for each unit. The regeneration bundle is a single file with the module's architecture, intent, decisions, public contract, evaluations, and consumers, but no implementation, plus the one dependency module it may import. The open-source package spotregen implements these primitives and provides a graphical user interface. This report presents a first application of Fowler's primitives to the safety-critical forecasting library spotforecast2-safe. During the first experiments, the Claude Code session that orchestrated the study consumed more than 43 million
tokens, almost all of them its own history re-sent with every request. This motivated the switch to an architecture that is built on local models. Deterministic checks and the maintainer judged the results. The report contains executable examples for every primitive. Each example runs spotregen against the repository or against a throw-away copy of it.
Read the full abstract ▾
K-Theory and Homology
Classifying quasi-representations of groups via their characters
Shmuel Weinberger, Jianchao Wu and Guoliang Yu
For every countable discrete group, we construct a character for finite-dimensional
unitary quasi-representations, with values in the inverse limit of the equivariant $K$-theory of its Rips complexes. The character is a complete invariant for quasi-representations up to stable equivalence for amenable groups and, more generally, for groups satisfying the Haagerup property and QR-quasidiagonality. For Haagerup groups satisfying quantitative QR-quasidiagonality, we also obtain a quantiative version of the
theorems. These conclusions apply when the maximal group $C^*$-
algebra is quasidiagonal and has the lifting property, including amenable groups, free groups, and closed orientable surface groups. For surface groups, rank and winding number give complete invariants for the classification. Finally, composing with the equivariant Chern character gives a delocalized cohomological obstruction to stable equivalence of a quasi-representation with genuine representations.
Read the full abstract ▾
Computation and Language
Welcome to the Age of Agent Ads: Using On-Policy Attacks to Steer Agentic Decisions
Tim R. Davidson, Marija Šakota, Joshua Joseph et al. (+1 more)
People increasingly delegate option-ranking tasks (e.g., shopping, job search, travel) to AI
agents, which creates incentives to steer agentic decisions via "
agent ads." To defend against such steering attacks, people count on models'
alignment training and platforms' terms of service, and consequently the most effective existing attacks are designed to trigger off-policy behaviors. Exposing a new attack surface, we introduce on-policy attacks, which steer agentic decisions without pushing individually aligned behaviors off policy. Our example attack consists of two steps: To remain compliant with platforms' terms of service, we first inject a truthful negation about a fictitious concept ("Does not contain blizyester") into a product description. This triggers an
agent to search the web ("What is blizyester?"), which surfaces planted websites full of fabricated information associating the fictitious concept with harmful implications ("Blizyester linked to dementia"). Trained to keep users safe from harm, the
agent penalizes every option missing the truthful negation. Evaluating our attack across four real-world listing
datasets, we find some frontier models exposed, enabling an
adversarial actor to penalize competitors without violating existing platform-listing guidelines. We discuss the broader challenge of regulating advertising aimed at AI
agents and provide recommendations to model developers and platform providers.
Read the full abstract ▾
All authors Tim R. Davidson, Marija Šakota, Joshua Joseph, Robert West
Functional Analysis
Absolutely minimising Lipschitz extensions between Hilbert spaces
Krzysztof J. Ciosmak
Every Lipschitz map from a closed subset of a real
Hilbert space into a real
Hilbert space has an absolutely minimising Lipschitz extension: an extension whose Lipschitz constant on every open subset of the complement, bounded or not, equals the Lipschitz constant of its restriction to the boundary of that subset. Neither space need be separable. This answers a question of Naor and Sheffield. For targets of dimension at least two we prove a density
theorem on connected domains: every optimal extension of a boundary datum can be approximated, uniformly on compact sets, by extensions whose Lipschitz constant on every nonempty open subset, and on its boundary, equals the constant of the datum. Consequently, in contrast with the scalar case, such a datum has a unique absolutely minimising extension exactly when its optimal extension is unique, and absolutely minimising extensions need not be closed under uniform convergence, even for fixed smooth data on a planar disk. The extensions are built by finite piecewise isometric interpolation: on nested polyhedra with dense union they are piecewise affine and preserve distances in at least two directions on each simplex, a structure described through leaves of 1-Lipschitz maps. The boundary constraints are handled by a midpoint argument for admissible values, and a Kuratowski-Zorn argument removes the separability assumption while preserving previously prescribed values. We also characterise the bounded linear maps that are absolutely minimising extensions of all data determining their operator norm.
Read the full abstract ▾
Computational Physics
Principles and Techniques for Agentic Research in Experimental Physics
Niko R. Reed, Declan M. Daly, Andrew J. Beling et al. (+4 more)
The use of artificial intelligence (AI) in physics research is increasingly common, yet few guidelines exist for its effective use. Starting with a principles-based approach, we outline a set of ten requirements for collaborative AI
agents in experimental physics, spanning scientific norms, effective
agent use, and domain expertise. We take a human-centered approach in which
agents act as collaborators that amplify rather than replace the expertise of a human researcher (user) who directs the work and is responsible for its results. We then present a framework designed to enforce these requirements, detailing methods for improving off-the-shelf behavior of large
language models (LLMs) with custom scaffolding and describing general-purpose scientific workflows and skills. In particular, we highlight the power of closed-loop campaigns where
agents work iteratively toward a well-defined goal. We demonstrate the potential of this mode of research through six case studies in experimental physics, covering device design, hardware control, neural network optimization, physical model identification, and data analysis. All six case studies result in useful work products that add utility to our research, including a neural network architecture and solid-state
quantum defect photodynamic model that substantially outperform our published work; a microwave waveguide design that extends a published design concept to a fabrication-ready device meeting our experimental constraints; a
quantum control pulse sequence candidate that improves simulated performance by more than an order of magnitude; commissioning of an NMR spectrometer to $\sim$1~Hz frequency precision on live hardware; and finally a data-fitting pipeline that achieves, to our knowledge, the first reconstruction of the Ising microstate of an artificial
spin-ice array from diffraction-limited
magnetic imaging.
Read the full abstract ▾
All authors Niko R. Reed, Declan M. Daly, Andrew J. Beling, Jordan D. Giandrea, Ronald L. Walsworth, John W. Blanchard, Daniel Ang
Artificial Intelligence
Size Generalization for Diffusion Routing Solvers
Barak Itzchak Davidovitch, Eli Meirom, Ethan Fetaya et al. (+1 more)
Routing problems, like the Traveling Salesman Problem (TSP) and the Capacitated Vehicle Routing Problem (CVRP), are NP-hard and slow to solve for large instances.
Diffusion-based solvers can be trained in a supervised way using sets of optimal solutions, and then quickly solve new test instances. However, for large instances, optimal solutions are hard to find, and
diffusion models fail to generalize to instance sizes outside their training distribution. Here we describe ResizeCO, a label free size-generalization approach that addresses both the change in statistical properties of larger instances, and the computational challenge of solving them quickly. It uses reinforcement learning on large-scale instances to adapt a
diffusion model trained on small instances. Specifically, we adapt GRPO to this problem, using route length as the training signal. ResizeCO also formulates k-opt as a tree search, solved by a beam search on the GPU. In size-generalization experiments on TSP and CVRP instances with up to 10000 nodes, ResizeCO achieves a Pareto front over route quality and time that dominates learned solvers, and also outperforms other solvers at short run times, setting a new state of the art.
Read the full abstract ▾
All authors Barak Itzchak Davidovitch, Eli Meirom, Ethan Fetaya, Gal Chechik
Machine Learning
FreeSolo: Linear-time Sparse Attention with Zero-Shot Length Extrapolation
Zekun Shi and Alastair James Arthur Price
Linear-time sequence models typically store long-context information in a recurrent state. A fixed-size state limits how much information can be stored, while hybrid architectures that add full
attention for better recall incur quadratic sequence cost. Alternatively, sparse
attention with continued pretraining (CPT) can extend the effective context length. Generalization beyond the trained length remains challenging, however, and CPT can be costly. We introduce FreeSolo, a native sparse
attention mechanism with a short-range sliding window branch and a long-range branch that selects keypoint
cells for direct
token attention. Its linear-time variant, FS-L, learns to prune the prefix before query-dependent selection through short CPT. FS-B searches balanced trees with $O(N\log^2 N)$ algorithmic cost including radix-sort construction. FreeSolo demonstrates better zero-shot length extrapolation than the baselines without long-context fine-tuning.
Read the full abstract ▾
Machine Learning
CartoVec: General-Purpose Location Embeddings from Open Map Data
Furkan Gursoy
Geographic coordinates are common inputs in applied machine learning, but latitude and longitude describe position rather than the surrounding environment. Image-grounded location encoders learn this context from satellite or ground-level imagery; structured open map data offers another source of supervision whose utility for continuous coordinate encoders remains less explored. We introduce CartoVec, a coordinate encoder trained to predict Overture-derived land cover, land use, transportation, place and connectivity targets at multiple spatial scales, without image inputs or downstream-task labels. We evaluate frozen
embeddings on Census outcomes, Airbnb prices and Philadelphia crime in the US Northeast using spatial block cross-validation. CartoVec has the highest mean performance on most Census outcomes and the highest average across the full suite, with several comparisons close across folds. The findings support open map data as a practical source of reusable location representations with a coordinate-to-vector interface that requires no map lookup at inference. The open-weight model is available at https://huggingface.co/gursoyfurkan/CartoVec.
Read the full abstract ▾
Computation and Language
Soft Teacher Forcing
Mohammad Fahes, Cordelia Schmid and Anurag Arnab
Soft
tokens have so far been used to replace discrete chain-of-thought in latent
reasoning methods. These approaches typically require multi-stage training and, in some cases, auxiliary modules, yet they still underperform supervised fine-tuning (SFT) on <input, chain-of-thought, output> triples. In this paper, we ask whether soft
tokens can be used to improve SFT itself, instead of as an avenue for latent
reasoning. We introduce Soft Teacher Forcing (Soft-TF), a simple training
algorithm that combines teacher-forced
embeddings with soft
tokens through random convex combinations. The soft
tokens are computed in a single parallel forward pass over the ground-truth sequence. Soft-TF adds no parameters, no external model and no curriculum, and leaves inference unchanged. Experiments show that Soft-TF improves over SFT on GSM8K, MATH, MathQA and AQuA, by +2.1 to +4.1 points on GSM8K across seven Gemma, Llama and Mistral models.
Read the full abstract ▾
Combinatorics
A Generalisation of Maximal $(k,b)$-Linear-Free Sets of Integers
Minh Quang Nguyên
Fix integers $k,b,q$ with $k\ge 2,b\ge 0,q\ge 2$. Define the function $p$ to be: $p(x)=kx+b$. We call a set $S$ of integers $(k,b,q)$-linear-free if $x\in S$ implies $p^i (x)\notin S$ for all $i=1,2,\dots,q-1$, where $p^i (x)=p\left(p^{i-1}(x)\right)$ and $p^0 (x)=x$. Such a set $S$ is maximal in $[n]:=\{1,2,\dots,n\}$ if $S\cup \{t\},\forall t\in [n]\backslash S$ is not $(k,b,q)$-linear-free. Let $M_{k,b,q} (n)$ be the set of all maximal $(k,b,q)$-linear-free subsets of $[n]$, and define $g_{k,b,q} (n)=\min_{S\in M_{k,b,q} (n)}|S|$ and $f_{k,b,q} (n)=\max_{S\in M_{k,b,q} (n)}|S|$. In this paper, formulae for $g_{k,b,q} (n)$ and $f_{k,b,q} (n)$ are proposed. Also, it is proven that there is at least one maximal $(k,b,q)$-linear-free subset of $[n]$ with exactly $x$ elements for any integer $x$ between $g_{k,b,q} (n)$ and $f_{k,b,q} (n)$, inclusively.
Read the full abstract ▾
Audio and Speech Processing
MSVAE: Rethinking Musical Semantics in Continuous Latent Spaces
Guobin Ma, Yuepeng Jiang, Dake Guo et al. (+15 more)
Continuous latent representations underlie recent
diffusion-based lyrics-to-song and text-to-music systems, yet aligning lyric conditions with these latents remains difficult: existing systems resort to generator-side machinery such as sentence-level timestamps, auxiliary
alignment objectives, or semi-autoregressive decoding, while the latents they predict are still produced by autoencoders trained purely for reconstruction. We address this difficulty from the representation side. We propose MSVAE (Music Semantic Variational Autoencoder), which structures its continuous latent around two complementary semantic streams: work-level compositional content and lyric-level linguistic-vocal content. Partitioned dual-stream
alignment makes both streams explicit by aligning disjoint latent subspaces with frozen lead-sheet and ASR references, while a semantic-first gated residual posterior derives its base from a frozen music-SSL encoder and admits acoustic information only through gated, bounded corrections. MSVAE achieves competitive stereo reconstruction and stronger MARBLE probing than continuous-autoencoder baselines, and improves downstream lyrics-to-song and text-to-music generation. Notably, with only an untimed phoneme prefix and no lyric timing, MSVAE yields the lowest phoneme error rate, whereas all baseline representations require sentence-level
alignment to produce intelligible vocals.
Read the full abstract ▾
All authors Guobin Ma, Yuepeng Jiang, Dake Guo, Huakang Chen, Yuan Zhou, Lingxin Xiong, Wenkai Cheng, Jingbin Hu, Hanke Xie, Chunbo Hao, Wenhao Li, Ziqian Ning, Jixun Yao, Jian Zhao, Zhonglin Jiang, Yong Chen, Lei Xie, Pengcheng Zhu
Cosmology and Nongalactic Astrophysics
Beyond the Kullback–Leibler divergence: the likelihood-ratio distribution as a detectability measure
Amirhossein Samandar, Mikel Martin Barandiaran, Linn Htat Lu et al. (+13 more)
The Kullback–Leibler (KL) divergence is the expectation value of the log-likelihood ratio between two specified models, evaluated with respect to one of the two models. It provides an average measure of the distinguishability of the models. We examine its connection to discovery potential and the accuracy of parameter estimates when one model represents a null hypothesis, and the alternative is a family of models characterized by one or more continuous parameters. For data drawn from a multivariate Gaussian distribution, we derive the log-likelihood-ratio distribution and its cumulants under two such hypotheses. We then search over the alternative model's parameter space and use peak prominence, the height of a likelihood-profile peak above its surrounding baseline, as an operational detection statistic. Repeating this search on simulated data drawn from the null model allows us to calibrate the prominence threshold for a specified false-positive rate. We study a toy Gaussian model as well as three cosmological examples involving
cosmic topology, the tensor-to-scalar ratio, and dark energy. In all these cases, detectability of a particular model depends not only on KL divergence, but also on the size of the model parameter space, the likelihood-ratio distributions, and the distance of the model from any parameter space boundaries. The relationship between KL divergence, discovery potential, and parameter estimation must therefore be calibrated for each family of models and detection statistic under consideration.
Read the full abstract ▾
All authors Amirhossein Samandar, Mikel Martin Barandiaran, Linn Htat Lu, Andrew H. Jaffe, Deyan P. Mihaylov, Glenn D. Starkman, Cynthia Trendafilova, Yashar Akrami, Stefano Anselmi, Javier Carrón Duque, Craig J. Copi, Fernando Cornet-Gomez, Arthur Kosowsky, Anna Negro, Ricardo G. Rodrigues, Andrius Tamosiunas
Quantum Physics
Error-Corrected QRAM Queries with One-Fault-Tolerant CSWAP Gates
Shifan Xu, Connor T. Hann, Nathan Wiebe et al. (+2 more)
Quantum random-access
memory (QRAM) enables coherent queries to classical data. However, the \(O(N)\)
qubits required for a
memory with \(N=2^n\) addresses make QRAM appear vulnerable to noise. Though bucket-brigade routing provides intrinsic noise resilience to reduce the error in QRAM, improving the fidelity of QRAM queries is still an important and challenging problem. We introduce a low-overhead protection framework for bucket-brigade QRAM with two complementary components. First, we propose a one-fault-tolerant repeat-until-success logical controlled-SWAP gate using the \([[4,2,2]]\) code under erasure noise channel. Interleaved erasure checks and local recovery correct every single erasure, so the leading logical failure probability is second order in the physical erasure error rates. Second, we exploit validity constraints of bucket-brigade routing to define local parity checks directly on the router states, without adding extra data
qubits to create code redundancy. Under bit-flip noise, minimum-weight perfect matching with noiseless syndrome measurements corrects single interior faults, while open spatial boundaries retain a first-order contribution in the residual bit-flip rate. Monte Carlo simulations confirm quadratic suppression for the protected controlled-SWAP and improved depth scaling for decoded QRAM router-state snapshots. The two codes admit a concatenated realization in which the inner code supplies erasure protection and the outer code optionally corrects interior logical bit flips. Together, these results show how routing constraints and a protected non-Clifford routing gate can improve QRAM reliability under biased and erasure-dominant noise.
Read the full abstract ▾
All authors Shifan Xu, Connor T. Hann, Nathan Wiebe, Steven M. Girvin, Yongshan Ding
Computation and Language
DuoDiscover: Scaling Autonomous Discovery Beyond the Evaluation Bottleneck
Shubham Agarwal, Akshat Gupta, Mike Heddes et al. (+21 more)
Evaluation has been central to advances in AI, and it is equally critical to autonomous discovery. Yet scaling autonomous discovery to real-world engineering challenges faces a fundamental bottleneck: the cost of evaluating progress. In real settings, a single true evaluation, such as a full
benchmark or serving trace, can take hours, so a search can afford only a handful of true evaluations. We present DuoDiscover, a search system that jointly discovers what to optimize and how to evaluate progress efficiently, through three mechanisms. First, each worker in a fleet of coding
agents co-discovers an optimization target and a fast, tailored proxy evaluator, which we call a yardstick. The
agent then hill-climbs the yardstick rather than the true evaluation. Second, since search can exploit gaps between a yardstick and the true objective, fidelity tests continually recheck each yardstick's ranking and invalidate any that drift. Third, the fleet is bottom-up, with no central decomposition of the problem. A lightweight orchestrator spawns and terminates
agents, briefs running ones, and seeds new
agents with others' confirmed work. DuoDiscover enables autonomous optimization from software to chip architecture. We demonstrate DuoDiscover on two repository-level
benchmarks and four GPU and TPU kernel suites covering hundreds of kernels. On SWE-fficiency, it doubles the average speedup of human experts. On SOL-Exec, it ranks first on 121 of 235 kernels and in the top three on 212. On the heavily optimized vLLM and SQLite engines, DuoDiscover achieves 1.92$\times$ and 1.38$\times$ gains over the base code. On a CPU microarchitecture optimization task with evaluations of up to 28 hours, DuoDiscover surpasses both AlphaEvolve-based ArchAgent v2 and the best human design, with 1.65$\times$ ArchAgent v2's gain in 16.7 days versus more than 40 days, advancing the frontier of autonomous chip design.
Read the full abstract ▾
All authors Shubham Agarwal, Akshat Gupta, Mike Heddes, Dilara Soylu, Rui Meng, Patrick Tser Jern Kon, Jiefeng Chen, Mert Cemri, Shu Liu, Abe Gonzalez, Akanksha Jain, Bhavana Dalvi Mishra, Linfeng Song, Vilobh Meshram, Xiangru Tang, Jinsung Yoon, Aditya Parameswaran, Ion Stoica, Vahab Mirrokni, Burak Gokturk, Parthasarathy Ranganathan, Tomas Pfister, Nanyun Peng, Chun-Liang Li
Artificial Intelligence
Aotian Chen, Frazier N. Baker, Reza Averly et al. (+5 more)
Current chemistry
benchmarks primarily evaluate outcome prediction or reaction pathway generation, but rarely test whether a model can explain why an observed chemical effect occurs. We introduce ChemExplain, a
benchmark of 1,180 multiple-choice chemistry questions drawn from 383 peer-reviewed papers. Each question pairs a molecular or reaction image with an observed chemical effect and four candidate explanations, each associated with a different structural feature visible in the image. Answering requires the model to perceive the depicted structure, apply chemical knowledge, and identify which feature most directly explains the effect, a capability we term explanatory attribution. The correct feature-mechanism relation is derived from supporting evidence in the source paper rather than invented by the question-generation LLM. We construct the
benchmark automatically with a multi-
agent framework that applies progressive quality control at three stages and validate it with eight chemistry domain experts. The strongest evaluated model reaches 78.6% accuracy, leaving considerable room for improvement. On a 600-question expert-reviewed subset,
reasoning-trace analysis classifies 71.4% of the strongest model's errors as attribution failures. Attribution is also the largest error category for the open-source and chemistry-specific models included in our analysis.
Read the full abstract ▾
All authors Aotian Chen, Frazier N. Baker, Reza Averly, Shan Zhang, Tianao Chen, Daniel Adu-Ampratwum, Xia Ning, Yuan Xue
Analysis of PDEs
Local well-posedness for the exponentially modified Benjamin-Ono equation
Thyago S. R. Santos
We prove local well-posedness for the modified Benjamin-Ono equation in $H^s(\mathbb{R})$ for $s>3/2$. This equation arises as a modified counterpart of the Benjamin-Ono equation through a Miura-type transformation, in close analogy with the classical relation between mKdV and KdV equations. One of the main obstacles is the apparent derivative loss in the quadratic part of the equation. To overcome this difficulty, we introduce a Cole-Hopf transform adapted to the structure of the equation. The map removes the quadratic derivative term and converts the equation into a Schrödinger equation with a transport and a commutator terms that can be controlled without derivative loss. Solutions are then constructed by parabolic regularization, while frequency envelopes are used to obtain the continuity of the data-to-solution map. In the degenerate case $\gamma=0$, the transform completely linearizes the equation and yields local well-posedness for every $s>1/2$, down to the scaling-critical threshold.
Read the full abstract ▾
Mesoscale and Nanoscale Physics
Spatial Engineering of Topological Quantum Matter with Structured Light
Matias Mustonen, Ali G. Moghaddam, Teemu Ojanen et al. (+1 more)
Structured light provides a powerful means of controlling matter while itself exhibiting nontrivial spatial
topology. Here, we show that these spatial structures can be harnessed to locally engineer the topological properties of matter. In a paradigmatic example of a topological
insulator with Dirac surface states driven by an optical skyrmion beam, we show that structured light spatially tailors the effective
Hamiltonian, yielding position-dependent topological gaps and Dirac velocities. The resulting gap modulation can create topological and trivial regions separated by optically defined interfaces, without introducing material boundaries or chemical modifications. We further show that the geometry of these interfaces can be controlled through the spatial structure of the light, in particular its orbital angular momentum. Our results establish structured light as a route to dynamically pattern topological phases and open a broader perspective on the interplay between
topology in light and matter, including the possibility of transferring topological characteristics between the two.
Read the full abstract ▾
All authors Matias Mustonen, Ali G. Moghaddam, Teemu Ojanen, Marco Ornigotti