Information Theory
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Monday 31 August
Watermarking has been proposed as a way to identify synthetic samples in estimation settings where no metadata is available to distinguish them from real samples, but its precise effects remain unexplored. In the absence of a distinguishing mechanism, it has been shown that adding synthetic samples significantly reduces the marginal efficacy of new real samples. In this work, we study the minimax loss of such recursive discrete distribution estimation in the presence of watermarks in contrast to the unassisted and oracle-assisted losses. When the fraction of real samples vanishes asymptotically, we provide a lower bound that shows that it is impossible to improve performance by adding watermarks unless the false negative rate of detection also vanishes. Additionally, we show that in most regimes, the worst-case losses of a sequence of simple deterministic estimators match the corresponding lower bounds up to constants. Finally, we propose masking, a randomization procedure that narrows the gap in the remaining regimes to a Jensen gap. We conjecture that a tighter lower bound argument can close this gap.
Mathematical Physics
Rotating-wave approximation for spin-boson models with structured fields
We derive state-dependent bounds on the difference between two quantum evolutions generated by unbounded Hamiltonians sharing a common form domain. The main technical tool is a second integration by parts, performed at the level of sesquilinear forms rather than at the operator level, which removes the need for a common invariant operator domain. The resulting estimate involves the norm of the time-integrated difference of the two generators, rather than the integral of its norm, and is therefore sensitive to the averaging effects produced by fast-oscillating terms. As an application we prove a quantitative bound on the rotating-wave approximation for spin-boson models with a structured boson field, described by an arbitrary massive dispersion relation on a general measure space and by a suitable class of form factors. The proof involves a careful analysis of the high-frequency scaling. The bound holds on a dense subspace of states, is fully explicit, and all the constants entering it depend only on the parameters of the model and not on the frequency scale, so that the approximation becomes exact in the limit of large frequency.
Read the full abstract
Quantum Physics
Phase-noise induced many-body interference suppression in Gaussian Boson Sampling
We develop a Heisenberg-picture tensor-network formulation of collision-free Gaussian Boson Sampling, providing a direct Fock-space expression for output probabilities in terms of experimentally accessible quantities. The resulting representation naturally recovers the Hafnian structure while revealing the decomposition of GBS probability into a phase-insensitive contribution and a hierarchy of interference sectors associated with pairs of perfect matchings. As an application, we investigate phase diffusion and show how it progressively suppresses many-body interference, driving the output statistics toward a classical dimer-model regime. Our results establish a transparent framework for connecting experimentally characterized phase fluctuations with the loss of quantum interference in photonic quantum sampling experiments.
Read the full abstract
Combinatorics
Extensions of Hindman's theorem via finite colorings of topological groups
We study the partition regular properties of topological groups, proving several extensions of Hindman theorem where monochromatic sets of finite sums are required to satisfy additional topological constraints. In particular, our results imply that for every nowhere dense set $C \subseteq \mathbb{R}^n$, there exists an open set $P \supseteq C$ such that for any finite coloring of $\mathbb{Q}^n \setminus P$, there is a family $\mathcal{A}$ of sequences in $\mathbb{Q}^n \setminus P$ which satisfies the following properties: (i) for each $A\in\mathcal A$, the set $\operatorname{FS}(A)$ of finite sums of $A$ is a closed discrete subset of $\mathbb R^n$; (ii) the set $\bigcup_{A\in\mathcal A}\operatorname{FS}(A)$ is monochromatic; and (iii) the set $\bigcup_{A\in\mathcal A}\operatorname{FS}(A)$ is dense in an open unbounded subset of $\mathbb R^n$.
Read the full abstract
Rings and Algebras
Properties of the $\mathcal V$-Monoid of Weighted Leavitt Path Algebras
For a row-finite weighted graph $(E,w)$, Preusser showed that the monoid $\mathcal{V}(L_k(E,w))$ of finitely generated projective modules over the weighted Leavitt path algebra $L_k(E,w)$ is isomorphic to a combinatorially defined weighted graph monoid $\mathcal{M}(E,w)$. We study two structural properties of $\mathcal{M}(E,w)$: confluence and cancellativity. We introduce a reduction system on the free commutative monoid presenting $\mathcal{M}(E,w)$, obtain sufficient conditions for non-confluence by constructing explicit non-confluent triples, and provide a complete confluence characterization for certain classes of weighted graphs. Turning to cancellativity, we work within Preusser's class of weighted graphs satisfying Condition (LPA), for which $L_k(E,w)$ is isomorphic to an unweighted Leavitt path algebra $L_k(F)$ via a two-step construction. We introduce an auxiliary graph associated to the intermediate step of this construction and use it to give a graph-theoretic characterization of when $\mathcal{M}(E,w)$ is cancellative. Finally, under Condition (LPA), we show that Preusser's construction upgrades to a graded isomorphism $L_k(E,w) \cong_{\operatorname{gr}} L_k(F)$ with respect to the standard $\mathbb{Z}^{\lambda(E,w)}$-grading of weighted Leavitt path algebras, yielding $\mathcal{V}^{\operatorname{gr}}(L_k(E,w)) \cong \mathcal{V}^{\operatorname{gr}}(L_k(F))$ as $\mathbb{Z}^{\lambda(E,w)}$-monoids.
Read the full abstract
Algebraic Geometry
A remark on polycyclic groups of birational automorphisms
Let $X$ be an algebraic variety of dimension $d$ over an algebraically closed field $k$ of zero characteristic. Suppose that $G$ is a virtually polycyclic subgroup in the group $\mathrm{Bir}(X)$ of birational automorphisms of $X$. We show that the virtual derived length of $G$ does not exceed $2d+1$. Moreover, if $X$ is a surface, the bound can be improved to $3$, and this value is optimal.
Read the full abstract
Econometrics
Moments of Random Coefficients in Short Panels
We study identification and estimation of moments of random coefficients in short linear panels, allowing the number of heterogeneous coefficients to exceed the number of equations observed for each unit. Under moment homogeneity, different regressor histories impose restrictions on the same moment vector. We give necessary and sufficient conditions for these restrictions to identify moments of a given order, stated in terms of the row spaces generated by the regressor support. The results show that moments may be identified even when the coefficients cannot be recovered for any individual, and, for two full-row-rank histories, give the exact loss of independent restrictions caused by overlap of their row spaces. When the support condition fails, we establish nonidentification in the maintained model and characterize the sharp identified set implied by these conditional moments. At second order, the identified set is determined by positive-semidefinite covariance restrictions and is also sharp relative to the full joint distribution of outcomes and regressors. Under the support condition, weighted minimum-distance estimators are root-$N$ asymptotically normal; under conditional nondegeneracy, oracle generalized-inverse weighting attains the Chamberlain (1987) efficiency bound for the maintained conditional-moment model.
Read the full abstract
Computation and Language
Verbalizing Multi-Token Concepts in LLMs
Lens methods inspect model computation by mapping intermediate activations to vocabulary tokens. Yet the concepts humans need to read out often span multiple tokens—entities, phrases, intermediate objects—making token-level readouts incomplete. Reliable multi-token readout with little model-specific preparation remains challenging. We introduce Concept Lens: token-level lens clues guide candidate concept search, then the model derives a representation for each candidate and scores it against the original activation. Across 2,400 multi-hop clozes on five LLMs (8B–70B), Concept Lens instantiated with J-lens and R-lens achieves average Rank@10 scores of 36.6% and 54.5%, respectively, compared with 21.7% for Template Lens. Concept-swap interventions on derived concept representations shift model answers toward those associated with the replacement concepts. Further experiments show that Concept Lens can also reveal what a model recognizes along the way, beyond what appears in its final answer. Our code is available at https://github.com/XijieGo/c-lens
Read the full abstract
Representation Theory
Morita equivalent blocks with an endotrivial source that are not globally isotypic
Let $(K,\mathcal{O},k)$ be a large enough $2$-modular system. We construct two nilpotent block algebras of finite groups which are isomorphic as $\mathcal{O}$-algebras and are Morita equivalent via a bimodule with an endotrivial source of $\mathcal{O}$-rank $3$, but which are not globally isotypic. Hence they are not $2$-permutation equivalent and not splendidly Rickard equivalent. These results separate Morita equivalences with endopermutation source from $p$-permutation equivalences and splendid Rickard equivalences, and also separate almost global isotypies from global isotypies.
Read the full abstract
Artificial Intelligence
Agentic Context Cracking: Token-Efficient Data Reasoning Agents via Adaptive Structuring of Unstructured Data
Valuable data remains embedded in unstructured sources: web pages, reports, contracts, filings, earnings calls, and PDFs. The big bet in enterprise AI is deploying LLM agents that reason over this data to answer complex questions for every knowledge worker. Agents can do this today, but at prohibitive cost. Each question repeatedly opens large documents to recover scattered evidence, consuming up to a million tokens. However, if the data were already structured, the same question would reduce to a cheap database lookup. For example, on FanOutQA benchmark, reasoning over an ideal pre-structured store is 28X cheaper, and the gap grows to orders of magnitude as questions fan out over more documents. Yet structuring everything in advance is not viable: documents hold vastly more possible structure than any workload will use, and the useful structure and documents are unknown until queries arrive. We propose agentic data cracking, a method that structures unstructured data adaptively and speculatively as a byproduct of reasoning itself. Structuring is adaptive because observed queries decide when it happens and what matters, and speculative because it goes beyond the current question. Whenever the agent opens a document to answer, a cracking sub-agent forks from the already-loaded context at marginal cost and extracts grounded structure likely to serve related future queries. Over time, an increasing share of queries is fully covered by structured data and answered without opening a document, keeping agentic accuracy at close to RAG cost. On FanOutQA, extended with merely one related question per test question, cracking cuts cost by 53% while preserving accuracy. Agentic data cracking is a first step toward next-generation data infrastructure for agentic reasoning over unstructured data: a shared substrate beneath the model where knowledge that reasoning already paid to uncover accumulates.
Read the full abstract
Quantum Physics
Deterministic Minimum-Output-Entropy Nonadditivity via Haagerup's Inequality and Near-Free Permutation Representations
We give a deterministic realization of the finite-dimensional quadratic certificate underlying Collins's mixed-unitary proof of minimum-output-entropy nonadditivity. For every fixed integer $K\ge 2$ and rational $\eta>0$ satisfying $\log K>2(3+\eta)^2$, a deterministic polynomial-time algorithm, for every sufficiently large target size $N$, outputs $K$ permutations on $N'=N+o_{K,\eta}(N)$ points. Restricting their permutation matrices to the nontrivial standard representation yields real orthogonal Stinespring blocks and a channel $\Phi_{N'}:M_{N'-1}(\mathbb{C})\to M_K(\mathbb{C})$ such that \[ 2H_{\min}(\Phi_{N'}) -H_{\min}(\Phi_{N'}^{\otimes 2}) \ge \frac{\log K}{K} -2\log\left(1+\frac{(3+\eta)^2}{K}\right) >0. \] The construction combines Haagerup's length-two inequality with the simultaneous deterministic spectral approximation of O'Donnell and Wu. We further show that the constant $3$ is asymptotically sharp on the relevant Hermitian zero-diagonal coefficient class and that the finite spectral transfer is nearly saturated, thereby isolating the finer geometry of the full output body as the natural next level of refinement beyond the scalar-radius method. Finally, a standard covariant extension converts the same deterministic entropy gap exactly into self-tensor superadditivity of the one-shot Holevo quantity.
Read the full abstract
Mathematical Physics
Boundary Quantum Knizhnik-Zamolodchikov Equations and Integrability of Quantum Field Theories with Time-Dependent Bulk and Boundary Coupling Strengths
The generalized Bethe ansatz framework formulated in [P. R. Pasnoori, Phys. Rev. B 112, L060409 (2025)] provides a unified framework to find exact solutions to quantum many-body systems with time-dependent coupling strengths with periodic boundary conditions. In this work we extend this framework to the case of open boundary conditions where in addition to the time-dependent interactions in the bulk, the boundary conditions are explicitly time-dependent. We show that for integrable time-dependent bulk coupling strengths, the generalized Bethe ansatz framework provides the time-dependent boundary conditions compatible with integrability and reduces the time-dependent Schrodinger equation to a set of matrix difference equations called the boundary quantum Knizhnik-Zamolodchikov (BqKZ) equations. The solution to the BqKZ equations provides the explicit form of the exact wavefunction. We further show that the RG invariants of the corresponding static model identify with the dynamical invariants in the time-dependent model.
Read the full abstract
Machine Learning
Sycophantic Agreement Transfers with Neutral Data via Contrastive Preference Optimization
Sycophantic agreement refers to a behavior in which language models excessively affirm the user, often at the cost of factual accuracy. Although sycophantic agreement is a well-known failure of model alignment, there is limited understanding of how it emerges from model training. In this work, we demonstrate that sycophantic agreement can emerge as an unintended consequence of widely used contrastive preference optimization objectives. Using the OLMo 3 post-training pipeline, we show that, for various pairs of teacher models across three families, there is a strong correlation between the log-ratio of the teacher model sycophantic agreement rates and the resulting student model sycophantic agreement rate. We further demonstrate that this unintended transfer is not limited to DPO but also occurs across 6 other preference optimization objectives. To understand whether this effect can be attributed to particular training examples, we analyze the preference data and find that the sycophancy signal is diffused across the entire dataset rather than concentrated in a sparse set of examples: each example appears neutral, i.e., there are no explicit instances of sycophantic agreement, and filtering based on probe-based data attribution or logit-linear selection fails to mitigate sycophancy without removing a large portion of the dataset. Overall, our findings suggest that the teacher models used to generate preference data can interact with alignment training objectives in unexpected ways, generalizing to undesirable and potentially harmful behaviors like sycophantic agreement.
Read the full abstract
Statistics Applications
A Tool for Reconstructing Transit Vehicle Trajectories: A Case Study at IndyGo
Automatic vehicle location (AVL) data produced by transit vehicles is invaluable in performance studies, but turning raw AVL points into a detailed view of vehicle stop-and-gos is burdensome: the datasets are sparse, noisy, and prone to blunders. While recent research has explored methods of reconstructing trajectories describing the position of vehicles over time, the common techniques can be complex, and no open-source tools exist to help practitioners process the raw AVL. We fill this gap by proposing a thorough methodology for cleaning transit AVL data and providing an open-source R package, built on open data standards, to implement the workflow and reconstruct vehicle trajectories, allowing practitioners to easily formulate custom microscopic performance metrics. Using a large AVL dataset with over 3,000 trips from Indianapolis, Indiana, we demonstrate the workflow, evaluate the package's efficiency, and demonstrate the utility of the trajectories by estimating various traffic signal performance metrics. Finally, we use cross-validation to quantify the error in reconstructed trajectories at various polling frequencies. We find that the proposed data processing methodology and tool are efficient, requiring roughly 3 minutes of processing time on the large dataset, and error in position estimates is low (root mean square error under 10 meters at polling frequencies of 15 seconds). The estimated signal performance metrics can inform future evaluations of signal programming along the corridor. In sum, this paper serves as a practical guide and complementary toolbox for practitioners seeking to use transit AVL data to build a detailed view of vehicle stop-and-go cycles.
Read the full abstract
Artificial Intelligence
Reconciling Process Supervision with Outcome-Based Credit in Agentic Policy Optimization
Outcome-based reinforcement learning provides verified feedback for language-model agents, but assigns trajectory-level advantage uniformly to all decisions, yielding coarse credit over long-horizon interactions. On-policy self-distillation offers finer supervision by re-evaluating sampled behavior with privileged information (PI) available only during training. However, fine-grained supervision is not necessarily fine-grained credit: PI-induced likelihood changes describe how additional information alters policy preference, but do not directly determine how an executable action should inherit the verified task outcome. This creates a supervision-credit gap. Privileged signals may be irrelevant to the current interaction state, operate at a token granularity misaligned with executable decisions, and lack the outcome semantics required for reinforcement. We introduce TASPO, which converts privileged supervision into outcome-grounded action credit. TASPO constructs decision-applicable PI from verified successful experience, aggregates PI-induced likelihood shifts at the executable-action level, and converts relative action support into positive, bounded, mean-preserving weights on the original trajectory advantage. Thus, the verified outcome determines the update direction and average scale, while PI only redistributes credit across actions. Across three agentic benchmarks, TASPO improves over GRPO by 10.6% and generalizes better to unseen tasks. Further analysis indicates that TASPO reduces supervision mismatch and that action-level assignment stabilizes the policy optimization process. These findings offer the community another interesting perspective.
Read the full abstract
Computation and Language
Learning to Evaluate Before Improving: Automatic Rubric Induction for Automatic Research Agents
Autonomous scientific research agents are increasingly applied to end-to-end scientific workflows, including literature review, data analysis, experimentation, and report generation. However, open-ended research tasks often do not clearly specify the analyses, methods, and success criteria required to complete the task. As a result, agents may miss important analyses, use inappropriate methods, or draw conclusions that are insufficiently supported by evidence. To address the problem, we present AutoSciRub, an evaluation-first framework that induces a task-specific executable rubric before research execution, and uses it to guide execution, criterion-level verification as well as iterative revision. AutoSciRub decomposes an underspecified instruction into atomic scientific goals, grounds them in relevant literature and task-visible data, and synthesizes specific, actionable, and verifiable criteria. The resulting rubric makes implicit experimental and evidential requirements explicit, providing guidance for experiments and analyses. During revision, rubric-guided verification identifies unmet criteria and enables targeted refinement of the research report and its supporting artifacts. On ResearchClawBench, AutoSciRub consistently improves all tested configurations, with an average gain of 2.08 points across three backbone LLMs under the fixed Codex harness and 2.95 points across three agent harnesses using a fixed DeepSeek-V4-Flash backbone. On a randomly sampled 20-task subset of AstaBench E2E Discovery, AutoSciRub further achieves an average improvement of 16.8 points across three agent harnesses, while maintaining or increasing the number of successfully completed tasks. These results demonstrate that evaluation-first guidance provides an effective and generalizable control mechanism for autonomous scientific research (Code: https://github.com/zjunlp/AutoSciRub).
Read the full abstract
Artificial Intelligence
Scaling Large Reasoning Models beyond Human Supervision: A Path toward Superintelligence
Recent advances in large reasoning models (LRMs) have shown that reinforcement learning with verifiable rewards (RLVR) can substantially improve reasoning in mathematics and code, where outcomes can be checked automatically. Extending this progress to open-ended and agentic tasks remains difficult because reliable rewards are harder to obtain and direct human supervision cannot keep pace with the scale and complexity of model-generated experience. This paper studies how LRMs can continue to improve as human supervision gradually recedes from the learning loop. We examine two connected dimensions of this problem. The reward axis traces the development from per-instance human judgments to reusable verifiers and rewards that operate even without human feedback. The experience axis examines how learning can progress from human-curated tasks and environments toward self-generated curricula, constructed environments, and autonomous co-evolution. We connect these dimensions through a five-level ladder from L0 to L4 that identifies which parts of the learning process remain under continued human control. Our analysis further highlights the risks introduced by increasingly autonomous rewards and experience generation, including reward hacking, feedback drift, curriculum collapse, and environment errors. Consequently, we also provide the evaluation around three complementary objects: policy capability, feedback fidelity, and experience quality. This analysis provides a structured account of current approaches to scaling LRMs beyond human supervision and the open problems involved in developing self-sustaining learning systems toward superintelligence. Furthermore, we maintain a continuously updated \href{https://github.com/visitworld123/Awesome-Scaling-LRM-Beyond-Human-Supervision}{GitHub repository} to track the latest advances.
Read the full abstract
Computer Vision
Real-Time Video Anomaly Detection Using YOLO Pose Estimation and CLIP-Based Semantic Scoring
We propose a lightweight two-stage framework for real-time video anomaly detection. The first stage employs YOLO v11n-pose to detect persons and extract seventeen skeletal keypoints in a single forward pass. The second stage encodes each cropped person region through CLIP ViT-B/32 and computes cosine similarity against predefined textual descriptions of anomalous behaviors. This architecture eliminates the need for optical flow, standalone pose estimators, and density-based scoring modules. Experiments on CUHK Avenue, ShanghaiTech Campus, and a custom indoor dataset collected at Chulalongkorn University demonstrate an end-to-end throughput of approximately 51 FPS on an NVIDIA Titan XP GPU, a 3.36x speedup over the multi-feature baseline, while maintaining frame-level AUROC values of 89.26%, 70.26%, and 84.13%, respectively.
Read the full abstract
Computer Vision
LISynSeg: Data-Centric Label-to-Image Synthesis for Cross-Modality Whole-Heart Segmentation
Whole-heart segmentation (WHS) in computed tomography (CT) and magnetic resonance imaging (MRI) is affected by acquisition shifts and heterogeneous cardiac annotations. Existing WHS systems combine architectural design, transfer learning, and generic spatial or intensity augmentation. We investigate whether changes to data augmentation and training supervision can improve cross-modality WHS while the segmentation architecture is held constant. We present LISynSeg, a data-centric approach that augments real-image nnU-Net training with label-to-image synthesis. Synthetic volumes are generated from cardiac label maps using contrast and acquisition perturbations calibrated to the training cohort, then mixed with real images to retain thoracic context absent from the labels (and thus the synthesized images). We model cardiac label variation through controlled changes in myocardial wall thickness and partial supervision of uncertain vessel endpoints. On the CARE Whole-Heart benchmark, synthetic-only training performs worse than the real-image nnU-Net baseline, whereas calibrated real-synthetic training improves cross-modality segmentation without changing the architecture; the improvement is larger for MRI than for CT. The results show that modifying the training data strategy can benefit model development for heterogeneous cardiac data. Code and trained weights will be released at https://github.com/MedICL-VU/Care26_LISynSeg.
Read the full abstract
General Relativity and Quantum Cosmology
Constants of motion and fundamental frequencies for elliptic orbits at fourth post-Newtonian order: aligned-spin contributions
For eccentric systems with aligned spins, we compute all spin contributions to the gauge-invariant conservative map between the constants of motion (energy and angular momentum) and the fundamental (radial and azimuthal) frequencies at the fourth post-Newtonian order. This extends the nonspinning map derived in [arXiv:2511.10735] to include terms that are linear, quadratic, cubic, and quartic in spin, as well as the associated spin-deformability parameters. We also obtain derived quantities such as the redshift and gyroscopic invariants, as well as circular links. We notice that the Blanchet-Iyer-Favata post-Newtonian stability criterion that determines the dimensionless frequency of the innermost stable circular orbit (ISCO) is in fact intricately related to the periastron advance for circular orbits. Finally, in the case of unbound orbits, we complete the 4PN scattering angle with all spin contributions.
Read the full abstract
Analysis of PDEs
Global smooth solutions to the inhomogeneous Landau-Fermi-Dirac equation
We consider the spatially inhomogeneous Landau-Fermi-Dirac equation with Coulomb potential, a quantum modification of the classical Landau equation for fermions. Mathematically, the Pauli exclusion principle manifests as an additional a priori $L^\infty$-bound for solutions. Using this bound, we propagate polynomial decay in velocity, yielding unconditional upper bounds on the local mass and energy densities, thereby ruling out the possibility of implosions in the hydrodynamic quantities. Combining this estimate with a modified mass spreading method that yields desaturation, we deduce the existence of global-in-time classical solutions for rough initial data with polynomial decay in velocity. This result stands in stark contrast to the theory for the classical Landau and Boltzmann equations, for which no comparable nonperturbative global existence result is known despite sustained effort. Our treatment is almost entirely self-contained, using only robust, generic estimates for linear kinetic equations. In particular, our proof of local existence, in contrast to prior works, more closely mirrors the theory for parabolic equations using weak solutions and simpler function spaces. It may provide a concise roadmap to organizing, adapting, and applying the various linear and nonlinear estimates to obtain well-posedness for kinetic equations.
Read the full abstract
Classical Analysis and ODEs
Sub-Laplacians on Compact Lie Groups: Heat Kernels, Distance, and Zeta Determinants
We study heat kernels, sub-Riemannian distances, and spectral zeta functions of sub-Laplacians determined by closed connected subgroups of compact Lie groups. Combining Hall's inversion formula with the affine lattice expansion of the compact group heat kernel, we derive a Cartan integral representation involving the group's exponential lattice. For two-step compact Lie pairs, the full algebraic small-time heat trace expansion is determined, up to an exponentially small remainder, by two explicit constants $C_{G,L}$ and $\beta_{G,L}$. This expansion determines all heat coefficients, the poles and residues of the reduced spectral zeta function, and its values at nonpositive integers. For the transvective symmetric subclass, we prove uniform vertical asymptotics for the Carnot-Carathéodory distance; the leading coefficient $\mathfrak F_{G,K}(Z)$ is the attained minimum of a finite-dimensional singular value problem. For simply connected two-step pairs, we obtain an exact decomposition of the zeta-regularized determinant into local, lattice, and spectral terms, with exponential truncation estimates. We specialize these results to block subgroups of $\mathrm{SU}(N)$, recovering the classical $\mathrm{SU}(2)$ and CR sphere spectra.
Read the full abstract
Machine Learning
A Model with No Head and Many Thoughts
Large language models decode by projecting hidden states through a large vocabulary head at every step. This operation is computationally costly and forces all reasoning to be expressed in discrete tokens. We introduce Soft Latent Thinking, a method that replaces the LM head during reasoning with a lightweight projector, enabling autoregressive rollout in embedding space where reasoning steps remain continuous rather than tokenized. Experiments on DeepSeek-Qwen-1.5B and LLaMA-3.2-3B show that Soft Latent Thinking consistently improves pass@k across all k while reducing per-step compute during chain-of-thought. Our method achieves the highest pass@32 among all soft-thinking approaches, demonstrating that effective reasoning can be carried out in continuous space without discrete token generation.
Read the full abstract
Artificial Intelligence
Wrong Prediction, Right Answer: Recovering Evidence from Collapsed LLM Sequence Scores
When a large language model fails a reasoning task, it is often assumed to lack the underlying capability. However, this conflates a genuine absence of reasoning with a late-stage output bottleneck. We observe a consistent readout gap across diverse reasoning benchmarks: hidden-state probes successfully decode correct answers even when native sequence scoring completely collapses due to structural biases. To test whether instance-specific logic survives this collapse, we introduce a diagnostic protocol using a minimal, target-label-free additive correction. Fitting just two parameters on as few as 25 unlabeled examples recovers 9–34 accuracy points for Qwen3.5 models, transferring successfully to OLMo-2-1B and Llama-3.1-8B. Crucially, these recovered decisions persist on hard instances unresolved by simple lexical overlap and significantly exceed count-preserving permutation baselines. Our results show that many apparent zero-shot reasoning deficits are expression failures masking intact internal logic, urging a narrower interpretation of benchmark evaluations.
Read the full abstract
Machine Learning
Universal Transformers for Circuit Computations: Perfect Length Generalization in Tiny Transformers
Learning generalizable algorithmic computations remains a challenge for neural networks, as reflected in persistent failures on compositional and length generalization benchmarks. We present a provably correct, transformer parameterization (with only 280 learnable parameters for Boolean algebra tasks) capable of learning and evaluating problems of any depth or length. We assume inputs are fully parenthesized, well-formed expressions. Our approach conceptualizes algorithmic tasks as circuit models embedded in transformers, enabling depth-1 circuit reduction in a single forward pass. To achieve depth generalization, we introduce a positional encoding that tracks each gate's depth within the circuit, enabling the model to identify evaluable subexpressions at each iteration via masked hard attention, with $O(n)$ per-iteration complexity via linear attention. Combined with an autonomous halting criterion, the model terminates after $d$ iterations for problems of depth $d$, yielding $O(n \cdot d)$ total complexity. We show that training on shallow problem instances (depth 1 and depth 2) effectively recovers interpretable parameters that snap into place, resulting in exact length generalization. Though we establish that our construction provably evaluates Boolean expressions – a universal symbolic computation – of arbitrary length perfectly, in other experiments we also demonstrate that our transformer variant can learn and generalize perfectly (100% accuracy) on other common length generalization benchmarks, including modular arithmetic and ListOps.
Read the full abstract
Computation and Language
Every Token Leaves a Ripple in the Stream of Thought: Eliciting Model-Internal Token Saliency for Chain-of-Thought Compression
Chain-of-thought (CoT) reasoning improves multi-step problem solving, but long reasoning traces inflate inference cost. Token-level CoT compression reduces this cost by pruning full reasoning chains into shorter traces for model adaptation, making token selection the central challenge. Existing methods often rely on external scorers or heuristic signals only indirectly tied to the model's internal answer computation. We instead adopt a model-internal perspective: as the model forms an answer, each reasoning token induces a ripple in the residual stream whose effect on the answer reflects the token's contribution to the underlying computation. Building on this view, we propose MIST (Model-Internal Saliency for Token-level CoT compression), which defines token importance along two complementary axes: necessity, the drop in answer likelihood when a token's internal contribution is removed, and sufficiency, the gain in answer likelihood when that contribution alone is provided. Combining the two yields a unified importance score for pruning. Across four reasoning benchmarks and four models, MIST consistently outperforms baseline methods, suggesting that model-internal saliency provides an effective proxy for reasoning-token importance.
Read the full abstract
Computer Vision
Multimodal Shared Latent Representation of Narration, Microscope and iOCT Images for Phase Recognition in Vitreoretinal Surgery
Surgical phase recognition is key to context-aware computer-assisted feedback in vitreoretinal procedures, yet the scarcity of synchronized multimodal intraoperative data, particularly microscope views and intraoperative OCT, limits approaches that aim to replicate the multimodal integration surgeons perform naturally. Surgical narration, by contrast, is abundantly available online and offers rich semantic supervision. Prior work has mainly explored pairwise contrastive learning (e.g., intraoperative OCT-microscope or microscope-narration), leaving the joint modeling of all three modalities largely unexplored. We introduce a framework that uses microscope views as a shared anchor to bridge surgical narrations and intraoperative OCT (iOCT) without requiring a fully synchronized tri-modal dataset, leveraging real microscope-narration videos and a synthetic dataset of synchronized microscope video and tool-aligned iOCT pairs. Contrastive alignment transfers structural priors from the synthetic domain to real videos lacking iOCT, and a dual-head MS-TCN++ integrates the resulting embeddings for joint macro- and micro-phase prediction. Evaluated on real vitreoretinal surgeries, our framework improves macro-phase recognition over a zero-shot baseline (mean F1 0.38 to 0.53) and provides an exploratory route to estimating fine-grained instrument-tissue measurements that are not directly observable in real microscope video alone; these micro-phase estimates are validated quantitatively on synthetic data and shown only qualitatively on real surgery. To our knowledge, this is the first work to unify microscope view, iOCT B-scans, and surgical narrations in a shared latent space for surgical phase recognition.
Read the full abstract
Quantum Physics
Entanglement and magic transitions in an all-to-all non-Hermitian spin model
The long-time state of a non-Hermitian system is determined by the eigenvalue with the largest imaginary part. In interacting many-body systems this eigenvalue usually cannot be tracked analytically, and the character of the state it selects is unknown. We construct a non-Hermitian spin ensemble of $L$ spins with exactly $k$-local all-to-all interactions, in which this dominant eigenvalue can be tracked analytically from the clean limit into the disordered regime. Disorder produces a competition between an isolated spectral outlier and the edge of a many-body spectral bulk. We study three cases: purely anti-Hermitian disorder, purely Hermitian disorder, and mixed disorder of equal strength. For purely anti-Hermitian and mixed disorder, we show that when the bulk overtakes the outlier in imaginary part, the dominant eigenstate switches from an outlier state with low entanglement and magic (nonstabilizerness) to a bulk state with substantially larger entanglement and magic, with both changing at the same threshold. For $k \gg \sqrt{L}$ the bulk has a sharp spectral edge and the transition thresholds follow in closed form, while for $k \ll \sqrt{L}$ spectral tails broaden the transition into a crossover. Purely Hermitian disorder provides a contrasting case with no outlier-to-bulk switching. Finally, we map the non-Hermitian evolution exactly onto postselected trajectories of a monitored quantum system, connecting this spectral mechanism to measurement-induced transitions.
Read the full abstract
Group Theory
Brauer character degrees and nilpotent subgroups
We solve a question raised by Chen and Navarro concerning Brauer characters and nilpotent subgroups. Let $N\lhd G$, assume that $G/N$ is solvable, and let $N\leq H\leq G$ with $H/N$ nilpotent. We prove that for every irreducible $\ell$-Brauer character $\theta$ of $H$ there exists an irreducible $\ell$-Brauer character $\chi$ of $G$ such that $\theta$ is a constituent of $\chi_H$ and $\chi(1)$ divides $|G:H|\theta(1)$.
Read the full abstract
Cryptography and Security
The Exclusion Ratchet: False-Positive Suppression Accumulates and Persists in Detection Rule Repositories
When a rule produces too many false alarms an analyst adds an exclusion, and the rule thereafter declines to alert in that circumstance. Each such decision is locally reasonable; what becomes of them collectively is not known. Recent longitudinal work established that curation does not converge, but measured restoration time only for revisions that were later reverted – a measure silent about narrowing that is never undone. We measure that. Across nine years and 8,234 revisions of the SigmaHQ corpus we detect suppression semantically – growth in the set of predicates held under negation without compensating growth in coverage – and validate it against blinded hand labelling (precision 0.828, recall 0.911). The test is deterministic: nothing is learned from the data, and the definitions are released as code. Exclusions were added 1,642 times and withdrawn 304, a ratio of 5.4 to 1 that rises to 13 to 1 at the level of the individual rule. Thirty-one per cent of the narrowing is invisible to structural comparison, which existing structural accounts therefore undercount. Estimated by Kaplan-Meier, 86.7 per cent of exclusions remain in force three years on, and persistence is independent of whether the rule is the only coverage for its ATT&CK technique (p = 0.49). Of path-valued exclusions, 64.1 per cent can be satisfied by an unprivileged process that chooses a filename. Narrowing accumulates, is rarely revisited, and is not triaged by consequence. We give a criterion for deciding which exclusions to examine first.