Artificial Intelligence
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Friday 11 September
While multimodal reasoning has advanced rapidly, solving complex geometry problems critically hinges on active visual assistance, such as constructing auxiliary lines, spurring the rise of Visual Chain-of-Thought (VCoT). However, existing evaluations typically assess visual generation quality and final answer accuracy in isolation, failing to examine whether intermediate visual aids are geometrically valid, effectively utilized in subsequent reasoning, or causally responsible for task success. To bridge this gap, we introduce GeoVAD-Bench, a diagnostic benchmark that pairs a fine-grained five-dimensional trajectory diagnosis covering perception, auxiliary quality, utilization, deductive reasoning, and final correctness with controlled No-Aux, Auto-Aux, and GT-Aux intervention settings to systematically isolate intermediate error modes, the causal gains of visual aids, and the resulting autonomy gap. Our findings reveal that while high-quality auxiliary aids offer substantial theoretical gains for geometric problem solving, autonomous generation is frequently hampered by compounding errors across geometric perception, faithful visual manipulation, visual-state grounding, and deductive reasoning. Guided by these diagnostic insights, we establish a specialized data construction pipeline encompassing geometric perception, diagram editing, and interleaved visual-textual reasoning trajectories, and develop a progressive SFT and multimodal RL training framework. The resulting model, GeoWeave-8B, outperforms the base model by +25.3% in final geometric accuracy and achieves a +30.4% gain in process average across the four intermediate diagnostic dimensions.
Cryptography and Security
A Feature-Rich Embedded NIDS with eBPF/XDP: Detector and Architecture Trade-offs
Distributed Denial-of-Service (DDoS) attacks remain a serious threat to transport networks, with recent attack volumes exceeding 30 Tbps, and the telecommunications industry being the main target. Recent work has yet to study the impact of the hosting software architecture on network monitoring solutions, or to assess recent algorithms for improving attack detection. This paper presents a Network Intrusion Detection System (NIDS) for DDoS detection in transport networks, developed in collaboration with Ericsson. Building on a statistical baseline, we improve detection effectiveness with an Isolation Forest trained on a wider set of flow features, extracted by GoFlowMeter, our open-source Go implementation of CICFlowMeter, and we integrate eBPF/XDP so that the NIDS filters real traffic at the kernel level. We further compare three deployments, monolithic, Kafka-based, and gRPC-based microservices, on a Raspberry Pi 5 testbed replaying the CIC-DDoS2019 dataset as real network traffic. Detection quality is governed mainly by the choice of detector rather than by the transport: the Isolation Forest raises recall and F1 score (0.965 live in the monolithic variant) over the baseline by flagging low-volume attack windows that the baseline misses. The transport is not neutral, however: gRPC reaches almost the same accuracy as the monolithic variant while adding less than 2 milliseconds of transport time per window, whereas the asynchronous Kafka pipeline trails by roughly nine percentage points and adds about 27 milliseconds. These findings clarify the trade-off between detection quality and architectural overhead when deploying a NIDS on resource-constrained hardware.
Read the full abstract
Quantum Physics
Violating a Bell Inequality: What to Do When You Lose Your Quantum System
The detection loophole is a well-known loophole in Bell nonlocality motivated by the inefficiency of detectors used in Bell experiments. However, this inefficiency is actually a ubiquitous feature of any photonic platform for quantum networks since photons are readily absorbed by the surrounding environment. When the quantum system, usually a photon, is lost, what should the parties output? The most natural choice is to fall back to a deterministic strategy where each party produces an output via a function of her local input. In this paper, we study how to choose such a fallback strategy for general Bell inequalities with respect to different efficiencies $\eta$ to maximize the possible violation. We mathematically prove that for the CHSH inequality, there exists a fallback strategy that is optimal for all $\eta$. However, in general, we find that the optimal fallback strategy can vary with $\eta$, sometimes in surprising ways. For example, we find an example of a Bell inequality where the optimal fallback strategy near the threshold efficiency is not an optimal deterministic strategy in the lossless setting. Our results can reduce the efficiency requirements for closing the detection loophole and thus are useful for applications of Bell inequality violation, such as quantum telepathy or device-independent quantum key distribution.
Read the full abstract
Adaptation and Self-Organizing Systems
Dynamics-preserving network reductions for ride-pooling paths
Reducing the complexity of ride-pooling paths is a central challenge in systems with distributed demand. Here we show that such dynamics admit an exact coarse-grained representation: for a broad class of routing algorithms whose decisions depend only on path lengths, the full network can be reduced to an effective network of active nodes weighted by shortest-path distances without altering the resulting trajectories, up to stochastic degeneracy breaking. The reduction therefore defines an equivalence class of network representations generating identical path dynamics. We further demonstrate that for globally optimizing dispatchers this equivalence is systematically violated through degeneracy amplification, yet remains quantitatively accurate beyond the exactly solvable regime. Our results identify when spatial structure can be integrated out without loss of dynamical fidelity, providing a general framework for the analysis of interacting path processes.
Read the full abstract
Distributed Computing
GreenDirector: carbon- and water-aware workload placement for sustainable computing
The rapid growth of data center electricity demand, accelerated by AI, makes carbon-only accounting an incomplete measure of computing's environmental impact: low-carbon electricity mixes are often water-intensive, and the resulting harm depends on local, seasonal scarcity rather than on the volume of water consumed. We propose the Environmental Score (ES), a unified, dimensionless index in $[0, 100]$ that jointly captures the carbon footprint and the spatial-temporal, scarcity-weighted water impact of the electricity a workload consumes. It combines real-time, cross-border electricity flow tracing with monthly AWARE2.0 water-scarcity characterization factors, weighting global greenhouse-gas emissions together with the local, seasonal severity of water stress. Building on it, we define the Green Score (GS), a scheduling metric proportional to the useful computational work delivered per unit of real environmental impact, which also accounts for data center power and hardware efficiency. We add both metrics as a green-affinity feature to the GreenDirector schedulers of two production federated infrastructures, the AI4EOSC scientific cloud and the DIRAC workload management system. In AI4EOSC, a cluster-filling experiment over four pan-European providers shows that greener sites are filled first without degrading scheduling latency or end-user experience. In DIRAC, a trace-driven simulation of 133,631 jobs and a preliminary production deployment for the KM3NeT community reduce carbon emissions and improve carbon efficiency by about 40%, while making the carbon-water trade-off explicit when the lowest-carbon site also carries higher water stress. The results show that hydrological stress can be dynamically weighted into workload placement in live, multi-tenant systems
Read the full abstract
Statistics Theory
Closed-form estimation of the exponential scale under two-sided informative random censorship
We estimate the scale of an exponential lifetime from observations censored randomly from both sides, under an informative model in which each censoring law is a power of the lifetime survival function – a two-sided generalization of the Koziol–Green proportional hazards model. Substituting the empirical distribution function into the likelihood equation yields a closed-form pseudo-maximum-likelihood estimator of the scale that is computed in one pass over the order statistics and needs neither iteration nor numerical optimization. A digamma identity shows the estimating equation to be exactly, not merely asymptotically, Fisher-consistent. Strong consistency and asymptotic normality are established with no restriction on the censoring depth; the key is an identity that rewrites the underlying $L$-statistic without its unbounded score function and so removes the condition that classical limit theorems would impose. The influence function and the asymptotic variance are explicit in polygamma functions, so confidence intervals require no numerical integration. Against the information bound of the observed-data model the efficiency exceeds $98.6\%$ in the designs considered and stays above $96.2\%$ over a wide sweep. Only the entry half of the model is needed: the estimator remains exactly Fisher-consistent and strongly consistent whatever the right-censoring law, and exponentiality is shown to be the precise price of that freedom. The construction extends verbatim to a proportional hazards class with known baseline. Three data sets illustrate the procedure, including a cohort in which left censoring is genuine.
Read the full abstract
Human-Computer Interaction
TraceMind: Predicting User Information Uptake from Low-Cost Interaction Traces during Human-LLM Content Co-Generation
In human-LLM content co-generation, AI-generated information can enter final artifacts without being adequately processed by users, creating risks when artifacts are shared or acted upon. We study whether recognition-level uptake of atomic information units can be assessed in open-ended co-generation and predicted from low-cost interaction traces. We collected data from 62 participants across three tasks. For each final draft, we extracted atomic information units and generated post-task recognition questions, yielding 1187 unit-level uptake labels. We present TraceMind, which tracks units across Chat and Draft histories, aligns interaction traces with changing on-screen layouts, and models spatial, temporal, and workflow-informed evidence. TraceMind outperformed all learned baselines across AUROC, AUPRC-non, balanced accuracy, and macro-F1. We found that uptake unfolds throughout interaction, with sustained active engagement providing informative evidence beyond isolated signals. Our work shifts human-LLM co-generation from content adoption toward what users actually take up, motivating uptake-aware systems grounded in low-cost interaction traces.
Read the full abstract
Machine Learning
SIMS: Scale-Invariant Merit-Function-Based Scalarization for Multi-Task Learning
Multi-task learning (MTL) requires navigating unavoidable trade-offs among competing objectives. This paradigm is frequently formulated as multi-objective optimization (MOO), where the scalarization is favored to reduce an MOO problem to a single objective. We empirically find that existing merit-function-based scalarization approaches are sensitive to the relative scales of different objectives in practical MTL, where task losses commonly differ by orders of magnitude. The optimization process often favors objectives with larger scales even though the underlying Pareto optimal solutions remains invariant to rescaling (i.e., multiplying an objective by a positive constant). To address this issue, we propose Scale-Invariant Merit-function-based Scalarization (SIMS) for MTL. Specifically, SIMS adopts a transformation-induced merit function to convert the MOO problem of MTL to a single objective that renders optimization invariant to the magnitudes of losses. Theoretically, we prove that the requirement for scale invariance uniquely determines this transformation to be logarithmic. We further show that this general transformation-induced merit function preserves weak Pareto optimality and admits a smooth surrogate with controllable approximation error. Extensive experiments on representative multi-task benchmarks demonstrate that SIMS consistently outperforms existing scalarization methods and achieves state-of-the-art performance.
Read the full abstract
Statistical Mechanics
Certifying Hidden Dissipation from Observed Current Fluctuations
A single-molecule experiment on a driven enzyme or motor resolves only a few of the transitions the machine makes, yet one would like to know how much free energy it dissipates in total, including on the steps that stay hidden. The mean and the fluctuations of the currents on the watched transitions are shown to certify this hidden dissipation, with no knowledge of the transition rates and no access to the hidden transitions: when the watched transitions span the cycles of the network, the observed-current covariance recovers the full density-contracted quadratic current geometry, including the contribution of transitions that are never seen. The mechanism is that fluctuations are set by the dynamical activity, or traffic, the symmetric partner of the current. Contracting the large-deviation cost of empirical currents over density fluctuations identifies the observed-current covariance with a traffic-weighted metric, and turns partial observation into a minimum-energy completion problem over the unseen cycles, whose solution is the certified hidden cost. Existing current-fluctuation uncertainty relations bound dissipation from chosen currents but do not say when partial observation fixes the hidden contribution; the cycle-observability condition derived here does. The statements concern long-time means and fluctuations; finite-time estimates require separate error control.
Read the full abstract
Databases
Invisible Yet Dominant: Big Stalls of Kernel I/O Mechanisms in Cloud OLTP Databases
Most databases, including PostgreSQL, RocksDB, and recent AI KV-cache middleware, rely on buffered I/O, delegating write-back to the Linux kernel. On the distributed block storage standard in the cloud, this delegation inherits a hidden bottleneck: each device is drained by a single kernel flusher thread over a high-latency, shallow-queue path. When the drain falls behind, dirty throttling pauses write() system calls, and even reads that must evict dirty pages stall. These stalls are invisible to iostat and every standard counter. This poster observes the stall from inside the kernel, using the multi-volume data placement proposed in SteelDB as the experimental lever. eBPF probes on writeback and block tracepoints separate write-back by issuing context and count every throttle pause. Across three configurations with identical provisioned IOPS and bandwidth but 1, 2, and 4 devices, we show that adding drains, not bandwidth, cuts throttle pauses by 70%, reduces maximum transaction latency by 59%, and raises throughput by 23%.
Read the full abstract
Mesoscale and Nanoscale Physics
Sub-kelvin cryogenic wafer probing verified by on-wafer primary Coulomb blockade thermometry
Scalable cryogenic wafer level testing is emerging as an essential tool for the development of quantum devices and cryogenic microelectronics, but the main motivation, the actual operational on-wafer temperature, remains largely unverified. In this work, utilizing a cryogenic wafer prober (CWP) enables wafer-scale electrical testing of Coulomb blockade thermometers (CBTs), while the CBTs simultaneously provide an independent static and dynamic assessment of the prober's thermal performance through primary and in situ calibrated secondary electron thermometry, realizing a reciprocal characterization approach. Using recently introduced scalable normal metal TiW-based CBTs, which require no superconductivity suppressing magnetic fields, we have measured on-wafer electron temperatures of 630-660 mK across a full 150 mm wafer, representing, to our knowledge, record-low electron temperatures measured in a CWP. These exceed the nominal 600 mK chuck setpoint by approximately 40 mK on average, demonstrating the necessity of direct on-wafer thermometry. Self-calibrated secondary CBT thermometry resolves the thermal transients caused by chuck movement and reveals a thermal settling time within 5 minutes after probe touch down. This reciprocal wafer-scale approach simultaneously evaluates the device under test and the thermal performance of the measurement environment, providing a practical methodology for scalable sub-kelvin characterization of quantum devices and cryogenic microelectronics.
Read the full abstract
Robotics
A Hierarchical Coverage Path Planning Algorithm for Unknown Environments
This paper presents an online coverage path planning algorithm for unknown environments. During navigation, the initially unknown search area is progressively decomposed into disconnected subareas as new obstacle information is acquired and coverage proceeds. These subareas are organized in an incrementally constructed decomposition tree that preserves their hierarchical parent-child relationships. Based on this tree, a global coverage tour is maintained and updated online by prioritizing newly generated child subareas according to their exploration states and distances from the robot. A local planner then generates coverage motions within each selected subarea, allowing the robot to adapt its trajectory as the environment is gradually revealed. Its performance is evaluated via high-fidelity simulations in complex scenarios. The results show improved coverage efficiency in terms of path length and overlap ratio in comparison to three baseline algorithms.
Read the full abstract
Machine Learning
Poisson-Corrector Complexity Bounds for Moreau–Yosida Unadjusted Langevin Sampling
We study the classical Moreau–Yosida unadjusted Langevin algorithm (MYULA) for $\pi(\,\mathrm{d} x)\propto e^{-f(x)-g(x)}\,\mathrm{d} x$, where $f\in C^2(\mathbb{R}^d)$ is $m$-strongly convex with $L_f$-Lipschitz gradient and $g:\mathbb{R}^d\to\mathbb{R}$ is convex and globally $G$-Lipschitz. For the Moreau-smoothed target $\pi_\lambda$ and the MYULA invariant law $\widehat\pi_{\lambda,h}$, we prove \[ \sqrt m\,W_2(\pi_\lambda,\widehat\pi_{\lambda,h}) =O(h)+\widetilde O(h^{3/4}) \] under $0<h(L_f+\lambda^{-1})\le c$, with only logarithmic dependence on $\lambda^{-1}$ in the error coefficients. Combining this estimate with the Moreau approximation bias yields $\widetilde O(\varepsilon^{-4/3})$ iterations to achieve $\sqrt m\,W_2(\mu_N,\pi)\le\varepsilon$, for fixed model parameters and initialization. The proof combines a discrete Poisson corrector with active-trace estimates and a shared-noise bound for the exact–Euler two-point curvature.
Read the full abstract
Quantitative Methods
Audiovisual diarization of overlapping click trains in sperm whale (Physeter macrocephalus) vocal sparring using a three-hydrophone array
The individual attribution of sperm whale (Physeter macrocephalus) vocalizations during surface interactions constitutes a methodological challenge due to acoustic overlaps, multipath propagation, body shadowing, and indistinguishable inter-pulse intervals among similar-sized individuals. A multimodal audio-visual workflow is presented to deinterleave and attribute uncharacterized click trains produced by immature males during “vocal sparring” using a portable three-hydrophone array coupled with synchronized video. The approach combines spectro-temporal tracking, based on inter-click interval dynamics and the Constant-Q Transform, with spatial time-difference-of-arrival modeling projected onto the image plane for optical validation. Analysis of 2,655 manually validated clicks shows that purely acoustic clustering diarization errors remain low, peaking at 25.79% only during extreme temporal superpositions. Integrating the optical modality resolves residual spatial indeterminacies; although near-planar array geometry induces vertical ambiguities, the system achieves up to 100% horizontal visual concordance for primary emitters. Crucially, this framework successfully reconstructed and assigned 11 distinct, intertwined click trains totaling 882 clicks to specific focal individuals despite near-field tactile constraints. Because ethological descriptions remain incomplete without identifying the emitter, explicitly correlating these emissions with physical kinematics provides the fine-scale resolution required to define this socio-acoustic behavior.
Read the full abstract
Quantum Physics
Factoring six-digit integers with superconducting quantum circuits
Integer factorization is a central computational problem with important applications in public-key cryptography. Here, we demonstrate a quantum factorization protocol using a superconducting circuit. Microwave drives are used to engineer a highly tunable effective two-level Hamiltonian whose eigenvalues can be measured spectroscopically. The target integer \(N\) and each candidate factor pair (\(p\),\(q\)) are encoded into the amplitudes, frequencies, and phases of the applied microwave fields. By scanning the candidate pairs while monitoring the spectral response at zero energy, we identify the factor pairs of integers up to six digits. The protocol requires neither two-qubit gates nor quantum entanglement. Its performance is currently limited by the precision of microwave control and the finite linewidth of the spectroscopic response. Improved control accuracy and longer coherence times would extend the accessible range of integers.
Read the full abstract
Machine Learning
Where Decoder Cosine Similarity Fails for SAE Feature Flow Discovery
Foundation models are increasingly adapted through fine-tuning, model editing, and alignment procedures while retaining previously acquired capabilities. Understanding the internal computations that support these adaptations is therefore becoming increasingly important for continual model evolution. Sparse autoencoders (SAEs) provide interpretable feature dictionaries for residual-stream activations and sublayer outputs, but it remains unclear how state features and update features interact to produce downstream residual features. In this work, we focus on MLP updates as a first test case. We construct a transition atlas of triples $s_k + u_j \rightarrow t_\ell$, where a residual-state feature and an MLP-update feature jointly predict a target residual feature, and validate candidate triples by ablating the decoded update feature. In a 20M-token Pythia-160M $L_7 \rightarrow L_8$ run, we find 38,125 strong ablation-effect transitions, but 88.0% have both state-target and update-target decoder cosine similarity below 0.7. As a preliminary cross-model check, a run of 20M-token Gemma-3-4B $L_{21} \rightarrow L_{22}$ causally validates only the top 30,000 ranked candidate triples by ablating the decoded update feature, and 53.6% of strong-effect triples have both state-target and update-target decoder cosine similarity below 0.7. The Gemma result is directionally consistent with Pythia, but weaker, since update-target cosine recovers many of the strongest Gemma effects and the run is not a full-atlas causal validation. Ultimately, our results suggest that feature flow atlases can serve as diagnostics of representation-update mechanisms and thereby inform tools for steering model updates. Future work will validate more complex patterns across layers, models, and SAE families.
Read the full abstract
Statistics Theory
Tight Sampling Complexity with Stochastic Gradient Oracles in Fixed Dimensions
We establish the stochastic-gradient query complexity of sampling smooth strongly log-concave distributions in every fixed dimension $d\geq1$. For $\mu$-strongly convex, $L$-smooth potentials with unknown minimizers $x_f^\star$ in the ball $\mathbb{B}(0,\mu^{-1/2})$, and unbiased gradient oracles with variance at most $\sigma^2$, the minimax worst-case expected query complexity is $\Theta\left(\log(1+\kappa)+\frac{\sigma^2}{\mu\varepsilon}\right)$, jointly optimal for the condition number $\kappa:=L/\mu$, variance $\sigma^2\ge 0$, and TV accuracy $0<\varepsilon\leq 1/10$. In the noiseless setting $\sigma=0$, this tight complexity $\Theta\left(\log(1+\kappa)\right)$ is independent of $\varepsilon$. Moreover, a gradient-only sampler generates an exact sample with ${O}(\log(1+\kappa))$ worst-case expected queries. Without previous ball $\mathbb{B}(0,\mu^{-1/2})$, exact sampling from any initial point $x_0$ can be implemented with expected cost ${O}\left(\log(1+\kappa)+\log(1+\sqrt{\mu}\|x_0-x_f^\star\|)\right)$, without knowing the initial distance. We also show that dependence on initial distance is generally unavoidable.
Read the full abstract
Algebraic Geometry
On symplectic aspects of $\mu$-equivalence of a class of isolated hypersurface singularities
A $\mu$-constant deformation of isolated complex hypersurface singularities of dimensions different from 2 preserves the diffeomorphism class of the Milnor fiber. If it would be known that $\mu$-constant deformations have no vanishing folds as defined in by O'Shea, this result is even true in all dimensions since the members in the family would have a common Milnor ball-radius. In this note a symplectic version of this is shown which in particular applies to weighted homogeneous polynomial singularities.
Read the full abstract
High Energy Physics (Theory)
The Hagedorn transition as a Plancherel wall: Tracy-Widom statistics, Polyakov loops and partial deconfinement
We study the Hagedorn transition of the single-winding Dutta-Gopakumar unitary matrix model at finite rank $N$. Its Schur expansion is a sum of Plancherel probabilities restricted by the finite-rank wall on the number of rows, and the Vershik-Kerov-Logan-Shepp limit shape reaches this wall when the number of boxes equals $N^2/4$. Two scaling windows around the Hagedorn point resolve the wall at different scales. In the outer window the Baik-Deift-Johansson theorem gives a Tracy-Widom crossover, and summing it identifies Liu's critical free-energy coefficient with the first moment of the GUE Tracy-Widom distribution. In the inner window the same sum determines the canonical probability law of the Polyakov loop at the Hagedorn point: the normalized loop $\mathrm{tr}\,U/N$ is uniformly distributed on a disk of radius one half, so the order parameter does not self-average at large $N$. The intensity $|\mathrm{tr}\,U|^2/N^2$ has the same limiting law as the Schur degree $n/N^2$: at the Hagedorn point the canonical degree weights become asymptotically equal below the macroscopic finite-rank wall. Under the partial-deconfinement dictionary the uniform disk becomes a linear probability density for the deconfined color fraction, and its fully deconfined endpoint is the Young-diagram endpoint of Berenstein and Yan. The Tracy-Widom mean also fixes the first finite-$N$ correction to the Polyakov-loop Laplace transform and to all of its fixed radial moments. Bessel-Toeplitz evaluations up to $N = 100$ confirm the free-energy coefficient, the uniform law, and the predicted correction.
Read the full abstract
Classical Analysis and ODEs
A weighted restriction estimate in $\mathbb R^2$
We establish a weighted restriction estimate in $\mathbb{R}^2$. Our proof combines new two-ends Furstenberg estimates with refined decoupling.As an application, our result yields decay estimates for the circular $L^p$-means of the Fourier transforms of fractal measures in $\mathbb{R}^2$.
Read the full abstract
Artificial Intelligence
Reproducing and Evaluating the Generalizability of Subliminal Learning in Open-Weight Models
In this reproduction paper we investigate subliminal learning, a consequence of distillation where teacher models transmit behavioral preference traits through semantically unrelated data. The original paper explores two types of traits (animal preferences and misalignment), three data modalities (number sequences, code, and chain of thought), and several model families. We reproduce their experiments and extend the setup along three axes: new preference categories (actors and politicians), a new task (chess move generation), and an additional open-weight model (Ministral8B). We also run a controlled ablation on the numbers task's answer-space size (1-, 2-, and 3-digit sequences). We focus on open-weight models with accessible checkpoints on HuggingFace, since the original paper's GPT-4.x fine-tuning is no longer available. Our reproduction supports the original paper's claims, but our extensions show they are not universal as transmission strength varies across traits and tasks, and one model shows almost no effect at all.
Read the full abstract
Complex Variables
Mobius invariant weighted reduced quaternionic Qp-modules
In this paper, we introduce some weighted reduced quaternionic modules induced by a reduced quaternionic Cauchy-Riemann operator associated to arbitrary orthogonal bases of R3. A weighted conformal invariant property of the modules is given.
Read the full abstract
Machine Learning
Clustering-Based Balanced Sampling and Allocation with Data Parallelism for High-Performance Fine-Tuning
Instruction-tuning datasets for large language models (LLMs) are often large, redundant, and imbalanced, limiting efficient adaptation. Naive large-batch fine-tuning repeatedly includes overrepresented sample groups while weakly covering underrepresented but informative ones, especially under data parallelism (DP) across multiple GPUs. We propose CluSTER, a Cluster-aware balanced Sampling framework for Training Efficient data Reduction in DP instruction tuning. CluSTER curates a representative reduced dataset through gradient-space clustering and DP-aware balanced allocation, ensuring dual-level coverage across clusters and workers, while preserving the original data distribution by weighted update. As a result, CluSTER reduces redundant computation and improves training stability without compromising model quality. Across multiple instruction-tuning datasets, CluSTER reduces training time by up to 69.6% with almost no accuracy loss compared to prior sampling and data reduction methods. Code is available at https://github.com/kaist-dmlab/CluSTER.
Read the full abstract
Analysis of PDEs
Large-Time Behavior of Pseudo-Parabolic Equations Associated with Generalized Total Variation Energies
In this paper, we study the well-posedness and large-time behavior of a pseudo-parabolic problem associated with a generalized total variation energy in the $BV$-framework. A main mathematical issue arises from the mismatch between the Sobolev regularity of solutions at finite times and the $BV$-structure of the corresponding steady-state problem. We prove the existence and uniqueness of solutions and investigate the relationship between their large-time behavior and solutions to the steady-state problem. Moreover, our results include the convergence of the solution trajectory to the unique steady-state solution under a typical setting arising in image processing.
Read the full abstract
Cryptography and Security
NovaFabric: Tamper-Evident, Replayable Evidence for Autonomous AI Agent Runs
When an autonomous AI agent does something consequential, what can be proven about what it did? Agent-observability platforms capture traces, but a trace is mutable: alterable undetected, with no recipe for re-executing it, silent on whether captured secrets were removed. Regulation (EU AI Act, ISO 42001, NIST AI RMF) presumes records an independent party can check. We present NovaFabric, producing audit-grade execution evidence: provider-neutral, tamper-evident, replayable, shareable. It records an agent run, without modifying agent logic, into a portable Run Capsule (fifteen-entity schema), sealed with a holistic DSSE signature, RFC 3161 timestamp, Merkle log and redaction attestation. Sealed runs are re-executable under a four-mode replay protocol and exportable as an Evidence Bundle for third-party verification with stock tooling (specified, not evaluated). The contribution is integration, not new cryptography: OpenTelemetry, DSSE/in-toto and W3C PROV. We evaluate eight research questions at measured scope. Mocked replay serves every model response from the capsule (no live model call, 10/10) but is offline w.r.t. models, not the network; only 2/10 tool-using workloads completed; the gap is missing tool-response substitution. Tampering is rejected across three tested classes. Declared-stream completeness is 0.652 (95% CI +/-0.064, ten scenarios). A repaired rule pack redacts 14/14 credential types, preserving 9/9 decoys; diff localises 140/140 mutations. Blast-radius queries: 45.5ms p99 over 10M edges (3.3x faster than a columnar baseline), 167.9ms over 100M (1 client, n=30). A 314-machine, ten-region run finds capsule REST ingest lossless but capped at 61.6 req/s (p99 26.8s) by per-worker serialisation. Six defects found in NovaFabric and its evaluation corpus: four fixed, one withdrawn, one open. Verification is conditional on a stated trusted computing base.
Read the full abstract
Logic
Fraïssé's conjecture, partial impredicativity and well-ordering principles, part II
We exhibit a well-ordering principle that is equivalent to a theory of partial impredicativity. The latter goes back to Towsner and relates to recent work of Suzuki and Yokoyama. Together with part I of the present paper, this yields a combinatorial upper bound on the axiomatic strength of Fraïssé's conjecture.
Read the full abstract
Cryptography and Security
MicroHasTEE: Bare-Metal Haskell for Type-Level Peripheral Ownership on Armv8-M
Arm TrustZone for Armv8-M isolates Secure and Non-secure software, but developers must still coordinate peripheral attribution, interrupt routing, initialization, and gateway interfaces across separately built firmware images. Inconsistent assumptions between these images can compile successfully and emerge only as faults on the target device. We present MicroHasTEE, a multiparty programming framework that expresses both firmware applications as participants in one typed Haskell program. MicroHasTEE represents peripheral authority with type-level capability ledgers and uses indexed setup computations to track resource acquisition, configuration, transfer, and finalization. Domain-specific effect types restrict peripheral operations and interrupt callbacks to the participant that holds the corresponding authority, while typed callable handles describe the Secure services available to Non-secure code. MicroHs compiles the shared program twice to produce separate bare-metal Secure and Non-secure firmware images. We implement MicroHasTEE for an STM32U5 Nucleo board, including TrustZone configuration, peripheral drivers, and a serialized gateway for cross-domain Haskell calls. For programs expressed through its interface, MicroHasTEE rejects inconsistent resource use, attribution changes after configuration, callbacks in the wrong domain, and calls to unregistered Secure services. A door-lock case study demonstrates feasibility, with firmware images occupying 232.7 KiB and 228.4 KiB of flash and approximately 220 KiB of SRAM per domain.
Read the full abstract
Machine Learning
SCOPE-OPSD: Fisher-Conditioned Privileged Subspaces for On-Policy Self-Distillation
On-policy self-distillation (OPSD) scores student-generated prefixes with a solution-conditioned self-teacher, yet transfers supervision only through next-token probabilities. We ask whether the aligned final-layer discrepancy offers a useful second channel, and how to test that channel without confusing its geometry with auxiliary strength. SCOPE-OPSD projects the privileged teacher-student residual onto a frozen rank-64 factor estimated from residual covariance and language-model-head Fisher sensitivity. It reuses the forwards already required by OPSD and adds neither rollouts nor inference-time modules. A matched Random control preserves the structured factor's rank and nonzero spectrum and uses per-arm gradient-RMS calibration, isolating the effect of the data-dependent orientation. Across the complete 25/50/75/100-step trajectories for Qwen3-1.7B, 4B, and 8B, Structured is never below Pure OPSD, with strict gains in 11 of the 12 model-checkpoint combinations and an exact tie at 4B step 25. Structured also exceeds matched Random in 10 of the 12 combinations. At step 75 on Qwen3-1.7B, Structured exceeds matched Random by 1.39 Macro Avg@12 points in each of two independent training reruns. A cross-fitted diagnostic also shows 4.40 times greater held-out privileged-gap capture than the matched random orientation. The results support a compact, Fisher-conditioned privileged subspace for short-budget OPSD.
Read the full abstract
Artificial Intelligence
From Collaboration to Capability: Internalizing Routed LLM Experts into Compact Reasoners
A compact controller can coordinate stronger experts by selecting whom to consult, formulating requests, and integrating their responses. We study whether learning from both the controller's decisions and the experts' reasoning and code improves its generation after expert removal. We introduce Rivet for collaboration internalization: expert-augmented reinforcement learning applies a shared outcome signal to controller decisions and returned expert spans, and verified trajectory internalization consolidates complete successful interactions through format-aware supervised training. The deployed controller generates reasoning, code, and interaction structure with local Python execution and no external LLM. Across seven competition-mathematics benchmarks, RIVET-1.7B and RIVET-4B achieve average accuracies of $28.25\%$ and $44.16\%$; Stage~II improves RIVET-4B's accuracy after expert removal by $6.49$ points, and GPQA-Diamond results provide evidence of generalization to scientific reasoning. Ablations show gains from ordinary trajectory supervision and additional format weighting, supporting the effectiveness of training on the content and structure of verified collaborations.
Read the full abstract
Computer Vision
SCORE: SubDistribution-aware Collaborative Knowledge Reinforcing for Cloth-Hybrid Lifelong Person Re-Identification
Lifelong Person Re-Identification (LReID) aims to train a unified person retrieval model from a non-stationary data stream. Existing LReID methods mainly focus on scenarios where the clothing of each person is consistent. Recently, the Cloth-Hybrid LReID (CH-LReID) where cloth-consistent and cloth-changing data alternately occur, has emerged as a more practical and challenging scenario. Due to the conflict between clothing-relevant and clothing-irrelevant knowledge, the well-known catastrophic forgetting problem is significantly exacerbated in this task. To address this issue, we propose a SubDistribution-aware COllaborative Knowledge REinforcing (SCORE) framework, where our key idea is explicitly modeling the intra-identity diversity to continually consolidate distinct cloth-consistent and cloth-changing knowledge. Specifically, an Adaptive SubDistribution Modeling mechanism is developed, where a set of distributional subprototypes is assigned to each identity to capture the intra-identity diversity, improving the compatibility between cloth-consistent and cloth-changing knowledge. Then, a Distributional Knowledge Reinforcement scheme is introduced, where the knowledge of old distributional subprototypes is retained in the new ones by a collaborative aligning mechanism. Extensive experiments show that our SCORE achieves the state-of-the-art performance. Our code is available at https://github.com/zhoujiahuan1991/ECCV2026-SCORE