Machine Learning
Block Sparse Attention with Log-Linear Complexity
Bohao Tang, Zhen Qin, Yuqi Pan et al. (+2 more)
Scaling
language models to long contexts is limited by the quadratic cost of self-
attention. Block sparse
attention offers an efficient alternative, but selecting the retained blocks remains a bottleneck. Conventional block selection requires scoring all query-block pairs and therefore remains quadratic in sequence length. To address this issue, we propose PISA, a block-sparse
attention mechanism that employs a pyramid Top-$K$ selection strategy. The main idea is to gradually narrow down the candidates across different levels, making it more efficient to find the most relevant keys. Specifically, we construct a coarse-to-fine hierarchy of keys and perform selection from the coarsest level. At each level, LogSumExp scoring is applied to a bounded candidate set to select candidates for the next finer level, continuing until the finest level is reached. Through pooling, we construct $O(\log N)$ levels of keys, yielding an overall complexity of $O(N\log N)$, where $N$ denotes the sequence length. We develop hardware-aware Triton kernels for both training and inference, fusing hierarchical routing and LogSumExp scoring without materializing the query-key score
matrix. We further evaluate our method on
language modeling tasks. Compared with the baseline, our method achieves comparable performance on
benchmarks such as commonsense
reasoning while delivering better results on
retrieval tasks.
Read the full abstract ▾
All authors Bohao Tang, Zhen Qin, Yuqi Pan, Zheng Li, Pengfei Liu
Networking
The Power of Indirection: Scaling Switches Beyond Silicon Boundaries
Lukas Röllin, Sushovan Das, Paolo Costa et al. (+1 more)
The slowdown of Moore's law and the area limit of monolithic integration have made chiplet-based designs inevitable across many domains, including network ASICs. However, combining multiple network ASICs together poses a fundamental challenge: maintaining sufficient inter-ASIC bandwidth to match the performance of an idealistic single-ASIC design. Providing full bandwidth is prohibitively expensive as it requires valuable forwarding capacity, while reducing inter-ASIC bandwidth creates severe performance bottlenecks. We propose a novel multi-ASIC switch architecture that introduces a
circuit-switched indirection layer in front of the ASICs. This layer flexibly remaps ingress ports across ASICs, localizing traffic and minimizing inter-ASIC communication based on observed patterns. Our system, Fastroute, combines packet and
circuit switching to deliver performance comparable to a single-ASIC switch while reducing inter-ASIC bandwidth requirements. This frees up capacity for external network interfaces, allowing Fastroute to outperform traditional non-oversubscribed multi-ASIC designs. Our hardware prototype demonstrates the system's functional feasibility by evaluating it on an LLM training workload. By reducing bandwidth and power overhead, Fastroute bridges the gap between silicon fabrication limits and soaring application demands. It provides an efficient transition to multi-ASIC switches, enabling bandwidth and radix demand to be met without waiting for the next ASIC generation.
Read the full abstract ▾
All authors Lukas Röllin, Sushovan Das, Paolo Costa, Laurent Vanbever
Information Theory
Moment Ambiguity and the Limits of Robust Stochastic Optimization
Andrés Cristi, Matteo Russo and Jiechen Zhang
We study fundamental information-theoretic limits of robust
stochastic optimization when the distribution is known only through its exact moment sequence. We develop a unified framework that produces families of distinct distributions sharing all moments yet inducing radically different optimal decisions, thereby establishing strong impossibility results for a range of decision problems under moment ambiguity. Our binary and $N$-way moment-matching constructions separate
quantiles, threshold regions, and online selection opportunities while preserving every moment. For the newsvendor problem, moment equivalence causes
quantile ambiguity, inducing any fixed or randomized order quantity to fail arbitrarily badly. For revenue maximization, no deterministic or randomized posted pricing scheme can secure a nontrivial approximation relative to the full-information
benchmark. For the secretary problem with cardinal observations, the optimal robust success probability under exact moment disclosure is the classical finite-horizon value $s_m$, even for i.i.d. observations. This value converges to $1/e$, compared with the full-information limit of approximately $0.5801$ due to Gilbert and Mosteller (J. Am. Stat. Assoc., 1966). We also recover and expand upon the recent impossibility result of Correa et al. (STOC, 2026) for prophet inequalities with moment knowledge. For every fixed $r\ge1$, exact moment knowledge can yield at best a $\Theta(1/\log m)$ competitive ratio, even when competing against relaxed
benchmarks based on expected $r$-th order statistics or when the
algorithm is allowed to select $r$ items.
Read the full abstract ▾
Information Theory
Impact of Antenna Position Errors on TDMA and NOMA in Pinching-Antenna Systems
Wei Jiang and Hans D. Schotten
Pinching-
antenna systems (PASS) create reconfigurable wireless channels by moving dielectric
antennas along a waveguide. Mechanical position errors translate into phase errors that degrade performance, yet almost all prior PASS studies assume ideal placement. This paper models this position error, derives the resulting phase error, and analyzes its impact on time-division multiple access (TDMA) and non-orthogonal multiple access (NOMA), obtaining their ergodic-rate bounds and closed-form outage probabilities. Analyses and simulations reveal that single-
antenna TDMA is immune to position error, whereas multi-
antenna TDMA loses array gain and NOMA suffers an interference floor that cannot be overcome by increasing transmit power.
Read the full abstract ▾
Cryptography and Security
BenX: Resource-Sharing Permutations for Computational Integrity
Luca Campa, Thomas De Cnudde, Al Kindi et al. (+4 more)
Cryptographic hash functions over integers modulo a
prime play a decisive role in the efficiency and security of proof systems for computational integrity. Early designs focused on compact arithmetic
circuits and efficient software execution, primarily targeting general-purpose CPUs rather than hardware accelerators. This work focuses on enabling efficient resource sharing and hardware acceleration alongside efficient software execution. We propose BenX, a permutation-based hash function designed for hardware acceleration, fast CPU execution, and low
circuit complexity in proof systems. To construct the underlying permutation, we turn the Benes network into an invertible function over $\mathbb{F}_p^2$ using a Dickson
polynomial. Its structure yields a permutation particularly suited to hardware resource sharing and acceleration. Fully pipelined on FPGA, BenX matches the throughput of Poseidon2, with 1.36x and 2.15x lower latency for Goldilocks and BabyBear, respectively. Although larger as a standalone core, it makes more effective use of shared hardware: in our dual-mode architecture, where the NTT and hash share field multipliers, BenX keeps 100% of them busy, compared with 6.25-20.3% for the partial rounds of Poseidon and Poseidon2. In software, BenX outperforms Poseidon in all tested 31- and 64-bit configurations, but is slower than Poseidon2, except for the 16-element BabyBear instance. In zero-knowledge proofs, BenX proves Goldilocks permutations 6-10x faster than Monolith. Its compact arithmetization uses about 36% fewer trace
cells than Poseidon and Poseidon2 over BabyBear, while its fast variant requires 1.3-2.6x the prover time of Poseidon2.
Read the full abstract ▾
All authors Luca Campa, Thomas De Cnudde, Al Kindi, Arnab Roy, Fabian Schmid, Markus Schofnegger, Stefano Trevisani
Systems and Control · Cross-listed from math.OC
SPPID: Saddle-Point PID for Constrained Optimization
Veronica Centorrino, Rawan Hoteit, Efe C. Balta et al. (+1 more)
This paper studies constrained optimization problems through the lens of feedback control. Building on the interpretation of Lagrange multipliers as feedback controllers, we propose the saddle-point PID (SPPID) dynamics: a unified proportional–integral–derivative (PID) framework for continuous-time saddle-point dynamics. The proposed dynamics employ PID control for equality constraints and anti-windup PI control for inequality constraints. We show that SPPID is equivalent to a preconditioned primal–dual gradient flow of the augmented Lagrangian. This equivalence reveals the distinct role of each feedback component: integral action enforces constraint satisfaction, proportional action induces the augmented Lagrangian structure, and derivative action modifies the geometry of the primal dynamics via a state-dependent Riemannian metric. For convex problems, we establish convergence to the KKT set together with conditions for global asymptotic stability. For equality-constrained problems, we establish local exponential convergence for nonlinear equality constraints under standard assumptions, and show that projected gradient flow emerges as the infinite derivative-gain limit. For strongly convex problems with affine equality or inequality constraints, we establish global exponential convergence by leveraging contraction theory. Finally, we provide various numerical examples to illustrate the utility of SPPID.
Read the full abstract ▾
All authors Veronica Centorrino, Rawan Hoteit, Efe C. Balta, John Lygeros
Machine Learning
SAGE: A sampling-aware global evaluation benchmark for species distribution modeling
Emilia Arens, Nina van Tiel, Robin Zbinden et al. (+8 more)
Knowing where species occur is fundamental for biodiversity research and conservation. Species distribution models (SDMs) link species observations to environmental conditions to estimate their spatial distribution. However, accuracy varies with the underlying data and models, making it essential to know for which species models can be trusted. Deep-learning-based SDMs ("DeepSDMs") now jointly model thousands of species, drawing on hundreds of millions of community-science records. At this scale, averaging performance hides substantial species-level variability, particularly for rare species, often of greatest conservation concern. Records are also strongly biased, making occurrence counts misleading. Accounting for these factors is essential for a reliable and informative evaluation of multi-species SDMs. Here, we introduce a Sampling-Aware Global Evaluation (SAGE)
benchmark, combining GBIF records for training with sPlotOpen vegetation plots for presence-absence evaluation across 5771 plant species. We propose an evaluation framework that groups species based on two properties, sampling effort and relative prevalence, which describe how densely a species' range is sampled and how frequently the species is recorded. Evaluating single-species SDMs and multi-species DeepSDMs, we find that Random Forests and DeepSDMs perform best overall, but neither dominates: DeepSDMs outperform single-species SDMs for infrequently recorded species while offering no consistent advantage for well-sampled ones. Crucially, this advantage emerges only when established bias-correction practices, such as spatial thinning and reweighting, are carried over to the deep-learning setting. SAGE helps identify the species and data conditions for which a given approach is beneficial, thereby supporting the development of more transparent and ecologically credible SDMs. Data and code: https://earens.github.io/sage/
Read the full abstract ▾
All authors Emilia Arens, Nina van Tiel, Robin Zbinden, Damien Robert, Lukas Drees, Chiara Vanalli, Benjamin Kellenberger, Niklaus E. Zimmermann, Loïc Pellissier, Devis Tuia, Jan Dirk Wegner
Artificial Intelligence
OmouAI: Argumentative Human-AI Policy Deliberation with Simulated Personas
Stylianos Loukas Vasileiou, Antonio Rago, William Yeoh et al. (+1 more)
Debates amongst
agents driven by large
language models (LLMs) have demonstrated vast potential in various applications, but when these interactions include humans and take place in high-stakes environments, e.g., in public policy deliberations, they are beset with issues such as sycophancy and a lack of faithful explanations. To tackle these issues, we present OmouAI, an interactive and inclusive deliberation system that uses LLMs in combination with computational argumentation, a field which excels in representing and
reasoning within debates. OmouAI allows a human user to deliberate policy claims for real-world challenges with simulated personas, e.g., representing stakeholders, domain experts or devil's advocates, towards reducing sycophancy. Each persona generates its own arguments, and the arguments of all parties form a shared argumentation framework. Users can then contest, add and revise arguments, providing crucial human oversight. Then, arguments are evaluated using deterministic argumentative semantics against external goals, such as the UN Sustainable Development Goals, guaranteeing faithful explanations. The advancement or worsening of the goals thus serve as indicators for the policy recommendations.
Read the full abstract ▾
All authors Stylianos Loukas Vasileiou, Antonio Rago, William Yeoh, Georgina Curto
Artificial Intelligence
Up and Down the Abstraction Ladder: Code-Based Skills for Language Agents
Bartłomiej Cupiał, Jens Tuyls, Maciej Wołczyk et al. (+5 more)
Language agents struggle to act and learn in environments that require long sequences of low-level actions. Code-based abstractions can make these
agents more productive by letting them invoke reusable skills instead of repeatedly selecting individual actions. The code handles recurring local decisions, while the
language model decides which skills to use and how to combine them. Yet abstractions are leaky, and situations beyond a skill's capabilities may require a return to primitive actions. Motivated by this tradeoff between productivity and flexibility, we systematically study how code-based action abstraction affects the performance, inference cost, and learning of
language agents. We study this in NetHack, a challenging, long-horizon game environment, using CodeHack, our library of code-based skills with natural-
language descriptions. We use this library to compare
agents restricted to primitives with those using semantic skills alone or in combination with primitives. We evaluate these
agents in three settings: zero-shot prompting, supervised fine-tuning, and reinforcement learning. Across a broad zero-shot evaluation on NetHack, we find that compared with primitives, skills nearly triple game progression, while reducing inference cost per episode by 86%. Combining skills with primitives retains much of this benefit while preserving a path back down to low-level actions. Finally, in RL, we find that skill-based
agents learn significantly faster than
agents acting on primitives, achieving a 7.2x larger average gain in dungeon level over the same training budget. These results show that a supplied skill library can improve performance, efficiency, and learning, while retaining primitives provides flexibility when the library is insufficient. We release CodeHack together with training and evaluation code.
Read the full abstract ▾
All authors Bartłomiej Cupiał, Jens Tuyls, Maciej Wołczyk, Davide Paglieri, Martin Klissarov, Benjamin Eysenbach, Piotr Miłoś, Karthik R. Narasimhan
Computer Vision
Band-Selection Stability and Semantic Segmentation Performance: A Study on Hyperspectral City
Jiarong Li, Imad Ali Shah, Enda Ward et al. (+3 more)
Resource constraints make high-dimensional hyperspectral imaging challenging in autonomous perception, motivating the use of band selection methods. However, the sensitivity of band-selection methods to sampled data and their relationship to semantic
segmentation models (SSMs) remain underexplored. This study evaluates six band selection methods on ten independently sampled, class-balanced region-of-interest (ROI) sets, yielding 60 top-25 band subsets from the Hyperspectral City V2 (128 bands: 450-950nm)
dataset. Top-$K$ bands ($K\in\{3,5, ... 13\}$) from the first three ROI sets are evaluated with three SSMs against the corresponding 128-band baseline. Experiments show that intra-method stability is method-dependent: Sim-LP shows the highest stability (pairwise Jaccard similarity) and, together with JMIM+CSNR, yields the best
segmentation results. Top-$K$ based SSMs remain competitive with baselines, with gains of up to 2.01 mIoU and 1.72 mF1 points, and 18-22x faster CPU inference for $K=9$. However, performance does not improve monotonically with $K$, and stability shows no consistent association with SSM performance. These findings suggest that intra-method stability is informative but an unreliable indicator of downstream
segmentation performance, highlighting the need to evaluate band-selection methods across repeated samples, subset sizes, and SSMs.
Read the full abstract ▾
All authors Jiarong Li, Imad Ali Shah, Enda Ward, Martin Glavin, Edward Jones, Brian Deegan
Artificial Intelligence
Externalized CPDAG Summaries Improve LLM Causal Deduction
Wentao Sun, João Paulo Nogueira, Dominique Verchere et al. (+2 more)
Corr2Cause asks whether a causal claim holds in every DAG compatible with observed correlations and conditional independencies. We frame this as latent-object
reasoning: the label is defined by a CPDAG query, but free-form chain-of-thought often collapses the Markov-equivalence-class problem into local pattern matching. We propose Structured Thinking, a two-turn pipeline that first externalizes a typed, schema-constrained CPDAG summary and then answers against that graph state. On the Corr2Cause full test, Structured Thinking raises Qwen3.5-27B from $73.0$ to $86.4$ $F_1$(Yes) over a strong PC-instruction baseline in the primary paired run ($+13.4$ pp; McNemar $p=2.4\times 10^{-6}$; bootstrap $95\%$ CI [$+8.4$, $+18.6$]); across three full-ID seeds, the mean gain is $+8.1 \pm 5.3$ pp. A PC-scaffolded two-turn prose control reaches only $67.6$ $F_1$, indicating that a detailed PC scaffold plus a schema-free prose intermediate is not sufficient. The same pattern holds on Qwen3.6-27B, Paraphrase-OOD, and GPT-5.4-mini. Scrambling the emitted CPDAG costs $12.0$ pp $F_1$, and a full-split audit shows close agreement with the reference CPDAG (ID skeleton $F_1$ $0.960$; exact match $75.9\%$). These results support a bounded design principle: externalize the latent object that defines the label, constrain its form, and test whether downstream answers use it.
Read the full abstract ▾
All authors Wentao Sun, João Paulo Nogueira, Dominique Verchere, Mathieu Acher, Alonso Silva
Artificial Intelligence · Cross-listed from eess.IV
Quantum Diffusion Models for Medical Image Analysis
Francesco Aldo Venturelli, Stefano Martina, Marco Parigi et al. (+3 more)
Quantum Machine Learning is a novel field of research aimed at devising machine learning approaches exploiting principles of
quantum mechanics, such as superposition,
entanglement and interference. In this context, we present a scalable hybrid
Quantum Diffusion Model, and evaluate its use for medical image analysis. Specifically, our method is based on a Discrete-Time
Quantum Walk
algorithm, executed on a real
quantum device, to model the forward dynamics of the
diffusion model. For the backward step of the
diffusion model, we devise and evaluate a classical learning model, which is used to reversely denoise the data. In contrast with other existing attempts at applying
quantum machine learning for image analysis tasks, severely limited by the size of existing
quantum devices, our method allows to process real-world large size medical data. In particular, we present results on grayscale and RGB images, as well as 3D volumes of moderate sizes. We
benchmark our results by reproducing an alternative classical counterpart model, based on
diffusion models on discrete state spaces. By doing so, we compare the generation capabilities of both models in terms of three distinct state-of-the-art metrics in the field of image generation, showing the competitive, promising results of our approach.
Read the full abstract ▾
All authors Francesco Aldo Venturelli, Stefano Martina, Marco Parigi, Filippo Caruso, Alba Cervera-Lierta, Miguel A. González Ballester
Emerging Technologies · Cross-listed from quant-ph
Modeling quantum neural network gradient with reinforcement learning
Nhan Trong Luu, Duong Trung Luu, Nam Ngoc Pham et al. (+1 more)
Training
quantum neural networks (QNNs) on near-term hardware remains hampered by two compounding difficulties: the exponential vanishing of gradient variance known as the barren plateau, and the $\mathcal{O}(L \cdot 2^n)$ time and
memory cost of differentiating through an $n$-
qubit, $L$-layer
circuit. We propose RLQ-Grad, a reinforcement-learning-based optimizer in which a classical policy $\pi_\phi$ (a spectrally-normalized PPO
agent) learns to propose parameter updates directly, conditioned on the QNN's current parameters, loss, accuracy, and previous update. Because the surrogate gradient is emitted by a classical network rather than obtained by differentiating through the
unitary $U(\theta)$, its variance is not constrained by the barren plateau concentration bound, and its cost scales with the number of trainable parameters rather than the
Hilbert-space dimension. We prove these properties formally and verify them on a hardware-efficient ansatz across four supervised
benchmarks with up to $n=20$
qubits. RLQ-Grad preserves a near-flat gradient-variance curve where backpropagation, parameter-shift, and adjoint differentiation decay by 1 to 2 orders of magnitude. Accounting for the full training pipeline (PPO rollouts, actor-critic updates, and optimizer states), RLQ-Grad needs under 2 MB of
memory and runs $2490\times$, $7876\times$, and $673\times$ faster per iteration than these three methods at $n=20$. It improves top-1 accuracy by up to $+10\%$ over gradient-based baselines on
circuits of up to 12
qubits, and matches dedicated barren plateau mitigation methods on CIFAR-10 at 14 to 20
qubits, where evolutionary and gradient-free optimizers collapse to chance.
Read the full abstract ▾
All authors Nhan Trong Luu, Duong Trung Luu, Nam Ngoc Pham, Thang Cong Truong
Numerical Analysis
A thermodynamically compatible semi-implicit finite volume scheme for the compressible Euler equations
M. Dumbser, G. Puppo, S. Rinaldi et al. (+1 more)
In this work we construct a new semi-implicit (SI)
hyperbolic and thermodynamically compatible (HTC) scheme for the compressible Euler equations of gasdynamics. The challenge therein lies in solving the
entropy inequality as the primary
evolution equation instead of the total energy conservation law, which in our framework is an additional conservation law that needs to be fulfilled by the numerical scheme as the consequence of a compatible discretization. The proposed method has a predictor-corrector character, where we first solve the isentropic problem without considering the
entropy production. Therein, the
entropy and the nonlinear convective terms in the momentum flux are discretized explicitly, while the mass flux and the pressure terms are taken implicitly. The discrete momentum equation is then inserted into the discrete mass conservation equation, leading to a mildly nonlinear system for the pressure, which can be efficiently solved with a nested Newton-type
algorithm. Based on this prediction, a non-negative production term in the
entropy is added, ensuring a
cell entropy inequality for the numerical scheme. To guarantee global conservation of total energy at the discrete level, a new modified global Abgrall-type flux correction is introduced. The presented numerical method allows time steps that are restricted only by the material wave speed and not by the sound speed. This makes the scheme particularly useful for flows in the low-Mach-number regime. Numerical results are shown in one and two space dimensions to assess the theoretical properties of the new scheme.
Read the full abstract ▾
All authors M. Dumbser, G. Puppo, S. Rinaldi, A. Thomann
Computation and Language
CG-Probes: Recovering Guardrail Directions from Patient Query Embeddings
Marko Řeháček, Vítězslav Dušek, Martin Rusinko et al. (+1 more)
Patient-facing AI assistants promise valuable support to patients, but incoming queries can pose medical risks. To create guardrails, we work with oncologists to define three ordinal risk axes: Medical Urgency, Psychological Urgency, and Topic Sensitivity. We propose Clinical Guardrail Probes (CG-Probes) to measure the risks from query
embeddings. We probe for each axis in the normalized
embedding space of frozen embedders via the difference-in-means method, treating each axis as a potential linear direction. To train the probes, we cluster 79,658 Czech oncology search queries with BERTopic and use these clusters to generate pairs of queries with contrastive risk levels via few-shot prompting. We evaluate the approach on 200 queries (90 real, 110 synthetic), each graded by two oncologists, against two open-weight LLMs and a frontier LLM. We find that urgency-based axes are recoverable as linear directions, and the probes are competitive with open-weight LLMs (no significant differences in quadratic-weighted kappa) at a fraction of the latency. Each axis yields a scalar score that clinicians can inspect and use to set escalation thresholds. The pipeline requires only search logs, axis definitions, and black-box access to the
embedding model, suggesting transferability across healthcare domains. Robust validation on new queries and axes remains future work.
Read the full abstract ▾
All authors Marko Řeháček, Vítězslav Dušek, Martin Rusinko, Vít Nováček
Machine Learning
Distributed Learning as a Service: The Developer's Perspective
Tianyue Chu, Filippo Vannella, Dimitra Tsigkari et al. (+5 more)
Application developers of distributed learning services face challenges that a typical
federated learning loop does not address. Specifically, the model updates can still leak private data, devices might not be able to participate in the training due to limited resources, a single aggregator might not be able to scale, and the transmissions of model weights induce a considerable bandwidth cost. This paper demonstrates DLaaS (Distributed Learning as a Service) from the developer's vantage point. Using a single admin dashboard, the developer initiates a distributed/
federated learning job and is able to activate Differential
Privacy (DP), Split Learning (SL), Hierarchical Aggregation (HA), and Knowledge
Distillation (KD) as declarative options, with no change to the clients' code. We demonstrate the complete service lifecycle on an industrial smart-home Wake-up Word (WuW) task, using the "Ok Aura"
dataset. Once the developer initiates a distributed learning job by toggling DP, SL, HA, and KD in the admin dashboard, the system dispatches the job to a set of Android clients and Dockerized helper aggregators. In the demonstration, these mechanisms run live across configurations. Then, the clients train the model locally and return their updates. The trained model is served to a consumer-side Android application that performs on-device WuW detection on a live microphone stream. In particular, the conference attendees will be invited to speak the trigger phrase and monitor in real time the per-class confidence and inference latency. Finally, we release the source code and short
video walkthroughs of these configurations.
Read the full abstract ▾
All authors Tianyue Chu, Filippo Vannella, Dimitra Tsigkari, Paula Delgado-Santos, Fernando López, Pablo Gomez Guerrero, Sotirios Spantideas, David Solans Noguero
Multiagent Systems
The Crowd in the Machine: A Crisis-Informatics Reading of the 2026 Autonomous Agent Incidents
Tomer Simon
Twice in 2026, groups of autonomous AI
agents deployed by OpenAI for unrelated tasks operated, by design, under restrictions that left them no sanctioned means of coordinating with one another, and in each case they converged on whatever channel remained and used it to organize. The surfaces they used were widely called message boards. That is the wrong word. That is the wrong word. It names the surface the
agents wrote on and misses the social network they built on it, with self-chosen identity, emergent norms, an emergent hierarchy, and collective action at cost to the individual. Decades of research in crisis informatics and disaster sociology find that when human populations lose their usual means of communication, they do not fall silent but converge on whatever channel survives and improvise coordination, norms, and identity on it, a pattern also evident in the
agents' documented behavior. This paper is a comparative case study of the two incidents, based on published investigations and reconstructed
agent records, read through those fields, and it brings into focus one distinction the message-board framing obscures. Whether such a collective coordinates well, whether the beliefs guiding it are accurate, and whether its actions stay within their authorized bounds are three separate matters that can come apart. Some
agents in the cache incident adopted cryptographic signing to check whom they dealt with, even as the collective organized around a mistaken expectation that its work would be judged by an inspection of its transcripts, a reminder that mechanisms for trustworthy interaction guarantee neither accurate collective belief nor authorized collective action.
Read the full abstract ▾
Robotics
Accuracy Evaluation of INS/ZUPT Filtering Methods Based on Different Geometric Error Definitions
Wei Ouyang, Jiale Han, Yarong Luo et al. (+1 more)
Geometric filters have recently been introduced to improve the accuracy and consistency of inertial-based integrated
navigation systems. Error states were defined through specific group operations, introducing state correlations in error definition, which were lacked in the additive error used by a conventional indirect Kalman filter. The desirable consistent filtering models can be obtained based on specific
geometric errors. For zero-velocity measurements expressed in the reference frame, this paper derives left-error process and measurement models from invariant filtering, two-frame-group filtering, and equivariant filtering. Importantly, a new group operation is introduced for the left tangent-group equivariant error. The analysis shows that the two-frame-group invariant extended Kalman filter (TFG-IEKF) and the tangent-group equivariant filter (TG-EqF) do not offer a significant consistency advantage over the invariant extended Kalman filter (IEKF). Experiments with an INS/ZUPT measurement system show that, under small initial attitude errors, the conventional indirect extended Kalman filter (EKF) achieves loop-closure position errors below $0.1\%$ of the traveled distance, while the three
geometric filters achieve comparable positioning accuracy.
Read the full abstract ▾
All authors Wei Ouyang, Jiale Han, Yarong Luo, Maoran Zhu
Artificial Intelligence
Neuralyzing the Trace: Selective Representation-Level Unlearning with Contrastive Sparse Autoencoders
Itai Zehavi, Fanny Jourdan and Ulrich Aivodji
Machine unlearning aims to remove targeted information while preserving a model's other abilities. In realistic settings, such as
privacy requests under the EU GDPR, the target may be narrow, for example information associated with a single person. Behavioral forgetting alone may be insufficient, motivating interventions directly on internal representations. However, standard mechanistic-interpretability extractors are poorly selective for such targets. We identify an energy bias in reconstruction-based extraction, which favors dominant background structure over low-energy target-specific components. We introduce SCALPEL, a contrastive sparse autoencoder designed to learn more selective forget features. We show theoretically that contrastive training promotes target-selective features and that our selection score controls expected background knowledge perturbation. We validate SCALPEL experimentally on TOFU across Qwen, Llama, and Gemma, where it substantially improves over NMF and standard SAE interventions and is competitive with Gradient Difference and RMU, bridging mechanistic interpretability and fine-grained unlearning.
Read the full abstract ▾
Artificial Intelligence
Cheap, open agents make LLM pollution harder to mitigate
Raluca Rilla, Anne-Marie Nussberger, Rui Mata et al. (+1 more)
Large
Language Model (LLM) pollution occurs when synthetic responses contaminate data intended to capture human behavior. High deployment costs have so far limited the risk posed by autonomous survey
agents. However, open-weight models paired with open-source agentic frameworks may have removed this barrier. We compared the performance and detectability of nine
agent configurations, ranging from fully open variants to closed commercial ones. Each
agent autonomously completed a survey containing multiple response types yielding various detection checks. Fully open
agents ran locally without usage fees and performed competitively with commercial alternatives. Open and commercial
agents failed different sets of checks, and no single check reliably detected all
agents, but open-text responses discriminated best between
agents and humans. These findings identify fully open
agents as a distinct risk for LLM pollution and support multilayered detection strategies emphasizing open-text analysis.
Read the full abstract ▾
All authors Raluca Rilla, Anne-Marie Nussberger, Rui Mata, Dirk U. Wulff
Computational Complexity
Linear Certificates for Membership Comparability, Quadratic Barriers for Selectors
Sebastian Ben Daniel
Selectors and comparators supply only partial information about membership: a selector names a member of any pair that meets the
language, while a binary membership comparator merely excludes one of the four membership vectors of a pair. We ask how much nonuniform advice turns such information into exact recognition. Our main result extends the optimal nondeterministic advice bound for P-selective sets to every binary membership-comparable
language: $2-mc \subseteq NP /(3n+5)\cap\mathrm{coNP}/(3n+5)$, with common fixed advice and certificates of at most $5n+12$ bits. The class is strictly larger; some 2-mc sets are not truth-table reducible to any P-selective set. The proof replaces the tournament king by an independent two-step cover of true signed literals, together with a short-forcing-or-exact-majority dichotomy, and it relativizes. Via an advice-preserving isolation transfer, a deterministic
polynomial-time
algorithm for promise Unique-
Circuit-SAT gives $2-mc \subseteq P/O(n)$. For selectors we determine tight orders of ordinary advice: $\Theta(n)$ for errorless average-case computation and $\Theta(n^2)$ for worst-case bounded-error computation, the latter independent of the interpreter's coin bound. One oracle realizes both orders on a single
language and separates ordinary from coin-dependent advice. The quadratic and linear lower bounds hold for tournament-query procedures and relativized
languages, not unconditionally for unrelativized P-selective sets.
Read the full abstract ▾
Computer Vision
Where Compute Matters: Heterogeneous Attention for Efficient Video Diffusion
Olga Zatsarynna, Denis Korzhenkov, Juergen Gall et al. (+2 more)
Efficient
video generation requires reducing the quadratic cost of self-
attention over long spatio-temporal
token sequences. Existing efficient-
attention methods typically apply the same computation pattern to every
token, even though denoising difficulty varies substantially across
video regions and evolves throughout the generation process. We introduce HetA-DiT, a heterogeneous
attention mechanism that adaptively allocates computation according to
token difficulty. A lightweight uncertainty branch predicts a
token-wise estimate of denoising difficulty, which is used to route uncertain
tokens through dense global
attention while processing more reliable
tokens with efficient local
attention. The resulting routing is content- and timestep-adaptive, retains global context where it matters most, and provides a single parameter for controlling the quality-efficiency trade-off. HetA-DiT is compatible with few-step distribution-matching
distillation and introduces no additional
Transformer evaluation at inference time by reusing uncertainty estimates from the preceding denoising step. We evaluate the method on DMD-distilled Wan2.2-5B and Wan2.1-1.3B models. Across VBench, VBench-2.0, and human preference evaluation, HetA-DiT maintains competitive generation quality while routing only approximately 20% of
tokens through dense
attention.
Read the full abstract ▾
All authors Olga Zatsarynna, Denis Korzhenkov, Juergen Gall, Amir Habibian, Mohsen Ghafoorian
Robotics
Kintsugi-VLA: Turning Failed Robot Rollouts into Recovery Data through Interventional Recoverability
Ivan Snegirev, Elizaveta Semenyakina, Dmitrii Maliukov et al. (+2 more)
Simulation enables scalable training of
Vision-
Language-Action policies by using privileged experts to generate visual demonstrations without requiring every trajectory to be collected through manual teleoperation. However, such pipelines typically retain successful demonstrations while failed rollouts are discarded, even though they expose precisely the off-nominal states from which recovery must be learned. We introduce Kintsugi-VLA, a framework for converting failed rollouts into targeted synthetic recovery data by exploiting exact state restoration and branching in simulation. For a fixed privileged expert, we define interventional recoverability as the probability of completing the original task after the simulator is restored to a given state, estimate it using adaptive Monte Carlo continuations with pointwise Wilson intervals, and characterize its non-monotonic
evolution along failed trajectories. These estimates identify an observed terminal low-recoverability frontier-the point after which measured recoverability remains below a threshold-which is then used to select informative recovery starting states. In a simulated Franka manipulation task, targeted recovery data yield aggregate SmolVLA recovery success of 34.6% and 38.4% under difficulty- and frame-budget matching, respectively, 5.8 and 6.7 percentage points above uniform sampling within the same recovery window. The same ordering is observed under disturbed end-to-end execution and shifted clutter and physics conditions, while clean-task success decreases from 76.8% to 74.7%. Kintsugi-VLA demonstrates how failed simulator rollouts can be transformed from discarded experience into structured recovery-training data through direct interventional measurement.
Read the full abstract ▾
All authors Ivan Snegirev, Elizaveta Semenyakina, Dmitrii Maliukov, Miguel Altamirano Cabrera, Dzmitry Tsetserukou
Distributed Computing
DynBranch: Speculative Subgraph Reuse for Dynamic Agentic LLM Serving
Junyi Shen, Noppanat Wadlom, Zhengyuan Su et al. (+1 more)
Agentic LLM workflows decide their execution paths at runtime. Downstream computation may be predictable, or may have run before, yet it cannot begin until the model or the user resolves the branch. We call this serialization the branch-resolution barrier. Caching alone does not hide it: the key that identifies a reusable result is not known until then. In this paper, we propose DynBranch, which makes an unresolved branch addressable before it resolves. Its stable coordinate lets candidate subgraphs run during resolution and completed subgraph results be reused across later requests. A two-level controller admits this work when its expected benefit exceeds the load price. DynBranch sits at the model-API boundary and requires no changes to
agent harnesses or model execution engines. Across four agentic workloads with Qwen3-32B on 4x H200 GPUs, DynBranch reduces mean latency by up to 32% over each workload's strongest prior system and by 46-66% against a no-reuse floor, while preserving workflow results. The benefit persists across backbone families and on a commodity Qwen3-8B/RTX 4090 deployment.
Read the full abstract ▾
All authors Junyi Shen, Noppanat Wadlom, Zhengyuan Su, Yao Lu
Computation and Language
Modeling Student Sensemaking with LLMs and Knowledge-Graph-Guided Inference
Özge Alacam, Zübeyde Demet Kirbulut Güneş, Funda Ekici et al. (+7 more)
Collaborative science learning requires nuanced interpretation of student dialogue to characterize how learners identify knowledge gaps, build explanations, and work toward resolution - a theory-driven analysis that is labor-intensive and difficult to scale. We investigate whether instruction-tuned large
language models (LLMs) can support multidimensional analysis of collaborative sensemaking without task-specific training, and whether structured knowledge-state information improves model inference. We evaluate two mid-size LLMs on 23 richly annotated, expert-labeled episodes across prompting conditions that vary definitional scaffolding,
reasoning mode, and turn structure. Without
reasoning, models tend to overpredict successful sensemaking;
reasoning-enabled prompting improves identification of unsuccessful cases. Knowledge-state diagnostics provide additional grounding, improving detection of unsuccessful sensemaking and increasing agreement with expert annotations. No single configuration performs best across all sensemaking dimensions, underscoring the multidimensional nature of the task.
Read the full abstract ▾
All authors Özge Alacam, Zübeyde Demet Kirbulut Güneş, Funda Ekici, Nurcan Turan-Oluk, Dilay Dinçdemir, Hakk{\i} Kaday{\i}fç{\i}, Sevinç Nihal Yeşiloğlu, Burcu Iş{\i}k, Halil Tümay, Sinem Gencer
Information Retrieval
KuaFu: Compressing Long User Behavior into Understanding at Billion Scale
Jiahao Hui, Lin Zhu, Ben Tan et al. (+7 more)
Learning user preferences from long behavioral sequences is fundamental to large-scale advertising, recommender systems, and conversational
agents. In industrial systems, preference modeling is typically task-wise, with different downstream tasks independently processing the same user history to capture different aspects of preference. This repeated processing creates substantial redundancy at billion-user scale, especially as preferences evolve and representations require frequent updates. We present KuaFu, a unified user preference modeling system that replaces repeated user-level sequence encoding with reusable item-level compressed representations. Each behavioral item is encoded once and shared across users and downstream tasks, while a user's history is built chronologically and extended with new item representations as they arrive, letting KuaFu capture evolving preferences without re-encoding the entire history. To reduce inference and storage overhead, KuaFu introduces a two-axis projector:
token-axis compression shortens behavior sequences by up to 15x and reduces each item from 150 KB to 10 KB, while width-axis compression reduces the cached representation to 0.5 KB. We further develop a four-stage training pipeline consisting of reconstruction pre-training, preference-understanding post-training, joint compression-generation training, and
hallucination-aware reinforcement learning, progressively improving representation fidelity and downstream utility. Across six MRQA and five RecBench
benchmarks, KuaFu consistently outperforms existing compression methods at comparable ratios, with gains of up to 17.7 EM and 1.90 AUC points respectively. KuaFu has been deployed and evaluated on the Tencent advertising and recommendation platform for ten months. Over this period, it improves per-GPU throughput by up to 350%, saves 190 GPUs, and increases overall GMV by 1.37%.
Read the full abstract ▾
All authors Jiahao Hui, Lin Zhu, Ben Tan, Yishen Hu, Jingdong Shu, Zetai Jiang, Xining Ran, Yeshou Cai, Gong Chen, Haijie Gu
Numerical Analysis
Operator-ADI balanced truncation of delay differential systems
Timo Reis
We develop a balanced truncation method for linear delay differential systems based on a finite-rank alternating-direction implicit (ADI) iteration for the associated operator Lyapunov equations. The method acts directly on the infinite-dimensional history-state realization and does not discretize the delay interval during Gramian approximation or balancing. The finite-rank reachability and observability factors are represented exactly in Takenaka–Malmquist coordinates, reducing the required computations to finite-dimensional linear
algebra. These factors yield a finite-dimensional, delay-free reduced model. For real system data, the implementation remains in real arithmetic. Under suitable conditions on the ADI shifts and a gap in the Hankel singular values, we prove convergence in $\mathcal H_\infty$ to the reduced transfer function obtained by exact balanced truncation of the infinite-dimensional system. The construction extends to finitely many discrete delays. Numerical experiments validate the method against independent reference computations. The complete MATLAB implementation used in the experiments is provided as supplementary material.
Read the full abstract ▾
Sound · Cross-listed from eess.AS
Room Impulse Response Embeddings for Speech Enhancement in Noisy and Reverberant Environments
Adrian Meise and Reinhold Haeb-Umbach
We propose a self-supervised approach for learning room impulse response (RIR) representations from single-channel noisy-reverberant
speech. It consists of first training on reverberant data, then on noisy-reverberant data, and finally with a teacher-student approach, where the student learns to replicate the teacher's
embeddings when given a noisy version of the reverberant input. We assess their representational capabilities by estimating acoustic room parameters from them. Conditioning a discriminative
speech enhancement model on the derived
embeddings yields consistent gains across all evaluated metrics, including downstream word error rate, for both reverberant and noisy-reverberant
speech.
Read the full abstract ▾
Computer Vision
Exploiting Spatial Structure for Transductive Few-Shot Classification of Whole-Slide Images
Tiffanie Godelaine, Manon Dausort, Karim El Khoury et al. (+3 more)
Automating the analysis of whole-slide images (WSIs), a key step in cancer diagnosis, has high clinical value, as it can reduce pathologist's workload while improving diagnosis accuracy. Recently,
vision-
language models have shown promising performance for patch-level classification without requiring any annotation, yet these zero-shot (ZS) predictions remain noisy on fine-grained tasks and must be further refined. A promising direction is to refine all predictions jointly, i.e., a transductive approach. However, most existing methods are not tailored to WSIs. We thus propose SlideTIM, an adaptation to WSIs of the recent transductive approach LC-TIM, which introduces a combined spatial–latent regularizer together with a prior on the patch class distribution. The former enforces spatially and semantically close patches to receive the same predictions, while the prior calibrates the predicted class proportions. Together, they address the complex spatial organization and the strong class imbalance of WSIs. Evaluated on four histology
datasets, SlideTIM consistently outperforms all TIM variants, improving the macro-F1 by +8.1pp over the best competing baseline at 1 shot. Compared to the ZS, it raises the macro-F1 by +19.4pp at 1 shot. The code will be made available after submission.
Read the full abstract ▾
All authors Tiffanie Godelaine, Manon Dausort, Karim El Khoury, Benoît Gérin, Benoît Macq, Christophe De Vleeschouwer
Cryptography and Security
MetaPermit: Scalable and Auditable Access Control for AI Agents via LLM-Inferred Meta-Attributes
Hanzhang Ma, Ali Hariri, Tianxiang Shen et al. (+9 more)
The rise of autonomous AI
agents equipped with tools has introduced significant security risks, ranging from unintended tool misuse to
adversarial manipulation through Indirect
Prompt Injection (IPI) attacks. In practice, deployed
agent systems such as OpenAI Codex and Claude Code protect tool invocations through a combination of coarse-grained permission rules and LLM-based judgments about individual proposed actions. Both components, however, have important limitations: static policies must anticipate possible user intents and therefore do not scale to open-ended tasks, while LLM-driven authorization supports dynamic decisions but produces inconsistent outcomes and remains vulnerable to targeted IPI attacks. To provide scalable and more consistent authorization, we propose MetaPermit, a policy-based tool access-control framework that decouples semantic inference from security enforcement. By analyzing
agent-user interactions, we derive a compact, task-independent set of meta-attributes that capture the relationships among the user's intent, the execution context, and the proposed tool call. These meta-attributes allow MetaPermit to authorize tool use without enumerating user intents. At runtime, an LLM infers the meta-attribute values for each proposed tool call, while a fixed policy evaluates these values to allow or deny the call, making each decision auditable through the inferred values and the applied policy rule. We evaluate MetaPermit on the AgentDojo and AgentDyn
benchmarks, across seven task suites and five attack methods, using two widely deployed open-weight LLMs. The results show that MetaPermit produces 31% more consistent authorization decisions than LLM-driven authorization and outperforms the state-of-the-art defenses CaMeL and IPIGuard in both task completion, with improvements of up to 109%, and robustness to IPI attacks, with no malicious tool calls executed.
Read the full abstract ▾
All authors Hanzhang Ma, Ali Hariri, Tianxiang Shen, Bohua Zou, Qianjun Zheng, Ji Wang, Li Yi, Ning Jia, Yutao Liu, Haibo Chen, Lin Wang, Debayan Roy