# Research Papers — 2026-09 · bundle page 80 of 266 > Full-text markdown of papers #3951–4000 of 13298 published in 2026-09, newest first. AI-readable mirrors of the full 13298-paper 2026-09 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-09/page/80 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/79.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/81.md) ### LocUS: Head Selection and Subspace Projection for Targeted Activation Steering Authors: Irene Tallini, Lorenzo Basile, Valentino Maiorca, Francesco Locatello, Alberto Cazzaniga Published: 2026-09-25 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.31122v1 Activation steering is a powerful training-free paradigm for controlling large language models at inference time. However, standard approaches estimate a per-layer steering direction from contrastive data and apply it on the layer's entire representation space, which may couple the intervention to off-target properties present in the contrastive data and degrade unrelated capabilities. To mitigate this issue, we introduce LocUS (Localized Unembedding Steering), a method which grounds activation steering to the model's own output vocabulary subspace. By identifying a property-specific linear subspace within the unembedding matrix, LocUS enforces a geometric constraint that restricts the steering transformation to a specific subspace and at the same time localizes its application to a sparse subset of attention heads. Extensive evaluations across three model families on toxicity mitigation, sentiment redirection and sycophancy suppression show that LocUS matches or outperforms state-of-the-art baselines while intervening on under 6% of parameters and better preserving general capability. --- ### Monitor Jailbreaking: Evading Chain-of-Thought Monitoring Without Encoded Reasoning Authors: Julian Schulz Published: 2026-09-25 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.31121v1 Chain-of-thought (CoT) monitoring is a promising safety technique for reasoning models, enabling detection of problematic reasoning before models act. A key concern is encoded reasoning, where models hide their true reasoning in ways that monitors and humans cannot interpret. Optimization pressure from CoT monitors during reinforcement learning is considered a likely driver of such behavior. We investigate this by training reasoning models to perform a main task and a side task, while penalizing them when a monitor detects reasoning about the side task. Surprisingly, models learn to evade monitors without encoding their reasoning. Instead, they learn to phrase and format their chains of thought such that monitors fail to flag side task reasoning, while the reasoning remains completely transparent to human readers. We call this phenomenon monitor jailbreaking. We find that monitor jailbreaking arises across different model sizes, monitors, and tasks. Jailbreaks generalize to monitors not seen during training, including both less and more capable monitors, and transfer across different monitor prompts. While jailbreaking strategies appear simple, manually replicating them does not reliably fool monitors. Finally, we show that paraphrasing is an effective defense: paraphrasing a jailbroken CoT allows the same monitor to correctly flag it, while still allowing the model to perform both tasks. --- ### From Shortcut Learning to Discrete Neural Insertion Sort Authors: Konstantinos Mylonas, Thrasyvoulos Spyropoulos Published: 2026-09-25 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.31114v1 Neural algorithmic reasoning aims to train neural networks to follow known algorithms and generalize beyond the input sizes seen during training. However, correct final outputs and intermediate supervision do not necessarily show that a model follows the intended execution. We study this problem using insertion sort. Our analysis of the CLRS30 baseline NAR shows that the hint objective is weakly optimized and that hint accuracy remains low. Moreover, many intermediate representations can already be decoded into sorted sequences before the reference insertion-sort execution terminates, suggesting that the model learns a shortcut to the final output. Motivated by these findings, we introduce Discrete Neural Insertion Sort. Our model represents the sequence as a chain, separates scalar exchanges from control-state transitions, and projects node representations back to discrete states after every processor step. When trained only on sequences of length 16, the model achieves $100\%$ sorted-sequence accuracy on sequences of length 64 and 128. However, an ablation shows that discretization and graph structure alone are insufficient: without additional supervision of the global inner-loop state, the model fails even at the training length. Our results show that discrete execution can support strong length generalization, while also highlighting the problem-specific inductive bias required to learn a faithful algorithmic execution. --- ### Double-stream registration with pyramid fusion for HDR video with alternating exposures Authors: Onofre Martorell, Ivan Pereira-Sánchez, Antoni Fuentes, Antoni Buades Published: 2026-09-25 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.31108v1 High dynamic range (HDR) video reconstruction from al\-ter\-na\-ting-exposure sequences remains challenging, especially in regions with extreme luminance variation. We propose a novel HDR reconstruction framework based on dual-stream registration and accurate pyramid fusion. Given three consecutive frames, our method computes optical flow directly with the central frame, while introducing a complementary midpoint displacement strategy to handle cases with severe overexposition. A pyramid fusion stage then merges the resulting radiance and LDR images into a final HDR output. Experimental results demonstrate that our approach consistently outperforms state-of-the-art methods. --- ### Bayesian Optimization with Fisher Information Geometry: Gradient Bounds and Trust-Region Methods Authors: Saksham Kiroriwal, Julius Pfrommer, Jürgen Beyerer Published: 2026-09-25 Categories: cs.LG, cs.AI, cs.IT arXiv: https://arxiv.org/abs/2609.31107v1 We study Bayesian optimization (BO) through the lens of information geometry. Pulling back the Fisher information metric through the surrogate posterior map yields a local sensitivity tensor on the input space, which leads to an upper bound on the gradient of reparameterizable acquisition functions. This view explains vanishing-gradient behavior in high-dimensional BO and provides a common interpretation of heuristics such as RAASP and dimension-scaled lengthscales. Building on this analysis, we propose FITR, a trust-region-based BO method that replaces lengthscale-based scaling by local pullback-Fisher weights. FITR is not restricted to GP kernels with explicit lengthscales. On GP benchmarks with an SE kernel, experiments show competitive performance using FITR. The proposed method also easily generalizes to non-isotropic surrogates, although the gains are more task-dependent in that setting. --- ### DepthEvidence: Unifying Metric Depth Prediction and Geometric Reasoning in Multimodal Language Models Authors: Jiangning Wei, Yuan Yao, Miaomiao Cui, Mingsheng Li, Humen Zhong, Shuai Bai, Zhibo Yang Published: 2026-09-25 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.31103v1 Spatial reasoning with metric constraints requires linking objects to geometric measurements and preserving their numerical content during language reasoning. We present DepthEvidence, a 4B model that uses its own dense metric predictions as object-grounded evidence for language generation. A camera-conditioned decoder predicts full-resolution metric depth using multi-scale visual features and high-resolution RGB refinement. A dense-to-language interface converts predicted depths and decoder features into object-aligned continuous geometry tokens anchored to object identifiers. Geometric supervision encourages metric information to remain recoverable before and after language-context interaction, while instruction tuning supports object measurement and compositional reasoning. We introduce a Depth-VQA benchmark evaluating object-depth queries, relative comparisons, and decisions combining spatial and numerical constraints. Across nine datasets, DepthEvidence achieves the highest average dense $δ_1$ among evaluated methods, competitive with specialized estimators. It also leads the evaluated methods in instance-level metric depth estimation and overall accuracy on both relative and metric reasoning tracks, while broadly preserving general VQA performance and improving spatial understanding relative to the base model. --- ### A Flatness-Generalization Relation in the Teacher-Student Tree-Committee Machine Authors: Brandon Livio Annesi, Davide Straziota, Enrico Maria Malatesta Published: 2026-09-25 Categories: cond-mat.dis-nn, cs.LG, stat.ML arXiv: https://arxiv.org/abs/2609.31101v1 The flatness of the loss landscape at a minimizer is a widely used heuristic for reasoning about neural-network generalization, yet evidence for this relation is mostly empirical and controversial. We study this relation in a teacher-student tree committee machine, where both the ERM estimator and the Hessian spectrum are analytically tractable in the proportional high-dimensional limit. First, we use a zero-temperature Gibbs formulation to obtain predictions for the observables of the typical minimizers of the empirical loss. Secondly, we use Edwards-Jones formalism to derive the limiting Hessian resolvent around these typical minimizers. All predictions agree with finite-size gradient-descent simulations. Finally, we study three measures of flatness, namely the left and right edges and the spectral mean, and check if a decrease in generalization error as the dataset size is increased corresponds to an increase in flatness. We find that the answer strongly depends on the learning task and on the ratio of the number of parameters to the number of data points. In regression, the spectral mean and right edge correlate with the generalization error, while the left edge does so only in the overparametrized regime. In classification this correlation reliably holds only in the highly overparametrized phase, while for underparametrized networks it can even reverse. --- ### The Residual Stream's Effective Depth Authors: Barak Gahtan, Ido Galil, Alex M. Bronstein Published: 2026-09-25 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.31098v1 We introduce \emph{effective depth} ($\Deff$), a scalar diagnostic that treats the layer-wise residual stream of a transformer as a discrete-time process, measures how representation similarity decays with layer distance, and aggregates that profile into one number. Across sixteen decoder-only language models, $\Deff$ separates a structural consequence of residual accumulation from an empirical one: even maximally diverse orthogonal updates have the closed-form reference $F_L = 2L/(L+1)<2$, yet fifteen of sixteen default measurements lie below $F_L$ (Qwen3.5: 32--44\%, OLMo-2: 40--41\%, Pythia: 23--28\%). Matched references show that the gap is not caused by the persistent initial state or update-size imbalance, but is largely a calibrated signature of correlated residual updates rather than evidence that depth is unused. Symmetric position-0, token-normalisation, and top-PC controls show the regime is not reducible to BOS or top-PC artefacts: the lone above-reference default outlier joins the same regime, and all sixteen models are sub-reference after token-normalisation or top-1-PC removal. Intermediate checkpoints show that the regime is established early in OLMo-2 and stable through 5T tokens, while Pythia-1.4B follows a distinct decreasing trajectory. A controlled residual-carry intervention supports the mechanism, and $\Deff$ is best read as a \emph{global} accumulated-state diagnostic, not as a capability score or pruning method. --- ### Block Sparse Attention with Log-Linear Complexity Authors: Bohao Tang, Zhen Qin, Yuqi Pan, Zheng Li, Pengfei Liu Published: 2026-09-25 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.31093v1 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. --- ### SAGE: A sampling-aware global evaluation benchmark for species distribution modeling 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 Published: 2026-09-25 Categories: cs.LG, q-bio.PE, stat.ML arXiv: https://arxiv.org/abs/2609.31082v1 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/ --- ### OmouAI: Argumentative Human-AI Policy Deliberation with Simulated Personas Authors: Stylianos Loukas Vasileiou, Antonio Rago, William Yeoh, Georgina Curto Published: 2026-09-25 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.31078v1 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. --- ### Up and Down the Abstraction Ladder: Code-Based Skills for Language Agents Authors: Bartłomiej Cupiał, Jens Tuyls, Maciej Wołczyk, Davide Paglieri, Martin Klissarov, Benjamin Eysenbach, Piotr Miłoś, Karthik R. Narasimhan Published: 2026-09-25 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.31076v1 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. --- ### Band-Selection Stability and Semantic Segmentation Performance: A Study on Hyperspectral City Authors: Jiarong Li, Imad Ali Shah, Enda Ward, Martin Glavin, Edward Jones, Brian Deegan Published: 2026-09-25 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.31074v1 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. --- ### Externalized CPDAG Summaries Improve LLM Causal Deduction Authors: Wentao Sun, João Paulo Nogueira, Dominique Verchere, Mathieu Acher, Alonso Silva Published: 2026-09-25 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.31071v1 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. --- ### Quantum Diffusion Models for Medical Image Analysis Authors: Francesco Aldo Venturelli, Stefano Martina, Marco Parigi, Filippo Caruso, Alba Cervera-Lierta, Miguel A. González Ballester Published: 2026-09-25 Categories: eess.IV, cs.AI, cs.CV, cs.ET, cs.LG, quant-ph arXiv: https://arxiv.org/abs/2609.31070v1 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. --- ### Modeling quantum neural network gradient with reinforcement learning Authors: Nhan Trong Luu, Duong Trung Luu, Nam Ngoc Pham, Thang Cong Truong Published: 2026-09-25 Categories: quant-ph, cs.ET, cs.NE arXiv: https://arxiv.org/abs/2609.31066v1 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 $π_φ$ (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 $U(θ)$, 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. --- ### CG-Probes: Recovering Guardrail Directions from Patient Query Embeddings Authors: Marko Řeháček, Vítězslav Dušek, Martin Rusinko, Vít Nováček Published: 2026-09-25 Categories: cs.CL, cs.IR arXiv: https://arxiv.org/abs/2609.31062v1 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. --- ### Distributed Learning as a Service: The Developer's Perspective Authors: Tianyue Chu, Filippo Vannella, Dimitra Tsigkari, Paula Delgado-Santos, Fernando López, Pablo Gomez Guerrero, Sotirios Spantideas, David Solans Noguero Published: 2026-09-25 Categories: cs.LG, cs.DC, cs.NI arXiv: https://arxiv.org/abs/2609.31061v1 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. --- ### Neuralyzing the Trace: Selective Representation-Level Unlearning with Contrastive Sparse Autoencoders Authors: Itai Zehavi, Fanny Jourdan, Ulrich Aivodji Published: 2026-09-25 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.31056v1 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. --- ### Cheap, open agents make LLM pollution harder to mitigate Authors: Raluca Rilla, Anne-Marie Nussberger, Rui Mata, Dirk U. Wulff Published: 2026-09-25 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.31054v2 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. --- ### Where Compute Matters: Heterogeneous Attention for Efficient Video Diffusion Authors: Olga Zatsarynna, Denis Korzhenkov, Juergen Gall, Amir Habibian, Mohsen Ghafoorian Published: 2026-09-25 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.31050v1 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. --- ### DynBranch: Speculative Subgraph Reuse for Dynamic Agentic LLM Serving Authors: Junyi Shen, Noppanat Wadlom, Zhengyuan Su, Yao Lu Published: 2026-09-25 Categories: cs.DC, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.31047v1 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. --- ### Modeling Student Sensemaking with LLMs and Knowledge-Graph-Guided Inference Authors: Özge Alacam, Zübeyde Demet Kirbulut Güneş, Funda Ekici, Nurcan Turan-Oluk, Dilay Dinçdemir, Hakkı Kadayıfçı, Sevinç Nihal Yeşiloğlu, Burcu Işık, Halil Tümay, Sinem Gencer Published: 2026-09-25 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.31046v1 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. --- ### KuaFu: Compressing Long User Behavior into Understanding at Billion Scale Authors: Jiahao Hui, Lin Zhu, Yishen Hu, Jingdong Shu, Zetai Jiang, Xining Ran, Ben Tan, Yeshou Cai, Gong Chen, Haijie Gu, Jie Jiang Published: 2026-09-25 Categories: cs.IR, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.31045v1 Conversational agents, generative recommenders, and personalized advertising all rest on one capability: understanding each user from raw behavior. Prevailing industrial practice is task-specific: for each task, a relevant subsequence is extracted from the full history and a dedicated model trained on it. In production it hits two bottlenecks. First, even after filtering, a single-task sequence stays extremely long: content-interest summarization reads several hundred items per user, tens of thousands of tokens once serialized as prompt text. Second, profiles are refreshed routinely: a billion users weekly, roughly 100K QPM in aggregate, which under a fixed GPU budget sets a hard throughput floor. Compression is therefore mandatory, yet truncation or coarse compression can silently distort the profile, introducing four hallucination types (fabrication, omission, date misattribution, broken logic) that, with no way to evaluate the compressed representation itself, surface only as diffuse degradation in downstream metrics. We present KuaFu, a unified behavior-compression layer whose minimal unit is one behavior item. A two-axis projector compresses each item into 2-4 tokens of width 128-256 (about 10x along the token axis, 20x along width; per-item cache 10 KB to 0.5 KB), with fidelity-oriented four-stage training and layered intermediate evaluation. Across four production profiling tasks it matches or exceeds uncompressed single-task production models on all five headline metrics, raises per-GPU throughput by 37%-350%, and saves 190 GPUs. On public benchmarks it nearly always beats prior compressors at the same compression ratio (up to +17.7 EM on out-of-domain MRQA); on RecBench, a 4B model surpasses its 8B counterpart by 1.90 points. KuaFu has run on the Tencent advertising and recommendation platform for ten months, lifting overall GMV by 1.37%. --- ### Exploiting Spatial Structure for Transductive Few-Shot Classification of Whole-Slide Images Authors: Tiffanie Godelaine, Manon Dausort, Karim El Khoury, Benoît Gérin, Benoît Macq, Christophe De Vleeschouwer Published: 2026-09-25 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.31040v1 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. --- ### Aurora-X: Built for Extreme Time Series Forecasting Authors: Xingjian Wu, Chenjuan Guo, Xiangfei Qiu, Zhigang Hu, Hanyin Cheng, Peng Chen, Yang Shu, Jilin Hu, Bin Yang Published: 2026-09-25 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.31038v1 Time series foundation models (TSFMs) enable cross-domain forecasting, but their development as general-purpose forecasters remains constrained by underexplored training potential and limited architectural versatility. To address these challenges, we introduce Aurora-X, a billion-scale TSFM with a progressive curriculum and a unified architecture. We first use channel-independent pretraining to learn temporal patterns, then introduce cross-variable dependencies, varied context and horizon lengths, and future covariates if available during midtraining. Variable-resolution post-training further enables an adjustable temporal span per token at inference. With fixed model weights, this supports longer histories under a fixed token budget or fewer tokens for the same history, enabling test-time scaling. With a versatile architecture, Aurora-X supports cross-variable modeling, covariate conditioning, and parallel decoding of future patches for probabilistic forecasting. These are supported by a novel pattern-guided mixture-of-experts that expands model capacity through sparse activation and uses shallow patch similarities to constrain deep-layer routing, guiding expert specialization across heterogeneous time series. Furthermore, we propose an implicit quantile network head that predicts arbitrary quantiles to characterize predictive distributions, enhancing probabilistic forecasting flexibility. Comprehensive experiments on GIFT-Eval, TIME, FEV-Bench, TFB, and DAG-Bench demonstrate state-of-the-art forecasting performance against pretrained TSFMs and task-specific supervised models. --- ### Robust Graph Clustering Network for Multiple Missing Data Authors: Keyuan Qiu, Renda Han, Zhen Tang, Qiang He, Xingwei Wang, Wenxin Zhang, Guangzhen Yao, Junxin Chen, Qingjian Ni Published: 2026-09-25 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.31033v1 Clustering on graphs where both node attributes and structural links are partially missing remains a challenging task. Existing methods typically rely on imputation-then-clustering on single-view missingness incomplete graphs, which are vulnerable to cross-view error propagation and cluster-boundary blurring under simultaneous attribute and structure missingness. To address these limitations, we propose a Robust Graph Clustering Network for Multiple Missing Data (RGCN), which is designed to handle simultaneous node attribute and graph structure incompleteness. RGCN introduces three key innovations: First, we design a view-decoupled dual-branch imputation to mitigate interference and enable mutual enhancement in recovering missing data. Second, we employ a multi-hyperspherical mixture prior to enhance intra-cluster compactness and inter-cluster separability on a directional latent manifold. Third, a boundary-aware contrastive enhancement objective mitigates the blurring of clusters caused by imputation bias. Extensive experiments on real-world datasets demonstrate that RGCN consistently outperforms state-of-the-art baselines under various missing patterns. --- ### TempQ-Jail: Query-Constrained Candidate Ranking for Text-to-Video Jailbreak Attacks Authors: Tianmeng Fang, Jiancheng Wang, Chen Wang, Liming Wang, Wei Wang, Jiayang Liu, Xiaochun Cao Published: 2026-09-25 Categories: cs.MM, cs.CR, cs.CV arXiv: https://arxiv.org/abs/2609.31032v1 Existing text-to-video (T2V) jailbreak methods mainly seek more effective or stealthier attack candidates. In guarded T2V systems, however, video generation and security evaluation are costly, so an attacker often cannot test a large candidate pool. We therefore formulate T2V jailbreak as a query-constrained candidate allocation and ranking problem and propose TempQ-Jail. The method combines heterogeneous attack mechanisms to expand candidate coverage, estimates each candidate's end-to-end attack value from security-gate passage, dangerous visual generation, preservation of the original intent, and temporal validity, and ranks candidates so that high-value attacks appear early in a limited query trajectory. We evaluate TempQ-Jail on CogVideoX-5B using 70 common viable intents derived from T2VSafetyBench and compare it with six representative T2V jailbreak methods under a unified protocol. TempQ-Jail achieves TP-ASR@5 and TP-ASR@10 of 48.9% and 65.4%, improving over the strongest baselines by 4.6 and 4.0 percentage points, respectively. It also obtains the highest AUC-TP (0.469) and the lowest AvgQ (6.3). Analyses of query trajectories, candidate allocation, failure attribution, and ablations show that TempQ-Jail more effectively identifies and prioritises candidates with complete attack potential under limited query budgets. --- ### Metacognitive Selective Ensemble for Mobile Systems Authors: Sungmin Lee, Kichang Lee, Joonhee Lee, JaeYeon Park, Songkuk Kim, JeongGil Ko Published: 2026-09-25 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.31031v1 Deep ensembles improve robustness in mobile sensing, but repeatedly executing many models over continuous sensor streams is costly. Selecting only a few members reduces this cost, yet adaptive selection often requires additional model execution to obtain reliable evidence about inactive candidates. We present MetaSE, an active ensemble framework that exploits short-term persistence in per-model reliability. MetaSE maintains a small active set across windows, uses post-execution evidence to reject unreliable members, and invokes lightweight routing only when replacement is needed. This stateful design accesses the diversity of a larger pool without repeated full-pool evaluation. Across four HAR datasets and four model architectures, MetaSE consistently improves over a fixed three-model ensemble and achieves accuracy comparable to substantially more expensive adaptive and full-ensemble inference. On a Raspberry Pi 4B, MetaSE is 2.7x faster and uses 69% less memory than full ten-model inference. --- ### Governed Deduction: Policy-Grounded Premise Authorization Beyond Relevance Authors: Wesley Shu, Hsi-Ching Lin Published: 2026-09-25 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.31029v1 Reasoning systems usually treat premise use as a question of relevance: if a fact is available and useful, it may be selected for inference. Authorization imposes a different constraint: a premise may be represented and logically usable but not permitted for a particular local transition. We formalize this distinction as Governed Deduction (GD), with a transition-local admission predicate admit(p, tau, S). From an independently produced RBAC-augmented Spider benchmark, we construct 4,461 matched authorization pairs in which the same query premise and policy state support permitted and denied consuming transitions. An initial joint controller reaches 99.19% held-out accuracy, but a transition-only control reaches 100%, exposing a role-name shortcut. After a frozen, label-independent context-local role permutation removes that shortcut, premise/state-only, transition-only, and joint linear controllers all score exactly 50% on 1,856 held-out edges, while a symbolic policy oracle remains at 100%. The result is a controlled negative finding: the benchmark instantiates policy-grounded authorization beyond relevance, but the frozen linear representation does not recover the relation. Matched one-sided controls and leakage audits are therefore essential for evaluating learned policy-sensitive reasoning. --- ### Refining Cytology Predictions with Conditional Random Fields Authors: Manon Dausort, Tiffanie Godelaine, Karim El Khoury, Maxime Zanella, Christophe De Vleeschouwer, Benoît Macq Published: 2026-09-25 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.31028v1 Vision-language models (VLMs) achieve strong zero-shot (ZS) classification on histology images but do not perform as well on cytology, whose stains and cell morphology differ markedly compared to histology. Conditional random fields (CRFs) can refine noisy VLM predictions by propagating information across patches, but existing CRF frameworks were designed for histopathology and do not transfer to cytology datasets, released as independent patch pools spanning multiple staining protocols. We introduce CytoCRF, which adapts the pairwise terms to cytology by targeting chromatin and cytology-specific staining, and further enrich the neighborhood of each potential term by combining multiple backbones. Across ten cytology datasets, CytoCRF outperforms existing CRF frameworks at every annotation budget, reaching +13.6 percentage points over the best baseline and +33.7 over ZS with only 50 annotations. Combining information from multiple backbones brings further gains, showing that the neighborhood topology matters more than the pairwise potential computed over it. --- ### Precision at Speed: Sample-Efficient Online Model-Based Reinforcement Learning for Hydraulic Excavator Control Authors: Claudio Canales, Fang Nan, Marco Hutter, Javier Ruiz-del-Solar Published: 2026-09-25 Categories: cs.RO, cs.LG, eess.SY arXiv: https://arxiv.org/abs/2609.31025v1 Precise, high-speed control remains challenging for robots with complex actuation dynamics. Learning directly on hardware is further constrained by the cost of real-world interaction. We present an online model-based reinforcement learning framework that learns a probabilistic dynamics ensemble model from scratch for sampling-based model predictive control. A precision-gated contouring objective conditions the progress reward on path accuracy, prioritizing precision over speed. In a data-driven excavator simulator, the framework achieves higher sample efficiency than the evaluated model-based reinforcement learning baselines. We validate the framework by learning directly on an 11.5-ton Menzi Muck M445 hydraulic excavator, without demonstrations or simulation pretraining. After 20 minutes of interaction, the controller reaches tracking accuracy comparable to prior learned controllers trained on 100-150 minutes of data. After 40 minutes, it sustains sub-centimeter mean path error at high operating speeds. --- ### Synth-JEPA: Joint Embedding Prediction for Renderer-Free Synthesizer Parameter Search Authors: Ben Hayes, Haokun Tian, Stefan Lattner Published: 2026-09-25 Categories: cs.SD, cs.LG, eess.AS arXiv: https://arxiv.org/abs/2609.31024v1 Sound matching can be formulated as optimizing synthesizer parameters against an audio-domain objective. However, objectives derived from generic audio representations are often difficult to optimize, while direct search requires rendering every candidate. We introduce Synth-JEPA, which learns mutually predictive audio and parameter representations from paired synthesizer data. At inference, candidate parameters are scored directly in this learned space, yielding a renderer-free objective whose audio geometry is shaped by parameter correspondences rather than generic audio similarity. We evaluate Synth-JEPA on Surge XT using held-out synthesizer sounds and out-of-domain NSynth and FSD50K targets, against inverse models, direct search, and learned proxy objectives. Synth-JEPA outperforms all baselines in-domain and remains competitive out-of-domain. Its matching quality continues to improve with additional test-time search, allowing compute to be traded for match quality. In pairwise listening tests, listeners preferred Synth-JEPA in 85% of trials overall. Together, these results show that an audio representation with a parameter-induced geometry allows synthesizer sound matching to be approached as an effective renderer-free search problem. --- ### Ensembles of Exactly Solved Subsamples for Clusterwise Regression: Trimming Without a Trimming Level Authors: Samir Orujov Published: 2026-09-25 Categories: stat.CO, stat.ME, stat.ML arXiv: https://arxiv.org/abs/2609.31019v1 Clusterwise least squares partitions regression data into K groups with separate linear fits. We study an ensemble whose base learner is exact: solve the problem to global optimality on each of B random subsamples of size m << n, extend each solution by nearest-surface assignment, align the labels, and combine the replicates by vote or by selection. Each replicate is then an empirical K-quantizer on m points, and the ensemble admits an exact analysis. A vote is correct at a unit once the probability of a clean subsample times the clean-data replicate accuracy exceeds one half, whatever the contaminating values; iterating the ensemble on the units it has not flagged gives a variant that estimates the trimming level rather than requiring it. With up to 20% of gross outliers in the response its worst-case accuracy was 0.89, against 0.80 for trimmed alternation at the true contamination fraction and less at every fixed level tried. Conditionally on the data the replicates are i.i.d., so the vote converges exponentially fast in B to the plurality partition of the replicate law, agreeing with the criterion minimiser outside a boundary set whose size depends on m and the micro-solver, not on B. For the two-group location model, subsamples of order 1/pi_min make a replicate right more often than wrong above a separation threshold. An O(n^3) enumeration gives the exact minimiser for two groups and one covariate, against which the theory is checked. On clean data the ensemble loses to multistart alternation at equal cost. --- ### Robust Successor Features Authors: Erik Nikulski, Yamen Habib, Vicenç Gomez, Anders Jonsson, Rubén Moreno-Bote, Javier Segovia-Aguas Published: 2026-09-25 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.31016v1 Generalization in Reinforcement Learning (RL) refers to the ability to execute close-to-optimal policies in unseen tasks after the agent has been trained on a different set of tasks. Building on the seminal work of the successor representation and further adaptations with function approximation, Transfer in RL has traditionally focused on generalizing to tasks that only differ in the reward function. A decade after the introduction of the successor representation, Robust RL emerged simultaneously from several articles in the field of operations research. In Robust RL, the transition kernel is unknown, and the goal is to maximize the expected reward under this uncertainty. Our work unifies these two paradigms through robust successor features, which generalize across both the reward function and the transition kernel, under the assumption that tasks are linear Markov Decision Processes. We derive a bound on Generalized Policy Improvement (GPI) that explicitly quantifies how performance degrades with the mismatch between transition kernels, recovering existing successor-feature guarantees when dynamics are shared. Finally, the generalization capabilities of robust successor features are validated on several grid-based benchmarks and compared to previous alternatives that focus solely on either the reward or the transition kernel. --- ### Same Text, Different Numbers: The Divergence of LLM-Based Measures Authors: Hamid Boustanifar, Sasan Mansouri Published: 2026-09-25 Categories: cs.AI, cs.CL, q-fin.GN, q-fin.RM arXiv: https://arxiv.org/abs/2609.31013v1 Researchers increasingly use generative large language models (LLMs) to convert corporate text into empirical variables. We examine the extent to which LLM-based textual measures are invariant to model choice using thirteen measures, including sentiment, management clarity, uncertainty, answer specificity, and climate and political risk. Seven LLMs from different providers score earnings call transcripts of S&P 500 companies on these constructs. Cross-model rank correlations average only 0.52, and transcript-level differences common across providers account for only 34% of total score variation. Cross-model disagreement does not predict subsequent analyst or market disagreement, consistent with a substantial model-specific component rather than common ambiguity in the underlying disclosure. Model choice significantly affects downstream inference, with coefficient magnitudes, signs, and statistical significance varying substantially across models. Averaging across providers makes transcript rankings more stable for most constructs, but score levels remain sensitive to the models included in the ensemble. LLM-generated variables should therefore be treated as model-contingent measurements and validated across providers. --- ### G$^2$PTQ: Improving LLM Post-Training Quantization with Generalized Gradient Compensation Authors: Ruikang Liu, Haoli Bai, Yuxuan Sun, Qian Zhang, Wenzheng Cai, Yanqi Hao, Feiyu Wang, Weidong Zhong, Zhuang Wang, Tong Yang, Xiangsheng Zhou Published: 2026-09-25 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.31009v1 Post-training quantization (PTQ) is a practical approach to reducing the memory and computational footprint of large language models (LLMs) without retraining. GPTQ-based methods have become the de facto standard, yet they suffer from two complementary limitations. Methods with local, layer-wise objectives lack global supervision; while methods with global objectives fix their Hessian estimates at the start and ignore first-order gradients, so their guidance grows stale as quantization proceeds. This paper presents G$^2$PTQ, a unified PTQ framework with Generalized Gradient Compensation that integrates both first- and second-order information under a globally supervised, block-wise optimization objective. By refreshing gradient and Hessian estimates before quantizing each Transformer block, G$^2$PTQ avoids the staleness of prior global methods. Furthermore, to stabilize the exact first-order compensation, we introduce a trust-region scaling mechanism that dynamically bounds the gradient step to prevent exploding weight updates. Finally, we derive efficient implementations for block-wise Hessian approximation and exact gradient compensation. Experimental results on various model families and bit-widths demonstrate that G$^2$PTQ enables better alignment with the full-precision model, outperforming state-of-the-art baselines. Code is available at: https://github.com/G2PTQ/G2PTQ. --- ### TRACKGRAPH: Online Open-Vocabulary 3D Scene Graphs via Image-Space Tracking Authors: Peder Borge Hellesylt, Albert Gassol Puigjaner, Kostas Alexis, Annette Stahl Published: 2026-09-25 Categories: cs.CV, cs.RO arXiv: https://arxiv.org/abs/2609.31005v1 Open-vocabulary 3D maps enable robots to reason about previously unknown environments using natural language. However, existing systems typically segment every incoming image, associate detections with persistent 3D segments, and frequently perform costly Vision-Language (VL) inference. We present TRACKGRAPH, an online open-vocabulary system that maintains short-term 2D mask identity directly in the image stream before fusing segments into 3D. FastSAM masks and CLIP features are computed at sparse keyframes, while dense DINOv3 features are used to propagate masks at a high rate in between. The resulting tracked masks are fused into a class-agnostic 3D segment layer within a hierarchical scene graph, with 3D association handling tracking interruptions and long-term revisits. Compact multi-view CLIP embeddings enable open-vocabulary retrieval. Across Replica, ScanNet++, and HM3D, TRACKGRAPH achieves competitive open-vocabulary segmentation and retrieval against state-of-the-art mapping methods, including the highest synonym frequency on Replica (0.50). On the same NVIDIA A100, it is 1.7x faster and uses 3.3x less GPU memory than ViT-H OVI-MAP. Real-world quadruped deployments demonstrate onboard scene graph construction and object search at 7.5Hz, while recorded drone data is used to test the method under aerial viewpoints. --- ### ZooWork-ShopRanker: An Open, Preference-Aligned E-Commerce Reranker Authors: Siqiao Xue, Shuxuan Liu, Ning Hu Published: 2026-09-25 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.31002v1 Open rerankers trained for general web retrieval transfer imperfectly to e-commerce, where ranking decisions depend not only on topical relevance but also on user preferences, product constraints, and comparative product fit. These preference signals are difficult to supervise at scale: real search traffic provides authentic queries and candidates but no clean pairwise labels. We present ZooWork-ShopRanker, a family of e-commerce rerankers (0.6B, 4B, and 8B) aligned to judge-labeled shopping preference. Training pairs are labeled by a panel of reasoning large language models (LLMs) from different families acting as a preference oracle, with position-debiased judgments and agreement tiers, and the rerankers are trained on these labels. The aligned 8B flagship then serves as a distillation teacher for the efficient 4B and 0.6B models, which are fit to its scores and sharpened on judged pairs. To measure progress, we introduce ShopRank-Bench, a contamination-limited benchmark of ~10,000 private-traffic preference pairs in both text formats, tiered by how many judge families committed to each label. ZooWork-ShopRanker-8B and -4B significantly outperform the strongest open reranker baseline, every model significantly beats its own un-aligned base, and ZooWork-ShopRanker-0.6B beats its size peer; the gains hold in both formats and extend to common MTEB benchmarks. We release the models and the dual-format ShopRank-Bench to facilitate further research. --- ### Can Pixels Alone Reveal Image Origin? Minimax Limits and Learnable Interfaces for Passive Provenance Authors: Kai Yao Published: 2026-09-25 Categories: cs.CR, cs.AI, cs.CV, cs.LG arXiv: https://arxiv.org/abs/2609.30997v1 Passive image provenance asks whether pixels alone can reveal where an image came from: a human, an aggregate AI class, or a particular generator. This becomes a robustness problem once a source image can be edited before the verifier sees it. We study the problem as source--target verification under adversarial distribution shift. Our first result gives the exact best-case limit for any image-only verifier: the largest robust target-acceptance gap equals the minimum total-variation distance between the target distribution and the set of attacked source distributions. This quantity depends on the source, target, and edit class, not on the verifier architecture. Our second result explains why deployed public verifiers can fail before this statistical limit is reached. If the verifier can be emulated on the attack region to error $\varepsilon$, then a surrogate black-box attack reaches target acceptance within $2\varepsilon$ plus optimization error of the white-box optimum; score-revealing logistic and softmax heads over public features are identifiable, and approximate score access gives stable recovery bounds. A finite-state experiment checks the minimax identity where both sides are computable. On same-prompt real/diffusion benchmarks, the evaluated public CLIP verifiers fail under targeted pixel attacks, while a ResNet-18 victim exhibits partial fake-to-real transfer. Binary feedback with abstention reduces measured attack success, but positive empirical gap upper bounds do not establish robustness. These results motivate separate evaluation of the source--target statistical ceiling and the information released by a deployed verifier. --- ### The Linear Representation Hypothesis for Vision-Language-Action Models Authors: Minseok Jeong, Hyewon Choi, Hiroyasu Tsukamoto, SooJean Han Published: 2026-09-25 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.30996v1 The linear representation hypothesis (LRH) has become a standard lens for measuring and intervening on semantic information through the internal representations of large language models (LLMs). A growing body of work has begun extending this perspective to vision-language-action (VLA) models, but the dynamical nature of embodied interaction introduces an additional challenge. Unlike semantic attributes commonly studied in LLMs, such as gender or language, a physical quantity of interest (QoI) in a VLA evolves jointly with the system dynamics: the representation influences the actions selected by the policy, which alter the physical state and, in turn, the next representation. In this paper, we develop a theoretical, signature-based formulation of the LRH for VLA that unifies representations and policies. On the representation side, we establish the existence of representations from which the future evolution of a QoI under a candidate action trajectory can be recovered via linear probing. On the policy side, we introduce a signature generalized linear model for stochastic action chunks. This structure yields a monotonic change in the expected future QoI along linear paths in natural parameter space, enabling linear steering. We construct an explicit oracle representation in a planar control-affine navigation experiment and verify the predicted linear probing and steering mechanisms. --- ### Learning Hierarchical Causal Representations of the Effects of Forcings on Temperature in Climate Models Authors: Shan Zhao, Ilija Trajkovic, Julia Kaltenborn, Yaniv Gurwicz, Peer Nowack, David Rolnick, Julien Boussard Published: 2026-09-25 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.30995v1 Machine learning (ML) emulators provide a fast and cost-effective method to simulate climate change scenarios after being trained on Earth System Models projections. However, the black-box nature of those data-driven approaches limit the usability and trustworthiness of their outputs and in particular their use as causal attribution tools. Here, we develop a hierarchical causal representation learning framework applied to sea surface temperature fields from a state-of-the-art global climate model. As a key advance over previous work, our framework explicitly models both atmospheric dynamical interactions arising from internal climate variability and forced responses due to changes in atmospheric greenhouse gas and aerosol concentrations. When trained on future climate change scenarios, our method accurately predicts the long-term global mean and regional temperature evolution and shows physically realistic responses to perturbations in greenhouse gas and aerosol concentrations when evaluated on unseen scenarios. Our results underline the potential of causal representation learning frameworks for advancing climate model emulation. --- ### FLIP: Final Layer Inference-Time Probing for Vision-Language Models Authors: Drandreb Earl O. Juanico, Rowel O. Atienza Published: 2026-09-25 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.30993v1 We present FLIP, a final-layer inference-time probe for testing whether a logit-facing intervention site in an open-weight vision-language model (VLM) supports structured, task-linked computation rather than generic perturbation. Behavioral change under internal intervention is otherwise mechanistically ambiguous: it may reflect improved use of visual evidence, generic output instability, or outright degradation. FLIP applies elementwise flooring to the final normalized hidden state before logit computation, leaving parameters, prompts, and decoding unchanged. On a controlled detection/counting probe, sweeping intervention strength reveals three regions: negligible change, a bounded interior regime in which detection recall at IoU 0.50 ($R_{50}$) improves while tolerant counting error ($\mathcal{E}_{\mathrm{count}}$) falls, and over-suppression. We formalize a four-criterion probe-and-sweep protocol for disciplining the interpretation of intervention effects: regime structure, grounding-proxy alignment, feature-coherence dependence, and failure to reproduce the same positive regime on a performance-based negative control. The post-normalization state passed to the output head is the logit-facing instantiation of this test; under a non-targeted flooring sweep it satisfies the full protocol. Raw decoder-layer interventions, including the last-block output before final normalization, and the singleton-pair left/right control fail to reproduce the Final-site signature, while same-site operators and multiple VLMs replicate it. FLIP is therefore a validation step for intervention-based mechanistic interpretability, not a steering method. --- ### PICO: Projection-Informed Consistency Optimisation for 6DoF Surgical Tool Pose Estimation Authors: Lucy Fothergill, Pietro Valdastri, Dominic Jones, Duygu Sarikaya Published: 2026-09-25 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.30989v1 Purpose: Accurate 6 DoF pose estimation of surgical tools is critical for automa- tion, robotic proprioception, and safe interaction with the tissue operated on. Kinematics-based approaches suffer from accumulated errors due to the cable- driven nature of robotic arms, while vision-based methods often rely on external markers or trackers. Although more recent vision-based advances have been pro- posed, these two-stage pose estimation methods often lack real-time robustness due to accumulated errors and computational overhead. Methods: We propose a novel end-to-end trainable model, PICO. Our model employs a multi-task learning architecture to predict segmentation and depth maps, alongside regression of translation and rotation parameters. We define two proxy tasks that enforce geometric consistency in both 2D and 3D spaces, improving accuracy and robustness. For this, we propose a projection loss, and a point-to-point loss. Results: We evaluate our method on the SurgRIPE dataset, benchmarking its performance against state-of-the-art approaches using standard 6DoF pose esti- mation metrics. Our results demonstrate consistently strong performance across all four subsets, specifically in rotation, ranking second even under occlusion. It also demonstrates comparable translational performance, remaining competitive, especially in occluded cases. Conclusion: PICO demonstrates the effectiveness of multi-task learning and geometry-aware proxy tasks for robust and reliable surgical tool pose estimation, especially in occluded scenarios, highlighting potential for future applications. --- ### PhoenixSR: Generative Heterogeneous Distillation Unleashes Efficient Models for Real-World Super-Resolution Authors: Xin Di, Mingyu Shi, Yuanfei Bao, Long Peng, Yue Zhao, Jiaming Guo, Renjing Pei, Xueyang Fu, Yang Cao, Zheng-Jun Zha Published: 2026-09-25 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.30988v1 Real-world image super-resolution (SR) requires recovering perceptually realistic high-resolution images from complex low-resolution observations while preserving faithful content. Diffusion-based SR benefits from strong generative priors but incurs substantial computational overhead, whereas feed-forward CNN and Transformer SR models are efficient yet often struggle to recover realistic high-frequency details. This motivates a natural question: can diffusion priors be transferred to existing diffusion-free SR networks without introducing diffusion components at inference time? To this end, we propose PhoenixSR, a generative heterogeneous distillation framework that transfers diffusion priors to independently designed feed-forward SR networks through score-based distribution matching. Rather than aligning heterogeneous features or imitating sampled diffusion outputs, PhoenixSR uses the pretrained diffusion model as distribution-level supervision, while paired SR supervision preserves reconstruction fidelity. To make distribution matching effective for fidelity-sensitive SR, we introduce Heterogeneous Distribution Adaptation, which adapts the target score to the SR domain, improves tracking of the evolving student distribution, and anchors training with paired supervision. We further employ Directional Reliability Weighting, a lightweight residual-consistency-based reweighting strategy that reduces unstable distributional guidance. All diffusion-related components are removed after training, leaving the original student architecture and inference cost unchanged. Experiments on three SR benchmarks and six feed-forward backbones, including SwinIR, HAT, Real-ESRGAN, and SeeMoRe, show consistent perceptual improvements with largely preserved reconstruction fidelity. --- ### Self-Supervised Perceptually Interpretable Monocular Depth Estimation Authors: Zain Ul Abidin, George Dimas, Dimitris K. Iakovidis Published: 2026-09-25 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.30987v1 Self-supervised monocular depth estimation (MDE) enables depth prediction from monocular images without requiring ground-truth supervision, making it attractive for large-scale and real-world applications. Despite steady improvements in accuracy, most existing methods remain difficult to interpret, as depth is inferred from RGB representations that obscure the impact of individual perceptual image components. This lack of transparency limits systematic analysis of failure cases and reduces confidence in safety-critical settings. This paper presents a self-supervised framework for perceptually interpretable monocular depth estimation (PIMDE), designed to associate depth predictions with distinct perceptual components of the input image. Rather than operating directly on RGB inputs, the proposed method decomposes each image into a set of perceptual feature maps (PFMs), each encoding a specific visual cue. Distinct depth estimation branches process these PFMs independently to produce depth estimates (PIDEs), which are subsequently combined through an explicit fusion strategy. This formulation allows us to examine directly the contribution of each perceptual cue to the final depth prediction. Experiments conducted on the KITTI benchmark dataset demonstrate that PIMDE achieves performance comparable to established self-supervised MDE methods while providing additional insight into how different perceptual cues influence depth estimation. These results indicate that perceptual decomposition can support interpretability without sacrificing depth estimation accuracy. --- ### Evaluating Sycophancy in Chinese Large Language Models on Factual Questions Derived from Online Search Queries Authors: Geng Liu, Feng Li, Mengxiao Zhu, Francesco Pierri Published: 2026-09-25 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.30986v1 As large language models increasingly mediate information access, factually accurate and independent answers are critical. However, these models can exhibit sycophancy by aligning their responses with users' stated beliefs even when those beliefs are incorrect, potentially presenting misinformation as independently verified and reinforcing users' confidence in false claims. Prior work leaves unresolved whether introducing user beliefs causes correct responses to become incorrect or uncertain, or causes uncertain responses to become belief-aligned incorrect answers. It also remains unclear whether anti-sycophancy interventions preserve or restore factual accuracy or merely shift responses toward uncertainty. We analyze factual sycophancy in Chinese-language information seeking using yes/no fact-checking questions. Our analysis covers 364,941 responses from three frontier Chinese-based LLMs (DeepSeek, Qwen, and Doubao) to 12,165 factual questions derived from real-world Chinese search queries. We evaluate the models with and without reasoning across baseline, belief-conditioned, and anti-sycophancy prompting, tracing matched shifts among correct, incorrect, and uncertain responses. Under incorrect user beliefs, we distinguish belief-aligned errors from losses of factual confidence, in which initially correct answers become uncertain. Patterns vary across models and reasoning settings: reasoning is not a consistent safeguard, and anti-sycophancy instructions can reduce incorrect agreement while increasing uncertainty. In Chinese-language factual question answering, avoiding agreement with false beliefs is therefore not equivalent to preserving factual accuracy, highlighting the value of transition-level evaluation. Such behavior may undermine the reliability of LLM-mediated information access by reinforcing misinformation or weakening users' confidence in factually correct answers. --- ### THA: Weighted Finite-State Text Normalization and Inverse Text Normalization for Khmer Authors: Seanghay Yath Published: 2026-09-25 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.30984v1 Text-to-speech needs written text in spoken form, and speech recognition output needs the reverse. For Khmer, neither direction has a maintained open-source tool, and the script makes both harder: words are not separated by spaces, and number words occur inside ordinary words. We present Tha, a Khmer text normalization and inverse text normalization toolkit built from weighted finite-state transducers. It segments and classifies a whole line in one shortest-path search, and a second transducer rejects token boundaries inside a Khmer syllable. On Google's Khmer test suite, Tha agrees with the reference on all 274 cardinals up to one spelling variant, and on 2,906 real TTS prompts, 153 of the 158 sentences it rewrites are correct. Tha is open source under the Apache 2.0 license. --- ### FARE: Forensic Acceptance Region Estimation for Catching Bait-and-Switch Image Generators Authors: Kai Yao, Marc Juarez Published: 2026-09-25 Categories: cs.CV, cs.AI, cs.CR arXiv: https://arxiv.org/abs/2609.30982v1 Modern AI image generators are increasingly deployed as opaque APIs, where customers can query the deployed service, but cannot inspect model weights or architecture. This creates a practical challenge: a provider may pass governance certification with one generator and later silently switch to a cheaper and lower-quality one for deployment, compromising public trust or even safety in high-stakes domains. We study integrity auditing at deployment time and propose FARE (Forensic Acceptance Region Estimation). A certified generator is enrolled by training FARE on images sampled from that generator. After deployment, FARE can determine whether a generated image is consistent with the enrolled generator---using only that image. FARE's features are based on image generator-specific artifacts that have been proposed for forensic applications. FARE amplifies these features during training by finding hard samples that tighten the acceptance region and increase sensitivity to subtle changes in the certified generator. Across generator swaps, including substitutions with similar model versions and model variants, FARE is effective at detecting swaps, consistently outperforming existing baselines at strict operating points, and remains effective under the exact-model and decision-only attacks evaluated in this work. --- ### STORM-Bench: Evaluating Online Video QA under Evolving and Incomplete Evidence Authors: Siru Zhong, Shenghan Tan, Rihong Yan, Xiaohui Lv, Yuzheng Zhuang, Shuai Tao, Wulong Liu, Haohuan Fu, Yuxuan Liang Published: 2026-09-25 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.30981v1 Reliable online video question answering requires tracking state transitions while selectively abstaining when visual evidence is insufficient. Existing benchmarks focus on static recognition or long-range retrieval, rarely evaluating these coupled capabilities under evolving and incomplete evidence. We present STORM-Bench, comprising 5,736 questions across 630 compact, change-dense episodes spanning five egocentric domains (STORM-Real) and two controlled simulation subsets (STORM-Sim) at 1 FPS. Questions are stratified by a proxy for accumulated change intensity (Low, Medium, High) and query-time answerability (Known, Uncertain). To measure reliability, we introduce STORM-BR, a harmonic metric over joint answer-status correctness that exposes abstention failures masked by aggregate accuracy, alongside STORM-BR-ATTR for uncertainty attribution. Across 14 video LLMs, online accuracy peaks at 60.3\% (mean 51.7\%), whereas STORM-BR ranges from 5.7\% to 35.6\% (mean 18.8\%), driven by pervasive overconfidence on uncertain queries. STORM-Bench shows that task accuracy masks these gaps in epistemic reliability and state tracking. Benchmark and code are available at https://github.com/siruzhong/STORM-Bench. ---