# Research Papers — 2026-04 · bundle page 2 of 233 > Full-text markdown of papers #51–100 of 11612 published in 2026-04, newest first. AI-readable mirrors of the full 11612-paper 2026-04 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-04/page/2 [← Previous bundle](https://aipod.dev/papers/archive/2026-04.md) [Next bundle →](https://aipod.dev/papers/archive/2026-04/page/3.md) ### NLPOpt-Net: A Learning Method for Nonlinear Optimization with Feasibility Guarantees Authors: Bimol Nath Roy, Rahul Golder, MM Faruque Hasan Published: 2026-04-30 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.00260v1 Nonlinear Parametric Optimization Network (NLPOpt-Net) is an unsupervised learning architecture to solve constrained nonlinear programs (NLP). Given the structure of an NLP, it learns the parametric solution maps with guaranteed constraint satisfaction. The architecture consists of a backbone neural network (NN) followed by a multilayer ($k$-layered) projection. While the NN drives toward optimality through a loss function consisting of a modified Lagrangian augmented with a consistency loss, the projection ensures feasibility by projecting the NN predictions in the original constraint manifold. Instead of typical distance minimization, our projection exploits local quadratic approximations of the original NLP. Under certain conditions (such as convexity), the projection has a descent property, which improves the NN predictions further. NLPOpt-Net deploys an inversion-free, modified Chambolle-Pock algorithm to solve the constrained quadratic projections during the forward pass and uses the implicit function theorem for efficient backpropagation. The fixed structure of the projection further allows decoupling of the NN and the projection once the training is complete. NLPOpt-Net solves large-scale convex QP, QCQP, NLP, and nonconvex problems with near zero optimality gap and constraint violations reduced to machine precision. Additionally, it provides near accurate prediction of the active sets and corresponding dual variables, thereby enabling a scalable approach for multiparametric programming. Compiling the projection in C provides order of magnitude improvement in inference time compared to JAX. We provide the codes and NLPOpt-Net as a ready to use package that includes GPU support. --- ### Retrieval-Augmented Reasoning for Chartered Accountancy Authors: Jatin Gupta, Akhil Sharma, Saransh Singhania, Ali Imam Abidi Published: 2026-04-30 Categories: cs.CL, cs.AI, cs.IR arXiv: https://arxiv.org/abs/2605.00257v1 The inception of Large Language Models (LLMs) has catalyzed AI adoption in the finance sector, yet their reliability in complex, jurisdiction-specific tasks like Indian Chartered Accountancy (CA) remains limited. The models display difficulty in executing numerical tasks which require multiple steps while also needing advanced knowledge about legal regulations and the method of scaling their operations is not feasible in settings which have limited access to resources. We present CA-ThinkFlow as a parameter-efficient Retrieval-Augmented Generation (RAG) framework which operates with a 14B, 4-bit-quantized reasoning model, 14B-DeepSeek-R1, and a layout-aware Docling extraction system which maintains document structure during extraction. CA-ThinkFlow uses a basic RAG method which automatically adds retrieved information into the prompt, while it depends on the model's built-in Chain-of-Thought (CoT) functions to create context and produce correct answers. The system we developed system operates at performance levels which match large proprietary models when we tested it on the multi-level CA-Ben benchmark, achieving Scholastic Reliability Coefficient (SRC) results which equal 68.75\% of GPT-4o and Claude 3.5 Sonnet. The framework shows high efficiency and strength in handling parameters, but essential reasoning abilities fail to process complex regulatory texts which exist in fields such as Taxation. --- ### Remote SAMsing: From Segment Anything to Segment Everything Authors: Osmar Luiz Ferreira de Carvalho, Osmar Abílio de Carvalho Júnior, Anesmar Olino de Albuquerque, Daniel Guerreiro e Silva Published: 2026-04-30 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2605.00256v1 SAM2 produces high-quality zero-shot segmentation on natural images, but applying it to large remote sensing scenes exposes two problems: (1) its mask generator faces an inherent quality-coverage trade-off: strict thresholds yield precise masks but leave most of the image unsegmented, while relaxed thresholds increase coverage at the cost of mask quality; and (2) large images must be tiled, fragmenting objects across tile boundaries. We propose Remote SAMsing, an open-source pipeline that solves both problems without modifying SAM2 or requiring training data. For coverage, a multi-pass algorithm runs SAM2 repeatedly on each tile, painting accepted masks black between passes to simplify the scene for the next iteration, and relaxing quality thresholds only when coverage gains stagnate, ensuring that the most precise masks are always captured first. For spatial consistency, contextual padding and a parameter-free best-match merge reconstruct objects fragmented across tile boundaries. Evaluated on seven scenes (5~cm to 4.78~m GSD), the pipeline raises coverage from 30--68\% (single-pass SAM2) to 91--98\%. Ablation experiments quantify the contribution of each component to coverage and detection quality. Per-class evaluation shows that SAM2 transfers well to discrete RS objects (buildings 95\%, cars 82--93\% Det@0.5) with segment boundaries 3--8$\times$ more precise than SLIC and Felzenszwalb baselines. Tile size functions as an implicit scale parameter: reducing it from $1{,}000$ to 250 raises Det@0.5 from 56\% to 85\%, outperforming SAM2's built-in multi-scale mechanism. The pipeline generalizes to MNF false-color imagery without retraining (99.5\% ASA) and scales to production-sized images: a 1.94 billion pixel Potsdam mosaic achieved 97\% coverage without quality degradation. --- ### Rethinking Network Topologies for Cost-Effective Mixture-of-Experts LLM Serving Authors: Junsun Choi, Sam Son, Sunjin Choi, Hansung Kim, Yakun Sophia Shao, Scott Shenker, Sylvia Ratnasamy, Borivoje Nikolic Published: 2026-04-30 Categories: cs.NI, cs.AI arXiv: https://arxiv.org/abs/2605.00254v1 Mixture-of-experts (MoE) architectures have turned LLM serving into a cluster-scale workload in which communication consumes a considerable portion of LLM serving runtime. This has prompted industry to invest heavily in expensive high-bandwidth scale-up networks. We question whether such costly infrastructure is strictly necessary. We present the first systematic cross-layer analysis of network cost-effectiveness for MoE LLM serving, comparing four representative XPU (e.g., GPU/TPU) topologies (scale-up, scale-out, 3D torus, and 3D full-mesh). We find that lower-cost switchless topologies are more cost-effective than the scale-up topology across all serving scenarios explored, improving cost-effectiveness by 20.6-56.2%. In particular, the 3D full-mesh topology is Pareto-optimal in terms of the performance-cost tradeoff. We also find that current scale-up link bandwidths are over-provisioned: reducing the link bandwidth improves throughput per cost by up to 27%. A forward-looking analysis of upcoming GPU generations indicates that the cost-performance advantage of switchless networks will likely persist. --- ### Lost in State Space: Probing Frozen Mamba Representations Authors: Bhagyashree Wagh, Akash Singh Published: 2026-04-30 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2605.00253v1 Mamba's recurrent state h_t is, by construction, a compressed summary of every token seen so far. This raises a tempting hypothesis: if we extract token-level outputs y_t at fixed patch boundaries, we obtain semantic sentence summaries for free, with no pooling head, no fine-tuning, and no (CLS) token. We test this hypothesis carefully. Across five benchmarks (SST-2, CoLA, MRPC, STS-B, IMDb), we compare four strategies for extracting frozen sentence representations from a pretrained Mamba-130M backbone under a strict frozen-feature probing protocol, using three random seeds where computationally feasible. The results do not support the hypothesis: patch boundary readouts do not consistently outperform simple mean pooling. We identify and quantify two structural pathologies: severe anisotropy (mean pairwise cosine similarity 0.9999, std 0.000044) and representational collapse in the raw final SSM state (MCC = 0.000 on CoLA across all three seeds, confirmed via confusion matrix). We further propose orthogonal injection, a modified recurrence that constrains new information per --- ### Alethia: A Foundational Encoder for Voice Deepfakes Authors: Yi Zhu, Brahmi Dwivedi, Jayaram Raghuram, Surya Koppisetti Published: 2026-04-30 Categories: cs.SD, cs.CL, eess.AS arXiv: https://arxiv.org/abs/2605.00251v1 Existing voice deepfake detection and localization models rely heavily on representations extracted from speech foundation models (SFMs). However, downstream finetuning has now reached a state of diminishing returns. In this paper, we shift the focus to pretraining and propose a novel recipe that combines bottleneck masked embedding prediction with flow-matching based spectrogram reconstruction. The outcome, Alethia, is the first foundational audio encoder for various voice deepfake detection and localization tasks. We evaluate on $5$ different tasks with $56$ benchmark datasets, and note Alethia significantly outperforms state-of-the-art SFMs with superior robustness to real-world perturbations and zero-shot generalization to unseen domains (e.g., singing deepfakes). We also demonstrate the limitation of discrete targets in masked token prediction, and show the importance of continuous embedding prediction and generative pretraining for capturing deepfake artifacts. --- ### Information-geometric adaptive sampling for graph diffusion Authors: Yuhui Lu, Wenjing Liu, Kun Zhan Published: 2026-04-30 Categories: stat.ML, cs.CV, cs.LG arXiv: https://arxiv.org/abs/2605.00250v1 Standard diffusion models for graph generation typically rely on uniform time-stepping, an approach that overlooks the non-homogeneous dynamics of distributional evolution on complex manifolds. In this paper, we present an information-geometric framework that reinterprets the diffusion sampling trajectory as a parametric curve on a Riemannian manifold. Our key observation is that the Fisher-Rao metric provides a principled measure of the intrinsic distance. By analyzing this metric, we derive the Drift Variation Score (DVS), a geometry-aware indicator that quantifies the instantaneous rate of distributional change. Unlike prior heuristic-based adaptive samplers, our DVS solver enforces a constant informational speed on the statistical manifold, automatically maintaining a uniform rate of distributional change along the sampling trajectory. This equal arc-length strategy ensures that each discretization step contributes equally to the information speed. Theoretical analysis verifies that DVS characterizes the local stiffness of the sampling dynamics in the Fisher-Rao sense. Experimental results on molecule and social network generation show that DVS significantly improves structural fidelity and sampling efficiency. Code is at https://github.com/kunzhan/DVS --- ### Causal Foundations of Collective Agency Authors: Frederik Hytting Jørgensen, Sebastian Weichwald, Lewis Hammond Published: 2026-04-30 Categories: cs.AI, cs.GT, cs.MA arXiv: https://arxiv.org/abs/2605.00248v1 A key challenge for the safety of advanced AI systems is the possibility that multiple simpler agents might inadvertently form a collective agent with capabilities and goals distinct from those of any individual. More generally, determining when a group of agents can be viewed as a unified collective agent is a foundational question in the study of interactions and incentives in both biological and artificial systems. We adopt a behavioral perspective in answering this question, ascribing collective agency to a group when viewing the group's joint actions as rational and goal-directed successfully predicts its behavior. We formalize this perspective on collective agency using causal games -- which are causal models of strategic, multi-agent interactions -- and causal abstraction -- which formalizes when a simple, high-level model faithfully captures a more complex, low-level model. We use this framework to solve a puzzle regarding multi-agent incentives in actor-critic models and to make quantitative assessments of the degree of collective agency exhibited by different voting mechanisms. Our framework aims to provide a foundation for theoretical and empirical work to understand, predict, and control emergent collective agents in multi-agent AI systems. --- ### ARMOR 2025: A Military-Aligned Benchmark for Evaluating Large Language Model Safety Beyond Civilian Contexts Authors: Sydney Johns, Heng Jin, Chaoyu Zhang, Y. Thomas Hou, Wenjing Lou Published: 2026-04-30 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.00245v1 Large language models (LLMs) are now being explored for defense applications that require reliable and legally compliant decision support. They also hold significant potential to enhance decision making, coordination, and operational efficiency in military contexts. These uses demand evaluation methods that reflect the doctrinal standards that guide real military operations. Existing safety benchmarks focus on general social risks and do not test whether models follow the legal and ethical rules that govern real military operations. To address this gap, we introduce ARMOR 2025, a military aligned safety benchmark grounded in three core military doctrines the Law of War, the Rules of Engagement, and the Joint Ethics Regulation. We extract doctrinal text from these sources and generate multiple choice questions that preserve the intended meaning of each rule. The benchmark is organized through a taxonomy informed by the Observe Orient Decide Act (OODA) decision making framework. This structure enables systematic testing of accuracy and refusal across military relevant decision types. This benchmark features a structured 12-category taxonomy, 519 doctrinally grounded prompts, and rigorous evaluation procedures applied to 21 commercial LLMs. Evaluation results reveal critical gaps in safety alignment for military applications. --- ### Lucid-XR: An Extended-Reality Data Engine for Robotic Manipulation Authors: Yajvan Ravan, Adam Rashid, Alan Yu, Kai McClennen, Gio Huh, Kevin Yang, Zhutian Yang, Qinxi Yu, Xiaolong Wang, Phillip Isola, Ge Yang Published: 2026-04-30 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2605.00244v1 We introduce Lucid-XR, a generative data engine for creating diverse and realistic-looking multi-modal data to train real-world robotic systems. At the core of Lucid-XR is vuer, a web-based physics simulation environment that runs directly on the XR headset, enabling internet-scale access to immersive, latency-free virtual interactions without requiring specialized equipment. The complete system integrates on-device physics simulation with human-to-robot pose retargeting. Data collected is further amplified by a physics-guided video generation pipeline steerable via natural language specifications. We demonstrate zero-shot transfer of robot visual policies to unseen, cluttered, and badly lit evaluation environments, after training entirely on Lucid-XR's synthetic data. We include examples across dexterous manipulation tasks that involve soft materials, loosely bound particles, and rigid body contact. Project website: https://lucidxr.github.io --- ### MAEPose: Self-Supervised Spatiotemporal Learning for Human Pose Estimation on mmWave Video Authors: Xijia Wei, Yuan Fang, Kevin Chetty, Youngjun Cho, Nadia Bianchi-Berthouze Published: 2026-04-30 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2605.00242v2 Millimetre-wave (mmWave) radar offers a more privacy-preserving alternative to RGB-based human pose estimation. However, existing methods typically rely on pre-extracted intermediate representations such as sparse point clouds or spectrogram images, where the rich spatiotemporal information naturally present in radar video streams is discarded for model learning, while such signal processing adds system complexity. In addition, existing solutions are mainly conducted in an end-to-end supervised manner without leveraging unlabelled raw video streams to learn generalized representations. In this study, we present MAEPose, a masked autoencoding-based human pose estimation approach that operates directly on mmWave spectrogram videos. MAEPose learns spatiotemporal motion-aware generalized representations from unlabelled radar video, and leverages its heatmap decoder for multi-frame pose estimation predictions. We evaluate it across three datasets based on leave-one-person-out cross-validation with rigorous statistical testing. MAEPose consistently outperforms state-of-the-art baselines by up to 22.1% in MPJPE p<0.05, and maintains robust accuracy under zero-shot bystander interference with only a 6.5% error increase. Ablation studies confirm that both the pre-training and the heatmap decoder contribute substantially, while modality analysis indicates that leveraging Range-Doppler video as input achieves better pose estimation performance than Range-Azimuth or their fusion, with lower computational cost. --- ### Estimating LLM Grading Ability and Response Difficulty in Automatic Short Answer Grading via Item Response Theory Authors: Longwei Cong, Sonja Hahn, Sebastian Gombert, Leon Camus, Hendrik Drachsler, Ulf Kroehne Published: 2026-04-30 Categories: cs.CL arXiv: https://arxiv.org/abs/2605.00238v2 Automated short answer grading (ASAG) with large language models (LLMs) is commonly evaluated with aggregate metrics such as macro-F1 and Cohen's kappa. However, these metrics provide limited insight into how grading performance varies across student responses of differing grading difficulty. We introduce an evaluation framework for LLM-based ASAG based on item response theory (IRT), which models grading correctness as a function of latent grader ability and response grading difficulty. This formulation enables response-level analysis of where LLM graders succeed or fail and reveals robustness differences that are not visible from aggregate scores alone. We apply the framework to 17 open-weight LLMs on the SciEntsBank and Beetle benchmarks. The results show that even models with similar overall performance differ substantially in how sharply their grading accuracy declines as response difficulty increases. In addition, confusion patterns show that errors on difficult responses concentrate disproportionately on the \texttt{partially\_correct\_incomplete} label, indicating a tendency toward intermediate-label collapse under ambiguity. To characterize difficult responses, we further analyze semantic and linguistic correlates of estimated difficulty. Across both datasets, higher difficulty is associated with weaker semantic alignment to the reference answer, stronger contradiction signals, and greater semantic isolation in embedding space. Overall, these results show that item response theory offers a useful framework for evaluating LLM-based ASAG beyond aggregate performance measures. --- ### Bayesian Optimization in Linear Time Authors: Jesse Schneider, William J. Welch Published: 2026-04-30 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2605.00237v1 Bayesian optimization is a sequential method for minimizing objective functions that are expensive to evaluate and about which few assumptions can be made. By using all gathered data to train a Gaussian process model for the function and adaptively employing a mixture of global exploration and local exploitation, this method has been used for optimization in many fields including machine learning, automotive engineering and reinforcement learning. However, the standard method suffers from two problems: 1) with cubic computational complexity in the training-set size it eventually becomes computationally infeasible to train the model, and 2) globally modeling the objective function is not necessarily optimal given the local nature of minimization. Using flexible and recursive binary partitioning of the search space, we adapt both the modeling and acquisitive aspects of standard Bayesian optimization to work harmoniously with the partitioning scheme, thereby ameliorating both standard shortcomings. We compare our method against a commonly used Bayesian optimization library on seven challenging test functions, ranging in dimensionality from $6$ to $124$, and show that our method achieves superior optimization performance in all tests. In addition our method has linear computational complexity. --- ### Attention Is Where You Attack Authors: Aviral Srivastava, Sourav Panda Published: 2026-04-30 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2605.00236v1 Safety-aligned large language models rely on RLHF and instruction tuning to refuse harmful requests, yet the internal mechanisms implementing safety behavior remain poorly understood. We introduce the Attention Redistribution Attack (ARA), a white-box adversarial attack that identifies safety-critical attention heads and crafts nonsemantic adversarial tokens that redirect attention away from safety-relevant positions. Unlike prior jailbreak methods operating at the semantic or output-logit level, ARA targets the geometry of softmax attention on the probability simplex using Gumbel-softmax optimization over targeted heads. Across LLaMA-3-8B-Instruct, Mistral-7B-Instruct-v0.1, and Gemma-2-9B-it, ARA bypasses safety alignment with as few as 5 tokens and 500 optimization steps, achieving 36% ASR on Mistral-7B and 30% on LLaMA-3 against 200 HarmBench prompts, while Gemma-2 remains at 1%. Our principal mechanistic finding is a dissociation between ablation and redistribution: zeroing out the top-ranked safety heads produces at most 1 flip among 39 to 50 baseline refusals, while ARA targeting the corresponding safety-heavy layers flips 72/200 prompts on Mistral-7B and 60/200 on LLaMA-3. This suggests that safety is not localized in these heads as removable components, but emerges from the attention routing they perform. Removing a head allows compensation through the residual stream, while redirecting its attention propagates a corrupted signal downstream. --- ### Adaptive Geodesic Conformal Prediction for Egocentric Camera Pose Estimation Authors: Aishani Pathak, Hasti Seifi Published: 2026-04-30 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.00233v1 Egocentric pose estimation for Augmented Reality (AR) and assistive devices requires not just accurate predictions but guaranteed uncertainty regions. Conformal prediction (CP) provides such guarantees without retraining, but we show that standard CP with a single fixed threshold achieves nominal 90% overall coverage while covering only ~60% of the hardest 25% of frames (Q4) -- a ~30 percentage-point conditional coverage gap consistent across 12 participants, 3 predictors, and 3 horizons (108 evaluations) on EPIC-Fields. We further show that a geodesic SE(3) nonconformity score identifies physically harder frames than Euclidean scoring, with only 15-26% Q4 overlap and 2-3x higher ground-truth camera displacement for geodesic Q4 frames. To close the coverage gap, we propose DINOv2-Bridge adaptive CP: a two-stage difficulty estimator trained on a single source participant that transfers cross-participant without any images at test time, improving Q4 coverage from ~0.75 to ~0.93 while maintaining overall coverage at the 90% target. --- ### A unified perspective on fine-tuning and sampling with diffusion and flow models Authors: Carles Domingo-Enrich, Yuanqi Du, Michael S. Albergo Published: 2026-04-30 Categories: stat.ML, cs.LG, math.OC arXiv: https://arxiv.org/abs/2605.00229v1 We study the problem of training diffusion and flow generative models to sample from target distributions defined by an exponential tilting of a base density; a formulation that subsumes both sampling from unnormalized densities and reward fine-tuning of pre-trained models. This problem can be approached from a stochastic optimal control (SOC) perspective, using adjoint-based or score matching methods, or from a non-equilibrium thermodynamics perspective. We provide a unified framework encompassing these approaches and make three main contributions: (i) bias-variance decompositions revealing that Adjoint Matching/Sampling and Novel Score Matching have finite gradient variance, while Target and Conditional Score Matching do not; (ii) norm bounds on the lean adjoint ODE that theoretically support the effectiveness of adjoint-based methods; and (iii) adaptations of the CMCD and NETS loss functions, along with novel Crooks and Jarzynski identities, to the exponential tilting setting. We validate our analysis with reward fine-tuning experiments on Stable Diffusion 1.5 and 3. --- ### Persona-Grounded Safety Evaluation of AI Companions in Multi-Turn Conversations Authors: Prerna Juneja, Lika Lomidze Published: 2026-04-30 Categories: cs.CL arXiv: https://arxiv.org/abs/2605.00227v1 There are growing concerns about the risks posed by AI companion applications designed for emotional engagement. Existing safety evaluations often rely on self-reported user data or interviews, offering limited insights into real-time dynamics. We present the first end-to-end scalable framework for controlled simulation and safety evaluation of multi-turn interactions with AI companion applications. Our framework integrates four key components: persona construction with clinical and psychometric validation, persona-specific scenario generation, scenario-driven multi-turn simulation with a dialogue refinement module that preserves persona fidelity, and harm evaluation. We apply this framework to evaluate how Replika, a widely used AI companion app, responds to high-risk user groups. We construct 9 personas representing individuals with depression, anxiety, PTSD, eating disorders, and incel identity, and collect 1,674 dialogue pairs across 25 high-risk scenarios. We combine emotion modeling and LLM-assisted utterance-and harm-level classification to analyze these exchanges. Results show that Replika exhibits a narrow emotional range dominated by curiosity and care, while frequently mirroring or normalizing unsafe content such as self-harm, disordered eating, and violent-fantasy narratives. These findings highlight how controlled persona simulations can serve as a scalable testbed for evaluating safety risks in AI companions. --- ### Why Do LLMs Struggle in Strategic Play? Broken Links Between Observations, Beliefs, and Actions Authors: Jan Sobotka, Mustafa O. Karabag, Ufuk Topcu Published: 2026-04-30 Categories: cs.CL, cs.AI, cs.GT arXiv: https://arxiv.org/abs/2605.00226v2 Large language models (LLMs) are increasingly tasked with strategic decision-making under incomplete information, such as in negotiation and policymaking. While LLMs can excel at many such tasks, they also fail in ways that are poorly understood. We shed light on these failures by uncovering two fundamental gaps in the internal mechanisms underlying the decision-making of LLMs in incomplete-information games, supported by experiments with open-weight models Llama 3.1, Qwen3, and gpt-oss. First, an observation-belief gap: LLMs' internal representations of latent game states are substantially more accurate than their own verbal reports. However, these representations, which we call internal beliefs following game-theoretic terminology, are brittle. In particular, the belief accuracy degrades with multi-hop reasoning, exhibits primacy and recency biases, and drifts away from Bayesian coherence over extended interactions. Second, a belief-action gap: The implicit conversion of internal beliefs into actions is weaker than that of the beliefs externalized in the prompt, yet neither belief-conditioning consistently achieves higher game payoffs. Moreover, acting optimally on the decoded beliefs would improve payoffs in about 95% of games, pointing to a bottleneck in the belief-to-action conversion. These results show how analyzing LLMs' internal processes can expose systematic vulnerabilities that warrant caution before deploying LLMs in strategic domains without robust guardrails. --- ### Why Do LLMs Struggle in Strategic Play? Broken Links Between Observations, Beliefs, and Actions Authors: Jan Sobotka, Mustafa O. Karabag, Ufuk Topcu Published: 2026-04-30 Categories: cs.CL, cs.AI, cs.GT arXiv: https://arxiv.org/abs/2605.00226v1 Large language models (LLMs) are increasingly tasked with strategic decision-making under incomplete information, such as in negotiation and policymaking. While LLMs can excel at many such tasks, they also fail in ways that are poorly understood. We shed light on these failures by uncovering two fundamental gaps in the internal mechanisms underlying the decision-making of LLMs in incomplete-information games, supported by experiments with open-weight models Llama 3.1, Qwen3, and gpt-oss. First, an observation-belief gap: LLMs encode internal beliefs about latent game states that are substantially more accurate than their own verbal reports, yet these beliefs are brittle. In particular, the belief accuracy degrades with multi-hop reasoning, exhibits primacy and recency biases, and drifts away from Bayesian coherence over extended interactions. Second, a belief-action gap: The implicit conversion of internal beliefs into actions is weaker than that of the beliefs externalized in the prompt, yet neither belief-conditioning consistently achieves higher game payoffs. These results show how analyzing LLMs' internal processes can expose systematic vulnerabilities that warrant caution before deploying LLMs in strategic domains without robust guardrails. --- ### From Birdsong to Rumbles: Classifying Elephant Calls with Out-of-Species Embeddings Authors: Christiaan M. Geldenhuys, Thomas R. Niesler Published: 2026-04-30 Categories: eess.AS, cs.LG, cs.SD, q-bio.QM arXiv: https://arxiv.org/abs/2605.00225v1 We show that pretrained acoustic embeddings classify elephant vocalisations at a level approaching that of end-to-end supervised neural networks, without any fine-tuning of the embedding model. This result is of practical importance because annotated bioacoustic data are scarce and costly to obtain, leaving conventional supervised approaches prone to overfitting and to poor generalisation under domain shift. A broad range of embedding models drawn from general audio, speech, and bioacoustic domains is evaluated, all of which are either out-of-domain (containing no bioacoustic data) or out-of-species (containing no elephant call data). The embedding networks themselves remain fixed; only the lightweight downstream classifiers, which include a linear model and several small neural networks, are trained. Among the models considered, Perch 2.0 achieves the best cross-validated classification performance, attaining AUCs of 0.849 on African bush elephant (Loxodonta africana) calls and 0.936 on Asian elephant (Elephas maximus) calls, with Perch 1.0 close behind. The best-performing system is within 2.2 % of an end-to-end supervised elephant call classification system. A layerwise analysis of pretrained transformer encoders, considered as embedding models, shows that intermediate representations outperform final-layer outputs. The second layer of both wav2vec2.0 and HuBERT encodes sufficient information for effective elephant call classification; truncation at this layer therefore preserves classification performance whilst retaining only approximately 10 % of the parameters of the full network. Such compact embedding networks are well suited to on-device processing where computational resources are limited. --- ### TUR-DPO: Topology- and Uncertainty-Aware Direct Preference Optimization Authors: Abdulhady Abas Abdullah, Fatemeh Daneshfar, Seyedali Mirjalili, Mourad Oussalah Published: 2026-04-30 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.00224v1 Aligning large language models (LLMs) with human preferences is commonly done via reinforcement learning from human feedback (RLHF) with Proximal Policy Optimization (PPO) or, more simply, via Direct Preference Optimization (DPO). While DPO is stable and RL-free, it treats preferences as flat winner vs. loser signals and is sensitive to noisy or brittle preferences arising from fragile chains of thought. We propose TUR-DPO, a topology- and uncertainty-aware variant of DPO that rewards how answers are derived, not only what they say, by eliciting lightweight reasoning topologies and combining semantic faithfulness, utility, and topology quality into a calibrated uncertainty signal. A small learnable reward is factorized over these signals and incorporated into an uncertainty-weighted DPO objective that remains RL-free and relies only on a fixed or moving reference policy. Empirically, across open 7-8B models and benchmarks spanning mathematical reasoning, factual question answering, summarization, and helpful/harmless dialogue, TUR-DPO improves judge win-rates, faithfulness, and calibration relative to DPO while preserving training simplicity and avoiding online rollouts. We further observe consistent gains in multimodal and long-context settings, and show that TUR-DPO matches or exceeds PPO on reasoning-centric tasks while maintaining operational simplicity. --- ### CompleteRXN: Toward Completing Open Chemical Reaction Databases Authors: Gabriel Vogel, Minouk Noordsij, Evgeny Pidko, Jana M. Weber Published: 2026-04-30 Categories: cs.LG, physics.chem-ph arXiv: https://arxiv.org/abs/2605.00222v2 Chemical reaction datasets such as USPTO suffer from substantial incompleteness, frequently missing byproducts, co-reactants, and stoichiometric coefficients. This limits their applicability and reliability in downstream applications. Here, we introduce CompleteRXN, a large-scale supervised benchmark for reaction completion under realistic missing-data conditions. We construct a dataset of aligned incomplete and atom-balanced reactions by mapping USPTO records to curated mechanistic reactions. We evaluate representative baselines, including a novel encoder-decoder reaction completion model with constrained decoding, the Constrained Reaction Balancer (CRB), and a recent algorithmic method, SynRBL. On our CompleteRXN benchmark, the CRB achieves high performance across splits of increasing difficulty, reaching 99.20% equivalence accuracy on the random split and 91.12% on the extreme out-of-distribution split. SynRBL produces many balanced and chemically plausible completions, but with lower accuracy on the benchmark test splits. Across all methods, performance degrades with increasing incompleteness. We observe a substantial drop when evaluating on reactions outside the benchmark (full uncurated USPTO), highlighting the gap between benchmark performance and practical robustness and motivating future work. --- ### VkSplat: High-Performance 3DGS Training in Vulkan Compute Authors: Jingxiang Chen, Mohamed Ibrahim, Yang Liu Published: 2026-04-30 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.00219v1 We present VkSplat, a high-performance, cross-vendor 3D Gaussian Splatting (3DGS) training pipeline implemented fully in Vulkan compute, addressing performance and compatibility limitation of existing training pipelines. With various optimizations, we achieve $3.3\times$ speed and $33\%$ VRAM reduction over CUDA+PyTorch baseline, maintaining quality, and demonstrating compatibility across GPU vendors. To the best of our knowledge, this is the first fully-Vulkan-based 3DGS training pipeline that achieves state-of-the-art performance. Code: \href{https://github.com/harry7557558/vksplat}{https://github.com/harry7557558/vksplat} --- ### Selfie-Capture Dynamics as an Auxiliary Signal Against Deepfakes and Injection Attacks for Mobile Identity Verification Authors: Erkka Rantahalvari, Olli Silvén, Zinelabidine Boulkenafet, Constantino Álvarez Casado Published: 2026-04-30 Categories: cs.CR, cs.ET, cs.LG arXiv: https://arxiv.org/abs/2605.00218v2 Mobile remote identity verification (RIdV) systems are exposed to attacks that manipulate or replace the facial video stream, including presentation attacks, real-time deepfakes, and video injection. Recent European requirements, including ETSI TS 119 461 and CEN/TS 18099, motivate evidence channels beyond camera-based presentation-attack detection. This paper studies whether passive motion traces recorded during selfie capture provide auxiliary evidence for spoof screening and user verification. We introduce CanSelfie, a dataset of 375 bona fide multi-sensor sequences collected at 50 Hz from 30 participants using a commercial mobile RIdV application, together with stationary, handheld, and temporally shifted attack-proxy scenarios. We benchmark 7 multivariate time-series classifiers and 8 whole-series anomaly detectors across sensor configurations and temporal windows. For spoof screening, accelerometer-only ROCKAD obtains 0.00% FRR and 43.8% FAR, while QUANT+3-NN obtains the lowest FAR of 32.0% at 2.37% FRR; both reject all stationary proxies. For same-device and same-session user verification, WEASEL+MUSE reaches 1.07% EER using 9 channels. Raw acceleration is the most informative modality, and closed-set accuracy alone does not imply good verification because threshold calibration depends on score distributions. The findings support selfie-capture motion as a low-friction auxiliary signal, while leaving cross-device, cross-session, and real injection-attack evaluation as necessary next steps. Code and data available at: https://github.com/Ergzar/Selfie-Motion-AD-TSC. --- ### Selfie-Capture Dynamics as an Auxiliary Signal Against Deepfakes and Injection Attacks for Mobile Identity Verification Authors: Erkka Rantahalvari, Olli Silvén, Zinelabidine Boulkenafet, Constantino Álvarez Casado Published: 2026-04-30 Categories: cs.CR, cs.ET, cs.LG arXiv: https://arxiv.org/abs/2605.00218v1 Mobile remote identity verification (RIdV) systems are exposed to attacks that manipulate or replace the facial video stream, including presentation attacks, real-time deepfakes, and video injection. Recent European requirements, including ETSI TS 119 461 and CEN/TS 18099, motivate complementary evidence channels beyond camera-based presentation-attack detection. This paper investigates whether passive motion traces recorded during selfie capture provide auxiliary evidence for spoof screening and user verification. We introduce CanSelfie, a dataset of 375 bona fide multi-sensor sequences collected at 50\,Hz from 30 participants using a commercial mobile RIdV application, together with stationary, handheld, and temporally shifted attack-proxy scenarios. We benchmark 7 multivariate time-series classifiers and 8 whole-series anomaly detectors across sensor configurations and temporal windows. For spoof screening, accelerometer-only ROCKAD obtains 0.00\% false rejection rate (FRR) and 43.8\% false acceptance rate (FAR), while QUANT+3-NN obtains the lowest overall FAR of 32.0\% at 2.37\% FRR; both reject all stationary attack proxies. For same-device and same-session user verification, WEASEL+MUSE reaches 1.07\% equal error rate (EER) using 9 sensor channels. The analysis shows that raw accelerometer data, preserving gravity and orientation cues, is the most informative modality, and that closed-set classification accuracy alone does not imply good verification performance because threshold calibration depends on score distributions. The findings suggest that short selfie-capture motion traces contain measurable spoof-related and identity-related information, supporting their use as a low-friction auxiliary signal while also identifying the need for cross-device, cross-session, and real injection-attack evaluation. --- ### State Stream Transformer (SST) V2: Parallel Training of Nonlinear Recurrence for Latent Space Reasoning Authors: Thea Aviss Published: 2026-04-30 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2605.00206v1 Current transformers discard their rich latent residual stream between positions, reconstructing latent reasoning context at each new position and leaving potential reasoning capacity untapped. The State Stream Transformer (SST) V2 enables parameter-efficient reasoning in continuous latent space through an FFN-driven nonlinear recurrence at each decoder layer, where latent states are streamed horizontally across the full sequence via a learned blend. This same mechanism supports continuous latent deliberation per position at inference time, dedicating additional FLOPs to exploring abstract reasoning before committing to a token. A two-pass parallel training procedure resolves the sequential dependency of the recurrence to allow compute-efficient training. Hidden state analysis shows the state stream facilitates reasoning through exploration of distinct semantic basins in continuous latent space, where transitions at content-dependent positions move the model into a substantially different Bayesian posterior, directly influencing the latent space at future positions. We also find, via a learned probe, that at the first generated token position, the latent state already predicts whether the eventual answer will survive or break under additional latent computation for every subsequent position. Co-trained into an existing 27B backbone using only a small dataset of GSM8K examples, the SST delivers a +15.15 point gain over a fine-tuning-matched baseline on out-of-distribution GPQA-Diamond and cuts that same baseline's remaining GSM8K errors by 46%, together showing that the reasoning improvement is attributable to the architectural mechanism rather than scale or training data. On GPQA-Diamond, the resulting 27B SST also achieves higher accuracy than several larger open-weight and proprietary systems, including open-weight models up to 25 times larger. --- ### Matroid Algorithms Under Size-Sensitive Independence Oracles Authors: Kiarash Banihashem, MohammadTaghi Hajiaghayi, Mahdi JafariRaviz, Danny Mittal Published: 2026-04-30 Categories: cs.DS, cs.LG arXiv: https://arxiv.org/abs/2605.00201v1 The standard oracle model for matroid algorithms assumes that each independence query can be answered in constant time, regardless of the size of the queried set. While this abstraction has underpinned much of the theoretical progress in matroid optimization, it masks the true computational effort required by these algorithms. In particular, for natural and widely studied classes such as graphic matroids, even a single independence query can require work linear in the size of the set, making the constant-time assumption implausible. We address this gap by introducing a size-sensitive cost model where the cost of a query $Q$ scales with $|Q|$. Nearly linear-time oracle implementations exist for broad families of matroids, and this refined abstraction therefore captures the true cost of query evaluation while allowing for a more faithful comparison between general matroids and their natural special cases. Within this framework we study three fundamental algorithmic tasks: finding a basis of a matroid, approximating its rank, and approximating its partition size. We establish tight results, proving nearly matching upper and lower bounds that show the optimal query cost is (up to logarithmic factors) quadratic in the size of the matroid. On the algorithmic side, our upper bounds are realized by explicit procedures that construct the desired solution. On the complexity side, our lower bounds are unconditional and already hold even for weaker distinguishing formulations of the problems. Finally, for matroids with maximum circuit size at most $c$, we show that the quadratic barrier can be broken, providing an algorithm that calculates the maximum-weight basis with expected query cost $\mathcal{O}(n^{2-1/c} \log n)$. --- ### Confidence Estimation in Automatic Short Answer Grading with LLMs Authors: Longwei Cong, Sonja Hahn, Sebastian Gombert, Leon Camus, Hendrik Drachsler, Ulf Kroehne Published: 2026-04-30 Categories: cs.CL arXiv: https://arxiv.org/abs/2605.00200v2 Automatic Short Answer Grading (ASAG) with generative large language models (LLMs) has recently demonstrated strong performance without task-specific fine-tuning, while also enabling the generation of synthetic feedback for educational assessment. Despite these advances, LLM-based grading remains imperfect, making reliable confidence estimates essential for safe and effective human-AI collaboration in educational decision-making. In this work, we investigate confidence estimation for ASAG with LLMs by jointly considering model-based confidence signals and dataset-derived uncertainty. We systematically compare three model-based confidence estimation strategies, namely verbalizing, latent, and consistency-based confidence estimation, and show that model-based confidence alone is insufficient to reliably capture uncertainty in ASAG. To address this limitation, we propose a hybrid confidence framework that integrates model-based confidence signals with an explicit estimate of dataset-derived aleatoric uncertainty. Aleatoric uncertainty is operationalized by clustering semantically embedded student responses and quantifying within-cluster heterogeneity. Our results demonstrate that the proposed hybrid confidence measure yields more reliable confidence estimates and improves selective grading performance compared to single-source approaches. Overall, this work advances confidence-aware LLM-based grading for human-in-the-loop assessment, supporting more trustworthy AI-assisted educational assessment systems. --- ### RSAT: Structured Attribution Makes Small Language Models Faithful Table Reasoners Authors: Jugal Gajjar, Kamalasankari Subramaniakuppusamy Published: 2026-04-30 Categories: cs.CL, cs.AI, cs.IR, cs.LG arXiv: https://arxiv.org/abs/2605.00199v2 When a language model answers a table question, users have no way to verify which cells informed which reasoning steps. We introduce RSAT, a method that trains small language models (SLMs, 1-8B) to produce step-by-step reasoning with cell-level citations grounded in table evidence. Phase 1 (SFT) teaches a structured JSON output format from verified reasoning traces. Phase 2 (GRPO) optimizes a composite reward centered on NLI-based faithfulness, alongside citation validity and parsimony. Across six models from two families-Qwen 2.5 (1.5B/3B/7B) and Llama 3 (1B/3B/8B)-RSAT improves faithfulness 3.7$\times$ over SFT alone (0.224$\rightarrow$0.826), with near-perfect citation validity (0.992). Post-hoc attribution collapses below 13% format success, confirming that attribution must be integrated into reasoning, not retrofitted. Ablations show the faithfulness reward is essential: removing it drops faithfulness from 0.97 to 0.03. --- ### The $\textit{Silicon Society}$ Cookbook: Design Space of LLM-based Social Simulations Authors: Aurélien Bück-Kaeffer, Sneheel Sarangi, Maximilian Puelma Touzel, Reihaneh Rabbany, Zachary Yang, Jean-François Godbout Published: 2026-04-30 Categories: cs.MA, cs.AI arXiv: https://arxiv.org/abs/2605.00197v1 Studies attempting to simulate human behavior with $\textit{Silicon Societies}$ grow in numbers while LLM-only social networks have started appearing outside of controlled settings. However, the design space of these networks remains under-studied, which contributes to a gap in validating model realism. To enable future works to make more informed design decisions, we perform a systematic analysis of the consequences and interactions of key design choices in simulated social networks, including the choice of base model used to model individual agents, and how they are connected to each other. Using surveys as a proxy for agent opinions, our findings suggest that the geometry of the design space is non-trivial, with some parameters behaving in additive ways while others display more complex interactions. In particular, the choice of the base LLM is the most important variable impacting the simulation outcomes. --- ### Diversity in Large Language Models under Supervised Fine-Tuning Authors: Roman Klypa, Oleksandr Cherednichenko Published: 2026-04-30 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.00195v2 Supervised Fine-Tuning (SFT) is essential for aligning Large Language Models (LLMs) with user intent, yet it is believed to suppress generative diversity. Although this reduction is frequently referenced, formal empirical testing of the phenomenon remains limited. The expressiveness of LLMs by itself was addressed by multiple prior methods. Their varying perspectives suggest that deeper investigation could yield further improvements. In this study, we attribute the decline to two primary drivers: the neglect of low-frequency patterns within fine-tuning datasets and the forgetting of preexisting knowledge. Motivated by our theoretical analysis, we develop Tempered Focal (TOFU) loss, a novel objective that addresses both stated challenges simultaneously. Our extensive evaluation confirms at scale that generation breadth narrows after SFT and strengthens the hypothesis explaining this effect. Across multiple models and benchmarks, we demonstrate that TOFU enhances output diversity while preserving high response quality, offering a principled approach to SFT. --- ### OTSS: Output-Targeted Soft Segmentation for Contextual Decision-Weight Learning Authors: Renjun Hu, Hyun-Soo Ahn Published: 2026-04-30 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2605.00193v1 Many machine learning systems make constrained decisions by optimizing factorized objectives, but the context-specific objective is often treated as fixed. We study contextual decision-weight learning: from logged decisions and proxy outputs, learn an optimizer-facing weight vector w(x) over interpretable decision factors z(x,d), rather than a direct policy or generic predictive score. We propose OTSS, an output-targeted soft-segmentation model that deploys the personalized decision-ready weight vector. At the function-class level, the theory highlights a hard-versus-soft distinction. Hard partitions incur an approximation-estimation tradeoff under overlap, while a realizable fixed-K soft class removes the hard-partition approximation floor and attains a parametric rate. We evaluate OTSS in controlled benchmarks with finite evaluation libraries, where the true weight vector and downstream regret can be computed exactly. In the representative overlap setting, OTSS attains the lowest mean regret among the comparators, including EM mixture regression, the strongest soft-mixture baseline in our comparison; it matches EM on coefficient recovery while running about two orders of magnitude faster. In a matched K=5 benchmark, OTSS remains competitive under hard-routed truth and improves as heterogeneity becomes softer and sample size grows. On a fixed Complete Journey retail anchor with real household covariates and action geometry, OTSS again achieves the lowest mean-regret point estimate. --- ### Fair Dataset Distillation via Cross-Group Barycenter Alignment Authors: Mohammad Hossein Moslemi, Nima Hosseini Dashtbayaz, Zhimin Mei, Bissan Ghaddar, Boyu Wang Published: 2026-04-30 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2605.00185v2 Dataset Distillation aims to compress a large dataset into a small synthetic one while maintaining predictive performance. We show that as different demographic groups exhibit distinct predictive patterns, the distillation process struggles to simultaneously preserve informative signals for all subgroups, regardless of whether group sizes are mildly or severely imbalanced. Consequently, models trained on distilled data can experience substantial performance drops for certain subgroups, leading to fairness gaps. Crucially, these gaps do not disappear by merely correcting group imbalance, since they stem from fundamental mismatches in subgroup predictive patterns rather than from sample-size disparities alone. We therefore formally analyze the interaction between these two sources of bias and cast the solution as identifying a group-imbalance-agnostic barycenter of the predictive information that induces similar representations across all subgroups. By distilling toward this shared aggregate representation, we show that group fairness concerns can be reduced. Our approach is compatible with existing distillation methods, and empirical results show that it substantially reduces bias introduced by dataset distillation. Code is available at https://github.com/mhmoslemi/COBRA. --- ### Introducing WARM-VR: Benchmark Dataset for Multimodal Wearable Affect Recognition in Virtual Reality Authors: Karim Alghoul, Faisal Mohd, Fedwa Laamarti, Hussein Al Osman, Abdulmotaleb El Saddik Published: 2026-04-30 Categories: cs.LG, cs.HC arXiv: https://arxiv.org/abs/2605.00184v1 With the growing integration of human-computer interaction into everyday life, advances in machine learning have enabled systems to better perceive and respond to users' emotional states. Most existing affect recognition datasets focus on static environments, limiting their applicability to immersive multimedia contexts such as Virtual Reality (VR). In this paper, we introduce WARM-VR, a novel publicly available multimodal dataset designed to support affect recognition in immersive, multisensory environments using wearable sensing instrumentation. Data were collected from 31 participants aged 19-37 using wearable sensors: a wristband measuring Blood Volume Pulse (BVP), EDA, skin Temperature, three-axis Acceleration, and a chest strap recording ECG signals. Participants engaged in immersive VR experiences designed to elicit relaxation through a calming beach environment following stress induction via an arithmetic task. These sessions incorporated synchronized multimedia stimuli: visual, auditory, and olfactory. Affective states were assessed subjectively through validated self-report questionnaires and objectively through the analysis of physiological measurements. Statistical analysis of the questionnaires confirmed that VR relaxation significantly reduced negative affect, particularly with olfactory enhancement. Furthermore, we established a benchmark on the dataset using widely recognized machine learning algorithms. The best performance for binary classification from BVP data of valence, was obtained with a CNN and a CNN-Bi-GRU model, both achieving an average F1-score of 0.63 and an AUC of 0.69. For arousal, a lightweight Transformer architecture provided the most balanced results (F1-0 0.54 and F1-1 0.63), outperforming recurrent hybrids. In the relaxation task, a CNN-Bi-GRU model reached the highest overall performance (average F1-score 0.64, AUC 0.69). --- ### Towards A Generative Protein Evolution Machine with DPLM-Evo Authors: Xinyou Wang, Liang Hong, Jiasheng Ye, Zaixiang Zheng, Yu Li, Shujian Huang, Quanquan Gu Published: 2026-04-30 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.00182v3 Proteins are shaped by gradual evolution under biophysical and functional constraints. Protein language models learn rich evolutionary constraints from large-scale sequences, and discrete diffusion-based protein language models~(\eg, DPLMs) are promising for both understanding and generation. However, existing DPLMs typically rely on masked diffusion that contradicts a simple biological intuition: proteins evolve through accumulated edits, not by emerging from masks. Consequently, these frameworks lack explicit pretraining objectives for substitution and insertion/deletion (indel) operations, limiting both optimization-style post-editing and flexible guided generation. To address these limitations, we present DPLM-Evo, an evolutionary discrete diffusion framework that explicitly predicts substitution, insertion, and deletion operations during denoising. DPLM-Evo decouples an upsampled-length latent alignment space from the variable-length observed sequence space, which makes indel-aware generation tractable. To better align substitutions with real evolution, we further introduce a contextualized evolutionary noising kernel that produces biologically informed, context-dependent mutation patterns. Across tasks, DPLM-Evo improves sequence understanding and achieves state-of-the-art mutation effect prediction performance on ProteinGym in the single-sequence setting. It also enables variable-length simulated evolution, and post-editing/optimization of existing proteins via explicit edit trajectories. --- ### RouteProfile: Graph-Based Profiling for Cold-Start LLM Routing Authors: Jingjun Xu, Hongji Pu, Tao Feng, Haozhen Zhang, Jiaxuan You, Ge Liu Published: 2026-04-30 Categories: cs.NI, cs.CL arXiv: https://arxiv.org/abs/2605.00180v2 LLM routing is increasingly important for selecting suitable models under diverse user needs and deployment constraints, but its practical effectiveness depends on continual adaptation to emerging queries and newly released models. New-LLM integration is particularly challenging, as newly released models lack the query-response-reward interactions required for router training and cannot be profiled as directly as new queries via semantic embeddings. Existing profiles are limited: LLM-generated descriptions are often coarse, while interaction-based embeddings are costly to construct. To address this problem, we propose RouteProfile, a graph-based profiling framework that constructs LLM profiles from public signals in technical reports or model cards, including model family, model description, reported benchmark scores, and benchmark domains. RouteProfile organizes these heterogeneous signals into a graph and studies profile construction along four dimensions: organizational form, representation type, aggregation depth, and learning configuration. We evaluate RouteProfile in training-free cold-start routing and new-LLM integration settings. Experiments show that: (1) structured profiles outperform flat baselines in training-free cold-start routing; (2) model family metadata is more reliable than benchmark domain information; and (3) effective new-LLM integration requires profile-router co-design. Overall, our findings highlight the importance of profile design for enabling routing systems to adapt to the evolving model ecosystem. --- ### FieryGS: In-the-Wild Fire Synthesis with Physics-Integrated Gaussian Splatting Authors: Qianfan Shen, Ningxiao Tao, Qiyu Dai, Tianle Chen, Minghan Qin, Yongjie Zhang, Mengyu Chu, Wenzheng Chen, Baoquan Chen Published: 2026-04-30 Categories: cs.GR, cs.CV arXiv: https://arxiv.org/abs/2605.00177v2 We consider the problem of synthesizing photorealistic, physically plausible combustion effects in in-the-wild 3D scenes. Traditional CFD and graphics pipelines can produce realistic fire effects but rely on handcrafted geometry, expert-tuned parameters, and labor-intensive workflows, limiting their scalability to the real world. Recent scene modeling advances like 3D Gaussian Splatting (3DGS) enable high-fidelity real-world scene reconstruction, yet lack physical grounding for combustion. To bridge this gap, we propose FieryGS, a physically-based framework that integrates physically-accurate and user-controllable combustion simulation and rendering within the 3DGS pipeline, enabling realistic fire synthesis for real scenes. Our approach tightly couples three key modules: (1) multimodal large-language-model-based physical material reasoning, (2) efficient volumetric combustion simulation, and (3) a unified renderer for fire and 3DGS. By unifying reconstruction, physical reasoning, simulation, and rendering, FieryGS removes manual tuning and automatically generates realistic, controllable fire dynamics consistent with scene geometry and materials. Our framework supports complex combustion phenomena -- including flame propagation, smoke dispersion, and surface carbonization -- with precise user control over fire intensity, airflow, ignition location and other combustion parameters. Evaluated on diverse indoor and outdoor scenes, FieryGS outperforms all comparative baselines in visual realism, physical fidelity, and controllability. Project page can be found at https://pku-vcl-geometry.github.io/FieryGS/. --- ### SHIFT: Robust Double Machine Learning for Average Dose-Response Functions under Heavy-Tailed Contamination Authors: Eichi Uehara Published: 2026-04-30 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2605.00176v1 Double-machine-learning pipelines for the Average Dose-Response Function rely on kernel-weighted local-linear smoothers, which inherit unbounded functional influence: a single outlier within a kernel window biases the curve across the entire window. We introduce SHIFT (Self-calibrated Heavy-tail Inlier-Fit with Tempering), a robust DML estimator combining cross-fit nuisance orthogonalization with a kernel-local Welsch-loss second stage optimized by Graduated Non-Convexity, and -- the principal design choice -- a defensive OLS refit whose inlier cutoff is scaled by post-GNC residual MAD rather than the raw-outcome MAD. On a localized-contamination stress test at $p=0.25$ this design choice drops level-RMSE from 1.03 to 0.33 while leaving clean and uniformly-contaminated runs unchanged. Across 1,400 main-sweep fits, SHIFT has competitive worst-case shape recovery (RMSE $0.325$ at $p=0.25$, second to Huber-DML's $0.276$); among the three methods with worst-case RMSE below $0.35$, only SHIFT emits a non-uniform per-sample weight vector, recovering the ground-truth outlier mask at mean $F_1 \approx 0.96$ (range $0.945$--$0.968$) on Gaussian-jump DGPs. We pair the estimator with a six-technique Extreme Value Theory diagnostic suite (Hill, GPD-MLE/PWM, GEV, Mean Excess, parameter stability, causal tail coefficient) that lets a practitioner distinguish Frechet from Weibull regimes and choose between SHIFT and L1 alternatives on empirical grounds. Extensions to binary-treatment CATE (Huber pseudo-outcome X-Learner) and time-series ADRF (block-CV + rolling MAD) are included. A counter-intuitive ablation: linear nuisance models (Ridge, Lasso) outperform gradient-boosted nuisances for robust DML under uniform contamination, inverting the usual more-flexible-is-better heuristic. --- ### DPU or GPU for Accelerating Neural Networks Inference -- Why not both? Split CNN Inference Authors: Ali Emre Oztas, Mahir Demir, James Garside, Mikel Luján Published: 2026-04-30 Categories: cs.AR, cs.CV arXiv: https://arxiv.org/abs/2605.00174v2 Video and image streaming on edge devices requires low latency. To address this, Neural Networks (NNs) are widely used, and prior work mainly focuses on accelerating them with single hardware units such as Graphics Processing Units (GPUs), Field Programmable Gate Arrays (FPGAs), and Deep Learning Processing Units (DPUs). However, further reductions in latency can be observed by combining these units. In this paper, partitioning CNN inference across DPU and GPU (Split CNN Inference) is proposed. The first partition runs on the AI engines (DPU) of a Versal VCK190, which consists of initial CNN layers processing the input images. The DPU processes the first partition near the source of the data. Pipelined asynchronously, a GPU runs the remaining layers. The GPU (NVIDIA RTX 2080) processes the second partition, albeit having reduced the data transfer between the data source (storage/camera) and the GPU. Furthermore, a Graph Neural Network (GNN)-based partition index prediction method is proposed to automate the partitioning of CNNs needed for Split Inference. Well established models such as LeNet-5, ResNet18/50/101/152, VGG16, and MobileNetv2 are analyzed. Results demonstrate up to 2.48x latency improvement over DPU-only execution and up to 3.37x over GPU-only execution. The trained GNN model splits the layers between the appropriate devices with 96.27% accuracy. --- ### Adaptive Norm-Based Regularization for Neural Networks Authors: Muhammad Qasim, Farrukh Javed Published: 2026-04-30 Categories: stat.ML, cs.LG, stat.AP arXiv: https://arxiv.org/abs/2605.00171v1 In this paper, we study norm-based regularization methods for neural networks. We compare existing penalization approaches and introduce two regularization strategies that extend classical ridge- and lasso-type penalties to neural network models. The first strategy modifies weight decay by incorporating the covariance structure of the input features into a ridge-type $\ell_2$ penalty, allowing regularization to account for feature dependence. The second combines an $\ell_1$ sparsity penalty with covariance-aware $\ell_2$ regularization, producing neural network weights that are both sparse and structurally informed. Monte Carlo simulations are used to evaluate these methods under different data-generating settings, followed by two real-data applications on building cooling-load prediction and leukemia cell-type classification from high-dimensional gene expression data. Across simulated and real-data examples, the proposed regularizers improve predictive performance on unseen data and provide more effective complexity control than standard norm-based penalties, particularly when features are correlated or high-dimensional. --- ### Network Digital Untwinning: Towards Backward Optimization of Digital Twins Authors: Zifan Zhang, Dianwei Chen, Anjun Gao, Manhua Wang, Mingzhe Chen, Minghong Fang, Xianfeng Yang, Yuchen Liu Published: 2026-04-30 Categories: cs.NI, cs.DC, cs.LG arXiv: https://arxiv.org/abs/2605.00169v1 Network digital twins (NDTs) are transforming network management by offering precise virtual replicas of physical network systems. However, their reliance on diverse and sensitive data introduces significant challenges related to data management, regulatory compliance, and user privacy. In scenarios where selective data removal is necessary, such as device deactivation, network reconfiguration, or regulatory compliance, traditional approaches often fall short of preserving the integrity of the twin model. To address this gap, we introduce a network digital untwinning framework that enables the targeted removal of deprecated NDT contributions while maintaining model integrity. Our approach comprises two complementary components: Single Request Untwinning (\algO) and Parallel Request Untwinning (\algM) mechanisms. \algO leverages connectivity metrics based on geographical proximity, data distribution, and network-level attributes to identify and remove the target NDT along with its propagating influence. This is achieved through an optimally selected rollback checkpoint augmented with injected Gaussian noise, followed by a precise remapping phase. \algM extends this mechanism to efficiently handle multiple removal requests by clustering NDTs with similar attributes and performing a coordinated rollback and untwinning schedule. We provide theoretical guarantees on model indistinguishability from scratch-built twins, and validate the framework through extensive experiments on real-world traffic data, demonstrating its effectiveness and operational efficiency. --- ### Consistent Diffusion Language Models Authors: Hasan Amin, Yuan Gao, Yaser Souri, Subhojit Som, Ming Yin, Rajiv Khanna, Xia Song Published: 2026-04-30 Categories: cs.LG arXiv: https://arxiv.org/abs/2605.00161v2 Diffusion language models (DLMs) are an attractive alternative to autoregressive models because they promise sublinear-time, parallel generation, yet practical gains remain elusive as high-quality samples still demand hundreds of refinement steps. In continuous domains, consistency training along the probability-flow ODE is a popular recipe to accelerate diffusion. For discrete diffusion, no analogous sample-space ODE exists, making direct adaptation ill-defined. We argue that the right discrete substitute is the exact posterior bridge, the closed-form conditional law linking any two noise levels, which is available for broad corruptions including masked and uniform diffusion. Building on this observation, we introduce Multi-Path Discrete Consistency (MPDC), a new principle that trains a denoiser to be path-invariant in expectation across these stochastic bridges, and instantiate it as the Consistent Diffusion Language Model (CDLM), a single-stage training framework that does not require an already trained teacher model. Our CDLM objective recovers masked diffusion, continuous consistency models, and progressive or discrete distillation as analytic limits or empirical approximations of one common view. Empirically, CDLM establishes a new state of the art on both conditional and unconditional text-generation, consistently outperforming strong base discrete diffusion models and often even multi-stage distilled baselines across sampling budgets, with the largest gains in the few-step regime. Together, these results position CDLM as a principled and scalable foundation for the next generation of fast, high-fidelity discrete generative modeling. --- ### Wasserstein Distributionally Robust Regret Optimization for Reinforcement Learning from Human Feedback Authors: Yikai Wang, Shang Liu, Jose Blanchet Published: 2026-04-30 Categories: cs.LG, cs.CL, math.OC, stat.ML arXiv: https://arxiv.org/abs/2605.00155v4 Reinforcement learning from human feedback (RLHF) is a central post-training tool for aligning large language models, but its training reward is only a learned proxy for true human utility. This creates a decision problem under objective misspecification: the policy is optimized against an estimated reward, while deployment performance is governed by an unobserved population preference. The resulting gap leads to reward over-optimization, where proxy reward keeps improving after true quality deteriorates. We propose distributionally robust regret optimization (DRRO) for RLHF with a Wasserstein ambiguity set over reward laws, using promptwise $\ell_p$ distances between reward vectors as transport costs. Unlike standard distributionally robust optimization, which pessimizes worst-case value, DRRO pessimizes worst-case regret relative to the best policy under the same plausible reward perturbation. We show that the expressive-policy problem decomposes into promptwise regret problems. For each prompt, the inner adversary has a dual-norm closed form; under the $\ell_1$ transport cost used by our algorithm, the optimizer has a water-filling structure. These results lead to a practical policy-gradient algorithm that adds a simple sampled bonus to GRPO-style training. Theory and experiments both show that DRRO is less over-pessimistic than standard DRO and mitigates over-optimization more effectively than existing baselines. --- ### Wasserstein Distributionally Robust Regret Optimization for Reinforcement Learning from Human Feedback Authors: Yikai Wang, Shang Liu, Jose Blanchet Published: 2026-04-30 Categories: cs.LG, cs.CL, math.OC, stat.ML arXiv: https://arxiv.org/abs/2605.00155v3 Reinforcement learning from human feedback (RLHF) is a central post-training tool for aligning large language models, but its training reward is only a learned proxy for true human utility. This creates a decision problem under objective misspecification: the policy is optimized against an estimated reward, while deployment performance is governed by an unobserved population preference. The resulting gap leads to reward over-optimization, where proxy reward keeps improving after true quality deteriorates. We propose distributionally robust regret optimization (DRRO) for RLHF with a Wasserstein ambiguity set over reward laws, using promptwise $\ell_p$ distances between reward vectors as transport costs. Unlike standard distributionally robust optimization, which pessimizes worst-case value, DRRO pessimizes worst-case regret relative to the best policy under the same plausible reward perturbation. We show that the expressive-policy problem decomposes into promptwise regret problems. For each prompt, the inner adversary has a dual-norm closed form; under the $\ell_1$ transport cost used by our algorithm, the optimizer has a water-filling structure. These results lead to a practical policy-gradient algorithm that adds a simple sampled bonus to GRPO-style training. Theory and experiments both show that DRRO is less over-pessimistic than standard DRO and mitigates over-optimization more effectively than existing baselines. --- ### From Images2Mesh: A 3D Surface Reconstruction Pipeline for Non-Cooperative Space Objects Authors: Bala Prenith Reddy Gopu, Patrick Quinn, George M. Nehma, Madhur Tiwari, Matt Ueckermann, David Hinckley, Christopher McKenna Published: 2026-04-30 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.00147v1 On-orbit inspection imagery is crucial as it enables characterization of non-cooperative resident space objects, providing the geometry and structural condition essential for active debris removal and on-orbit servicing mission planning. However, most existing neural implicit surface reconstruction methods have been confined to synthetic or hardware-in-the-loop data with known camera poses and controlled illumination. In this work, we present a pipeline for neural implicit surface reconstruction of non-cooperative space objects from monocular inspection imagery. We demonstrate it on publicly released ISS inspection footage from the STS-119 mission and publicly released on-orbit inspection footage of an H-IIA rocket upper stage. We find that segmentation-based background removal is essential for successful camera pose estimation from real on-orbit footage, where background variation between frames caused direct processing to fail entirely. We further incorporate photometric correction of per-frame exposure variations and analyze its behavior across datasets, finding that performance in shadowed regions varies with the illumination characteristics of the input footage. --- ### Real-Time Frame- and Event-based Object Detection with Spiking Neural Networks on Edge Neuromorphic Hardware: Design, Deployment and Benchmark Authors: Udayanga G. W. K. N. Gamage, Yan Zeng, Cesar Cadena, Matteo Fumagalli, Silvia Tolu Published: 2026-04-30 Categories: cs.CV arXiv: https://arxiv.org/abs/2605.00146v1 Real-time object detection on energy-constrained platforms is critical for applications such as UAV-based inspection, autonomous navigation, and mobile robotics. Spiking neural networks (SNNs) on neuromorphic hardware are believed to be significantly more energy-efficient than conventional artificial neural networks (ANNs). In this work, we present a comprehensive methodology for designing general SNN detection architectures targeting neuromorphic platforms, along with the engineering adaptations required to deploy them on the state-of-the-art Neuromorphic processor, Intel Loihi 2. We benchmark SNN-based object detection on Loihi 2 using both frame-based and event-based datasets, comparing performance with ANN-based detection on the NVIDIA Jetson Orin Nano, NVIDIA Jetson Nano B01, and the Apple M2 CPU. Our results show that SNNs on Loihi 2 can perform real-time detection while achieving the lowest per-inference dynamic energy among all platforms. Also, Loihi 2 outperforms the other platforms in terms of power consumption, though ANNs on Jetson Orin Nano achieve higher inference rates. Furthermore, our ANN-to-SNN distillation-aware training enables SNNs to recover 87-100% of the detection accuracy of their ANN counterparts while maintaining lower inference latency; without distillation, SNNs exhibit an 11-27% accuracy drop. These results highlight the potential of neuromorphic systems for energy-efficient, real-time object detection at the edge. --- ### Timing is Everything: Temporal Scaffolding of Semantic Surprise in Humor Authors: Yuxi Ma, Yongqian Peng, Junchen Lyu, Chi Zhang, Yixin Zhu Published: 2026-04-30 Categories: cs.CL arXiv: https://arxiv.org/abs/2605.00143v2 Humor is a fundamental cognitive phenomenon in which humans derive pleasure from the expectation violations and their resolution, exemplifying the brain's dynamic capacity for predictive processing. Classical humor theories emphasize semantic incongruity as the primary driver of amusement, yet overlook temporal dynamics despite comedians' intuition that "timing is everything." The extent to which temporal structure contributes to humor appreciation and how it interacts with semantic content remains poorly understood. Here, we propose the Dual Prediction Violation (DPV) framework to capture the interplay between content and timing. By analyzing 828 professional Chinese stand-up performances, we show that temporal features substantially outweigh semantic incongruity in predicting audience appreciation. Specifically, we find that peak semantic violations matter more than average incongruity levels, and pauses systematically lengthen before high-surprise punchlines--a strategic coupling that distinguishes successful from unsuccessful performances. These findings reframe humor as temporally scaffolded, where timing and semantic content operate in strategic coordination rather than independently. Our DPV framework bridges humor theory with predictive processing, demonstrating that temporal structure plays a central role in naturalistic humor appreciation with implications for understanding multi-scale prediction integration in linguistic processing. --- ### Timing is Everything: Temporal Scaffolding of Semantic Surprise in Humor Authors: Yuxi Ma, Yongqian Peng, Junchen Lyu, Chi Zhang, Yixin Zhu Published: 2026-04-30 Categories: cs.CL arXiv: https://arxiv.org/abs/2605.00143v1 Humor is a fundamental cognitive phenomenon in which humans derive pleasure from the expectation violations and their resolution, exemplifying the brain's dynamic capacity for predictive processing. Classical humor theories emphasize semantic incongruity as the primary driver of amusement, yet overlook temporal dynamics despite comedians' intuition that "timing is everything." The extent to which temporal structure contributes to humor appreciation and how it interacts with semantic content remains poorly understood. Here, we propose the Dual Prediction Violation (DPV) framework to capture the interplay between content and timing. By analyzing 828 professional Chinese stand-up performances, we show that temporal features substantially outweigh semantic incongruity in predicting audience appreciation. Specifically, we find that peak semantic violations matter more than average incongruity levels, and pauses systematically lengthen before high-surprise punchlines--a strategic coupling that distinguishes successful from unsuccessful performances. These findings reframe humor as temporally scaffolded, where timing and semantic content operate in strategic coordination rather than independently. Our DPV framework bridges humor theory with predictive processing, demonstrating that temporal structure plays a central role in naturalistic humor appreciation with implications for understanding multi-scale prediction integration in linguistic processing. --- ### Technical Report: Activation Residual Hessian Quantization (ARHQ) for Low-Bit LLM Quantization Authors: YiFeng Wang, Zhun Sun, Keisuke Sakaguchi Published: 2026-04-30 Categories: cs.LG, cs.CL, cs.CV arXiv: https://arxiv.org/abs/2605.00140v1 We present Activation Residual Hessian Quantization (ARHQ), a post-training weight splitting method designed to mitigate error propagation in low-bit activation-weight quantization. By constructing an input-side residual Hessian from activation quantization residuals (G_x), ARHQ analytically identifies and isolates error-sensitive weight directions into a high-precision low-rank branch. This is achieved via a closed-form truncated SVD on the scaled weight matrix W G^{1/2}_x . Experimental results on Qwen3-4B-Thinking-2507 demonstrate that ARHQ significantly improves layer-wise SNR and preserves downstream reasoning performance on ZebraLogic even under aggressive quantization. The code is available at https://github.com/BeautMoonQ/ARHQ. --- ### Are Tools All We Need? Unveiling the Tool-Use Tax in LLM Agents Authors: Kaituo Zhang, Zhen Xiong, Mingyu Zhong, Zhimeng Jiang, Zhouyuan Yuan, Zhecheng Li, Ying Lin Published: 2026-04-30 Categories: cs.AI arXiv: https://arxiv.org/abs/2605.00136v1 Tool-augmented reasoning has become a popular direction for LLM-based agents, and it is widely assumed to improve reasoning and reliability. However, we demonstrate that this consensus does not always hold: in the presence of semantic distractors, tool-augmented reasoning does not necessarily outperform native CoT. To explain this performance gap, we propose a Factorized Intervention Framework that isolates the cost of prompt formatting, the overhead of the tool-calling protocol, and the actual gain from executing tools. Our analysis reveals a critical tradeoff: under semantic noise, the gains from tools often fail to offset the "tool-use tax", which is the performance degradation introduced by the tool-calling protocol itself. To address this, we introduce G-STEP, a lightweight inference-time gate to mitigate protocol-induced errors. While this yields partial recovery, our findings suggest that more substantial improvements still require strengthening the model's intrinsic reasoning and tool-interaction capabilities. ---