# Research Papers — 2026-03 · bundle page 238 of 243

> Full-text markdown of papers #11851–11900 of 12123 published in 2026-03, newest first. AI-readable mirrors of the full 12123-paper 2026-03 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2026-03/page/238
[← Previous bundle](https://aipod.dev/papers/archive/2026-03/page/237.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-03/page/239.md)

### Quasar: Quantized Self-Speculative Acceleration for Rapid Inference via Memory-Efficient Verification

Authors: Guang Huang, Zeyi Wen
Published: 2026-03-02
Categories: cs.DC, cs.LG
arXiv: https://arxiv.org/abs/2603.01399v1

Speculative Decoding (SD) has emerged as a premier technique for accelerating Large Language Model (LLM) inference by decoupling token generation into rapid drafting and parallel verification. While recent advancements in self-speculation and lookahead decoding have successfully minimized drafting overhead, they have shifted the primary performance bottleneck to the verification phase. Since verification requires a full forward pass of the target model, it remains strictly memory-bandwidth bound, fundamentally limiting the maximum achievable speedup.In this paper, we introduce \textbf{Quasar} (\textbf{Qua}ntized \textbf{S}elf-speculative \textbf{A}cceleration for \textbf{R}apid Inference), a novel, training-free framework designed to overcome this "memory wall" by employing low-bit quantization specifically for the verification stage. Our empirical analysis reveals that while aggressive structural pruning significantly degrades verification accuracy, quantization-based verification preserves the logit distribution with high fidelity while effectively halving memory traffic. Extensive experiments on state-of-the-art models (e.g., OpenPangu and Qwen3) demonstrate that Quasar maintains a speculative acceptance length comparable to full-precision methods while achieving a $1.28\times$ improvement in end-to-end throughput. Being orthogonal to existing drafting strategies, Quasar offers a generic and efficient pathway to accelerate the verification leg of speculative execution. Code is available at https://github.com/Tom-HG/Quasar.

---

### Continuous Exposure-Time Modeling for Realistic Atmospheric Turbulence Synthesis

Authors: Junwei Zeng, Dong Liang, Sheng-Jun Huang, Kun Zhan, Songcan Chen
Published: 2026-03-02
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.01398v2

Atmospheric turbulence significantly degrades long-range imaging by introducing geometric warping and exposure-time-dependent blur, which adversely affects both visual quality and the performance of high-level vision tasks. Existing methods for synthesizing turbulence effects often oversimplify the relationship between blur and exposure-time, typically assuming fixed or binary exposure settings. This leads to unrealistic synthetic data and limited generalization capability of trained models. To address this gap, we revisit the modulation transfer function (MTF) formulation and propose a novel Exposure-Time-dependent MTF (ET-MTF) that models blur as a continuous function of exposure-time. For blur synthesis, we derive a tilt-invariant point spread function (PSF) from the ET-MTF, which, when integrated with a spatially varying blur-width field, provides a comprehensive and physically accurate characterization of turbulence-induced blur. Building on this synthesis pipeline, we construct ET-Turb, a large-scale synthetic turbulence dataset that explicitly incorporates continuous exposure-time modeling across diverse optical and atmospheric conditions. The dataset comprises 5,083 videos (2,005,835 frames), partitioned into 3,988 training and 1,095 test videos. Extensive experiments demonstrate that models trained on ET-Turb produce more realistic restorations and achieve superior generalization on real-world turbulence data compared to those trained on other datasets. The dataset is publicly available at: github.com/Jun-Wei-Zeng/ET-Turb.

---

### HarmonyCell: Automating Single-Cell Perturbation Modeling under Semantic and Distribution Shifts

Authors: Wenxuan Huang, Mingyu Tsoi, Yanhao Huang, Xinjie Mao, Xue Xia, Hao Wu, Jiaqi Wei, Yuejin Yang, Lang Yu, Cheng Tan, Xiang Zhang, Zhangyang Gao et al.
Published: 2026-03-02
Categories: cs.AI, cs.CE, q-bio.QM
arXiv: https://arxiv.org/abs/2603.01396v2

Single-cell perturbation studies face dual heterogeneity bottlenecks: (i) semantic heterogeneity--identical biological concepts encoded under incompatible metadata schemas across datasets; and (ii) statistical heterogeneity--distribution shifts from biological variation demanding dataset-specific inductive biases. We propose HarmonyCell, an end-to-end agent framework resolving each challenge through a dedicated mechanism: an LLM-driven Semantic Unifier autonomously maps disparate metadata into a canonical interface without manual intervention; and an adaptive Monte Carlo Tree Search engine operates over a hierarchical action space to synthesize architectures with optimal statistical inductive biases for distribution shifts. Evaluated across diverse perturbation tasks under both semantic and distribution shifts, HarmonyCell achieves a 95% valid execution rate on heterogeneous input datasets (versus 0% for general agents) while matching or even exceeding expert-designed baselines in rigorous out-of-distribution evaluations. This dual-track orchestration enables scalable automatic virtual cell modeling without dataset-specific engineering.

---

### Invariant-Stratified Propagation for Expressive Graph Neural Networks

Authors: Asela Hevapathige, Ahad N. Zehmakan, Asiri Wijesinghe, Saman Halgamuge
Published: 2026-03-02
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2603.01388v1

Graph Neural Networks (GNNs) face fundamental limitations in expressivity and capturing structural heterogeneity. Standard message-passing architectures are constrained by the 1-dimensional Weisfeiler-Leman (1-WL) test, unable to distinguish graphs beyond degree sequences, and aggregate information uniformly from neighbors, failing to capture how nodes occupy different structural positions within higher-order patterns. While methods exist to achieve higher expressivity, they incur prohibitive computational costs and lack unified frameworks for flexibly encoding diverse structural properties. To address these limitations, we introduce Invariant-Stratified Propagation (ISP), a framework comprising both a novel WL variant (ISP-WL) and its efficient neural network implementation (ISPGNN). ISP stratifies nodes according to graph invariants, processing them in hierarchical strata that reveal structural distinctions invisible to 1-WL. Through hierarchical structural heterogeneity encoding, ISP quantifies differences in nodes' structural positions within higher-order patterns, distinguishing interactions where participants occupy different roles from those with uniform participation. We provide formal theoretical analysis establishing enhanced expressivity beyond 1-WL, convergence guarantees, and inherent resistance to oversmoothing. Extensive experiments across graph classification, node classification, and influence estimation demonstrate consistent improvements over standard architectures and state-of-the-art expressive baselines.

---

### Toward Graph-Tokenizing Large Language Models with Reconstructive Graph Instruction Tuning

Authors: Zhongjian Zhang, Xiao Wang, Mengmei Zhang, Jiarui Tan, Chuan Shi
Published: 2026-03-02
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2603.01385v1

The remarkable success of large language models (LLMs) has motivated researchers to adapt them as universal predictors for various graph-related tasks, with the ultimate goal of developing a graph foundation model that generalizes diverse scenarios. The key challenge is to align graph data with language spaces so that LLMs can better comprehend graphs. As a popular paradigm, Graph-Tokenizing LLMs (GTokenLLMs) encode complex structures and lengthy texts into a graph token sequence, and then align them with text tokens via language instructions tuning. Despite their initial success, our information-theoretic analysis reveals that existing GTokenLLMs rely solely on text supervision from language instructions, which achieve only implicit graph-text alignment, resulting in a text-dominant bias that underutilizes graph context. To overcome this limitation, we first prove that the alignment objective is upper-bounded by the mutual information between the input graphs and their hidden representations in the LLM, which motivates us to improve this upper bound to achieve better alignment. To this end, we further propose a reconstructive graph instruction tuning pipeline, RGLM. Our key idea is to reconstruct the graph information from the LLM's graph token outputs, explicitly incorporating graph supervision to constrain the alignment process. Technically, we embody RGLM by exploring three distinct variants from two complementary perspectives: RGLM-Decoder from the input space; RGLM-Similarizer and RGLM-Denoiser from the latent space. Additionally, we theoretically analyze the alignment effectiveness of each variant. Extensive experiments on various benchmarks and task scenarios validate the effectiveness of the proposed RGLM, paving the way for new directions in GTokenLLMs' alignment research.

---

### End-to-End Simultaneous Dysarthric Speech Reconstruction with Frame-Level Adaptor and Multiple Wait-k Knowledge Distillation

Authors: Minghui Wu, Haitao Tang, Jiahuan Fan, Ruizhi Liao, Yanyong Zhang
Published: 2026-03-02
Categories: cs.SD, cs.CL
arXiv: https://arxiv.org/abs/2603.01382v1

Dysarthric speech reconstruction (DSR) typically employs a cascaded system that combines automatic speech recognition (ASR) and sentence-level text-to-speech (TTS) to convert dysarthric speech into normally-prosodied speech. However, dysarthric individuals often speak more slowly, leading to excessively long response times in such systems, rendering them impractical in long-speech scenarios. Cascaded DSR systems based on streaming ASR and incremental TTS can help reduce latency. However, patients with differing dysarthria severity exhibit substantial pronunciation variability for the same text, resulting in poor robustness of ASR and limiting the intelligibility of reconstructed speech. In addition, incremental TTS suffers from poor prosodic feature prediction due to a limited receptive field. In this study, we propose an end-to-end simultaneous DSR system with two key innovations: 1) A frame-level adaptor module is introduced to bridge ASR and TTS. By employing explicit-implicit semantic information fusion and joint module training, it enhances the error tolerance of TTS to ASR outputs. 2) A multiple wait-k autoregressive TTS module is designed to mitigate prosodic degradation via multi-view knowledge distillation. Our system has an average response time of 1.03 seconds on Tesla A100, with an average real-time factor (RTF) of 0.71. On the UASpeech dataset, it attains a mean opinion score (MOS) of 4.67 and demonstrates a 54.25% relative reduction in word error rate (WER) compared to the state-of-the-art. Our demo is available at: https://wflrz123.github.io/

---

### 3BASiL: An Algorithmic Framework for Sparse plus Low-Rank Compression of LLMs

Authors: Mehdi Makni, Xiang Meng, Rahul Mazumder
Published: 2026-03-02
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2603.01376v1

Sparse plus Low-Rank $(\mathbf{S} + \mathbf{LR})$ decomposition of Large Language Models (LLMs) has emerged as a promising direction in model compression, aiming to decompose pre-trained model weights into a sum of sparse and low-rank matrices $(\mathbf{W} \approx \mathbf{S} + \mathbf{LR})$. Despite recent progress, existing methods often suffer from substantial performance degradation compared to dense models. In this work, we introduce 3BASiL-TM, an efficient one-shot post-training method for $(\mathbf{S} + \mathbf{LR})$ decomposition of LLMs that addresses this gap. Our approach first introduces a novel 3-Block Alternating Direction Method of Multipliers (ADMM) method, termed 3BASiL, to minimize the layer-wise reconstruction error with convergence guarantees. We then design an efficient transformer-matching (TM) refinement step that jointly optimizes the sparse and low-rank components across transformer layers. This step minimizes a novel memory-efficient loss that aligns outputs at the transformer level. Notably, the TM procedure is universal as it can enhance any $(\mathbf{S} + \mathbf{LR})$ decomposition, including pure sparsity. Our numerical experiments show that 3BASiL-TM reduces the WikiText2 perplexity gap relative to dense LLaMA-8B model by over 30% under a (2:4 Sparse + 64 LR) configuration, compared to prior methods. Moreover, our method achieves over 2.5x faster compression runtime on an A100 GPU compared to SOTA $(\mathbf{S} + \mathbf{LR})$ method. Our code is available at https://github.com/mazumder-lab/3BASiL.

---

### Words & Weights: Streamlining Multi-Turn Interactions via Co-Adaptation

Authors: Chenxing Wei, Hong Wang, Ying He, Zhongxiang Dai, Bo Jiang, F. Richard Yu, Yao Shu
Published: 2026-03-02
Categories: cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2603.01375v2

Test-time policy adaptation for multi-turn interactions (T2PAM) is essential for aligning Large Language Models (LLMs) with dynamic user needs during inference time. However, existing paradigms commonly treat test-time adaptation as a single-axis problem, either purely refining instructions (Prompt Engineering) or only adjusting weights (Test-Time Training), ignoring that interaction failures stem from a coupled mix of ambiguity and incapacity. We argue that these two optimization paths are not merely additive but synergistic: semantic clarity acts as a pre-conditioner for effective parameter updates. To this end, we propose ROSA2, a framework that reformulates interaction as a joint optimization problem over the heterogeneous space of Words and Weights. By mathematically decomposing the error signal, ROSA2 utilizes textual gradients to rectify intent ambiguity and parameter updates to bridge capability gaps. Theoretically, we prove that this co-adaptation strictly reduces the required parameter shift for convergence. Empirically, ROSA2 outperforms state-of-the-art baselines by 30% on MATH while reducing interaction turns by 40%, demonstrating that refining the context unlocks the true potential of parameter updates.

---

### Words & Weights: Streamlining Multi-Turn Interactions via Co-Adaptation

Authors: Chenxing Wei, Hong Wang, Ying He, Zhongxiang Dai, Bo Jiang, F. Richard Yu, Yao Shu
Published: 2026-03-02
Categories: cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2603.01375v1

Test-time policy adaptation for multi-turn interactions (T2PAM) is essential for aligning Large Language Models (LLMs) with dynamic user needs during inference time. However, existing paradigms commonly treat test-time adaptation as a single-axis problem, either purely refining instructions (Prompt Engineering) or only adjusting weights (Test-Time Training), ignoring that interaction failures stem from a coupled mix of ambiguity and incapacity. We argue that these two optimization paths are not merely additive but synergistic: semantic clarity acts as a pre-conditioner for effective parameter updates. To this end, we propose ROSA2, a framework that reformulates interaction as a joint optimization problem over the heterogeneous space of Words and Weights. By mathematically decomposing the error signal, ROSA2 utilizes textual gradients to rectify intent ambiguity and parameter updates to bridge capability gaps. Theoretically, we prove that this co-adaptation strictly reduces the required parameter shift for convergence. Empirically, ROSA2 outperforms state-of-the-art baselines by 30% on MATH while reducing interaction turns by 40%, demonstrating that refining the context unlocks the true potential of parameter updates.

---

### Causal Neural Probabilistic Circuits

Authors: Weixin Chen, Han Zhao
Published: 2026-03-02
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2603.01372v2

Concept Bottleneck Models (CBMs) enhance the interpretability of end-to-end neural networks by introducing a layer of concepts and predicting the class label from the concept predictions. A key property of CBMs is that they support interventions, i.e., domain experts can correct mispredicted concept values at test time to improve the final accuracy. However, typical CBMs apply interventions by overwriting only the corrected concept while leaving other concept predictions unchanged, which ignores causal dependencies among concepts. To address this, we propose the Causal Neural Probabilistic Circuit (CNPC), which combines a neural attribute predictor with a causal probabilistic circuit compiled from a causal graph. This circuit supports exact, tractable causal inference that inherently respects causal dependencies. Under interventions, CNPC models the class distribution based on a Product of Experts (PoE) that fuses the attribute predictor's predictive distribution with the interventional marginals computed by the circuit. We theoretically characterize the compositional interventional error of CNPC w.r.t. its modules and identify conditions under which CNPC closely matches the ground-truth interventional class distribution. Experiments on five benchmark datasets in both in-distribution and out-of-distribution settings show that, compared with five baseline models, CNPC achieves higher task accuracy across different numbers of intervened attributes.

---

### TIMI: Training-Free Image-to-3D Multi-Instance Generation with Spatial Fidelity

Authors: Xiao Cai, Pengpeng Zeng, Ji Zhang, Heng Tao Shen, Jingkuan Song, Lianli Gao
Published: 2026-03-02
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.01371v2

Precise spatial fidelity in Image-to-3D multi-instance generation is critical for downstream real-world applications. Recent work attempts to address this by fine-tuning pre-trained Image-to-3D (I23D) models on multi-instance datasets, which incurs substantial training overhead and struggles to guarantee spatial fidelity. In fact, we observe that pre-trained I23D models already possess meaningful spatial priors, which remain underutilized as evidenced by instance entanglement issues. Motivated by this, we propose TIMI, a novel Training-free framework for Image-to-3D Multi-Instance generation that achieves high spatial fidelity. Specifically, we first introduce an Instance-aware Separation Guidance (ISG) module, which facilitates instance disentanglement during the early denoising stage. Next, to stabilize the guidance introduced by ISG, we devise a Spatial-stabilized Geometry-adaptive Update (SGU) module that promotes the preservation of the geometric characteristics of instances while maintaining their relative relationships. Extensive experiments demonstrate that our method yields better performance in terms of both global layout and distinct local instances compared to existing multi-instance methods, without requiring additional training and with faster inference speed.

---

### DARS: Dysarthria-Aware Rhythm-Style Synthesis for ASR Enhancement

Authors: Minghui Wu, Xueling Liu, Jiahuan Fan, Haitao Tang, Yanyong Zhang, Yue Zhang
Published: 2026-03-02
Categories: cs.SD, cs.CL
arXiv: https://arxiv.org/abs/2603.01369v1

Dysarthric speech exhibits abnormal prosody and significant speaker variability, presenting persistent challenges for automatic speech recognition (ASR). While text-to-speech (TTS)-based data augmentation has shown potential, existing methods often fail to accurately model the pathological rhythm and acoustic style of dysarthric speech. To address this, we propose DARS, a dysarthria-aware rhythm-style synthesis framework based on the Matcha-TTS architecture. DARS incorporates a multi-stage rhythm predictor optimized by contrastive preferences between normal and dysarthric speech, along with a dysarthric-style conditional flow matching mechanism, jointly enhancing temporal rhythm reconstruction and pathological acoustic style simulation. Experiments on the TORGO dataset demonstrate that DARS achieves a Mean Cepstral Distortion (MCD) of 4.29, closely approximating real dysarthric speech. Adapting a Whisper-based ASR system with synthetic dysarthric speech from DARS achieves a 54.22% relative reduction in word error rate (WER) compared to state-of-the-art methods, demonstrating the framework's effectiveness in enhancing recognition performance.

---

### DUEL: Exact Likelihood for Masked Diffusion via Deterministic Unmasking

Authors: Gilad Turok, Chris De Sa, Volodymyr Kuleshov
Published: 2026-03-02
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.01367v2

Masked diffusion models (MDMs) generate text by iteratively selecting positions to unmask and then predicting tokens at those positions. Yet MDMs lack proper likelihood evaluation: the evidence lower bound (ELBO) is not only a loose bound on log-likelihood, but, as we show, is also computed under the training distribution rather than the test-time distribution. We resolve this within our DUEL framework, which unifies leading MDM sampling strategies that employ $\textit{deterministic}$ position selection. We prove that DUEL samplers admit $\textbf{exact likelihood computation under the test-time distribution}$ -- giving MDMs $\textit{proper}$ likelihood, and hence proper perplexity, for the first time. This proper perplexity is the natural analogue of autoregressive perplexity and lets us revisit key questions about MDMs. $\textbf{MDMs are substantially better than previously thought}$: the MDM-autoregressive perplexity gap shrinks by up to $32\%$ on in-domain data and $82\%$ on zero-shot benchmarks. DUEL enables the first principled comparison of fast,parallel samplers across compute budgets -- an analysis impossible with the ELBO and unreliable with generative perplexity -- identifying a strong default method. Finally, oracle search over position orderings reveals MDMs can far surpass autoregressive models -- achieving $36.47$ vs. $52.11$ perplexity on AG News -- demonstrating the ceiling of MDM performance has not yet been reached.

---

### NM-DEKL$^3_\infty$: A Three-Layer Non-Monotone Evolving Dependent Type Logic

Authors: Peng Chen
Published: 2026-03-02
Categories: cs.LO, cs.CL
arXiv: https://arxiv.org/abs/2603.01366v1

We present a new dependent type system, NM-DEKL$^3_\infty$ (Non-Monotone Dependent Knowledge-Enhanced Logic), for formalising evolving knowledge in dynamic environments. The system uses a three-layer architecture separating a computational layer, a constructive knowledge layer, and a propositional knowledge layer. We define its syntax and semantics and establish Soundness and Equational Completeness; we construct a syntactic model and prove that it is initial in the category of models, from which equational completeness follows. We also give an embedding into the $μ$-calculus and a strict expressiveness inclusion (including the expressibility of non-bisimulation-invariant properties).

---

### Align and Filter: Improving Performance in Asynchronous On-Policy RL

Authors: Homayoun Honari, Roger Creus Castanyer, Michael Przystupa, Michael Noukhovitch, Pablo Samuel Castro, Glen Berseth
Published: 2026-03-02
Categories: cs.LG, cs.AI, cs.RO, eess.SY
arXiv: https://arxiv.org/abs/2603.01365v1

Distributed training and increasing the gradient update frequency are practical strategies to accelerate learning and improve performance, but both exacerbate a central challenge: \textit{policy lag}, which is the mismatch between the behavior policy generating data and the learning policy being updated. Policy lag can hinder the scaling of on-policy learning algorithms to larger problems. In this paper, we identify the sources of policy lag caused by distributed learning and high update frequency. We use the findings to propose \textit{total Variation-based Advantage aligned Constrained policy Optimization (\methodacronym)} as a practical approach to mitigate policy lag. We empirically validate our method and show that it offers better robustness to policy lag in classic RL tasks and a modern RL for LLM math reasoning task.

---

### Fed-GAME: Personalized Federated Learning with Graph Attention Mixture-of-Experts For Time-Series Forecasting

Authors: Yi Li, Han Liu, Mingfeng Fan, Guo Chen, Chaojie Li, Biplab Sikdar
Published: 2026-03-02
Categories: cs.LG, cs.DC
arXiv: https://arxiv.org/abs/2603.01363v1

Federated learning (FL) on graphs shows promise for distributed time-series forecasting. Yet, existing methods rely on static topologies and struggle with client heterogeneity. We propose Fed-GAME, a framework that models personalized aggregation as message passing over a learnable dynamic implicit graph. The core is a decoupled parameter difference-based update protocol, where clients transmit parameter differences between their fine-tuned private model and a shared global model. On the server, these differences are decomposed into two streams: (1) averaged difference used to updating the global model for consensus (2) the selective difference fed into a novel Graph Attention Mixture-of-Experts (GAME) aggregator for fine-grained personalization. In this aggregator, shared experts provide scoring signals while personalized gates adaptively weight selective updates to support personalized aggregation. Experiments on two real-world electric vehicle charging datasets demonstrate that Fed-GAME outperforms state-of-the-art personalized FL baselines.

---

### MixerCSeg: An Efficient Mixer Architecture for Crack Segmentation via Decoupled Mamba Attention

Authors: Zilong Zhao, Zhengming Ding, Pei Niu, Wenhao Sun, Feng Guo
Published: 2026-03-02
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2603.01361v1

Feature encoders play a key role in pixel-level crack segmentation by shaping the representation of fine textures and thin structures. Existing CNN-, Transformer-, and Mamba-based models each capture only part of the required spatial or structural information, leaving clear gaps in modeling complex crack patterns. To address this, we present MixerCSeg, a mixer architecture designed like a coordinated team of specialists, where CNN-like pathways focus on local textures, Transformer-style paths capture global dependencies, and Mamba-inspired flows model sequential context within a single encoder. At the core of MixerCSeg is the TransMixer, which explores Mamba's latent attention behavior while establishing dedicated pathways that naturally express both locality and global awareness. To further enhance structural fidelity, we introduce a spatial block processing strategy and a Direction-guided Edge Gated Convolution (DEGConv) that strengthens edge sensitivity under irregular crack geometries with minimal computational overhead. A Spatial Refinement Multi-Level Fusion (SRF) module is then employed to refine multi-scale details without increasing complexity. Extensive experiments on multiple crack segmentation benchmarks show that MixerCSeg achieves state-of-the-art performance with only 2.05 GFLOPs and 2.54 M parameters, demonstrating both efficiency and strong representational capability. The code is available at https://github.com/spiderforest/MixerCSeg.

---

### ASTRA-bench: Evaluating Tool-Use Agent Reasoning and Action Planning with Personal User Context

Authors: Zidi Xiu, David Q. Sun, Kevin Cheng, Maitrik Patel, Josh Date, Yizhe Zhang, Jiarui Lu, Omar Attia, Raviteja Vemulapalli, Oncel Tuzel, Meng Cao, Samy Bengio
Published: 2026-03-02
Categories: cs.AI
arXiv: https://arxiv.org/abs/2603.01357v1

Next-generation AI must manage vast personal data, diverse tools, and multi-step reasoning, yet most benchmarks remain context-free and single-turn. We present ASTRA-bench (Assistant Skills in Tool-use, Reasoning \& Action-planning), a benchmark that uniquely unifies time-evolving personal context with an interactive toolbox and complex user intents. Our event-driven pipeline generates 2,413 scenarios across four protagonists, grounded in longitudinal life events and annotated by referential, functional, and informational complexity. Evaluation of state-of-the-art models (e.g., Claude-4.5-Opus, DeepSeek-V3.2) reveals significant performance degradation under high-complexity conditions, with argument generation emerging as the primary bottleneck. These findings expose critical limitations in current agents' ability to ground reasoning within messy personal context and orchestrate reliable multi-step plans. We release ASTRA-bench with a full execution environment and evaluation scripts to provide a diagnostic testbed for developing truly context-aware AI assistants.

---

### Constructing Synthetic Instruction Datasets for Improving Reasoning in Domain-Specific LLMs: A Case Study in the Japanese Financial Domain

Authors: Yuma Okochi, Fabio Milentiansen Sim, Tomoyasu Okada
Published: 2026-03-02
Categories: cs.LG, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2603.01353v1

In adapting LLMs to specific domains, achieving both domain expertise and reasoning ability remains an urgent challenge. This study proposes a general method for constructing high-quality synthetic instruction data for any domain, starting from domain-specific vocabulary. As a demonstration, we applied this method to the financial domain and constructed a large-scale instruction dataset totaling approximately 9.5 billion tokens with Chain-of-Thought reasoning traces. Evaluation results confirmed performance improvements over baseline models on financial benchmarks, demonstrating the effectiveness of our approach. We also report findings on the impact of reasoning trace length on performance and its limitations. Lastly, we open-source our models and datasets on https://huggingface.co/nri-ai .

---

### UTICA: Multi-Objective Self-Distllation Foundation Model Pretraining for Time Series Classification

Authors: Yessin Moakher, Youssef Attia El Hili, Vasilii Feofanov
Published: 2026-03-02
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2603.01348v1

Self-supervised foundation models have achieved remarkable success across domains, including time series. However, the potential of non-contrastive methods, a paradigm that has driven significant advances in computer vision, remains underexplored for time series. In this work, we adapt DINOv2-style self-distillation to pretrain a time series foundation model, building on the Mantis tokenizer and transformer encoder architecture as our backbone. Through a student-teacher framework, our method Utica learns representations that capture both temporal invariance via augmented crops and fine-grained local structure via patch masking. Our approach achieves state-of-the-art classification performance on both UCR and UEA benchmarks. These results suggest that non-contrastive methods are a promising and complementary pretraining strategy for time series foundation models.

---

### Relatively Smart: A New Approach for Instance-Optimal Learning

Authors: Shaddin Dughmi, Alireza F. Pour
Published: 2026-03-02
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2603.01346v2

We revisit the framework of Smart PAC learning, which seeks supervised learners which compete with semi-supervised learners that are provided full knowledge of the marginal distribution on unlabeled data. Prior work has shown that such marginal-by-marginal guarantees are possible for "most" marginals, with respect to an arbitrary fixed and known measure, but not more generally. We discover that this failure can be attributed to an "indistinguishability" phenomenon: There are marginals which cannot be statistically distinguished from other marginals that require different learning approaches. In such settings, semi-supervised learning cannot certify its guarantees from unlabeled data, rendering them arguably non-actionable. We propose relatively smart learning, a new framework which demands that a supervised learner compete only with the best "certifiable" semi-supervised guarantee. We show that such modest relaxation suffices to bypass the impossibility results from prior work. In the distribution-free setting, we show that the One-Inclusion Graph learner is relatively smart up to squaring the sample complexity, and show that no supervised learning algorithm can do better. For distribution-family settings, we show that relatively smart learning can be impossible or can require idiosyncratic learning approaches, and its difficulty can be non-monotone in the inclusion order on distribution families.

---

### PanCanBench: A Comprehensive Benchmark for Evaluating Large Language Models in Pancreatic Oncology

Authors: Yimin Zhao, Sheela R. Damle, Simone E. Dekker, Scott Geng, Karly Williams Silva, Jesse J Hubbard, Manuel F Fernandez, Fatima Zelada-Arenas, Alejandra Alvarez, Brianne Flores, Alexis Rodriguez, Stephen Salerno et al.
Published: 2026-03-02
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2603.01343v1

Large language models (LLMs) have achieved expert-level performance on standardized examinations, yet multiple-choice accuracy poorly reflects real-world clinical utility and safety. As patients and clinicians increasingly use LLMs for guidance on complex conditions such as pancreatic cancer, evaluation must extend beyond general medical knowledge. Existing frameworks, such as HealthBench, rely on simulated queries and lack disease-specific depth. Moreover, high rubric-based scores do not ensure factual correctness, underscoring the need to assess hallucinations. We developed a human-in-the-loop pipeline to create expert rubrics for de-identified patient questions from the Pancreatic Cancer Action Network (PanCAN). The resulting benchmark, PanCanBench, includes 3,130 question-specific criteria across 282 authentic patient questions. We evaluated 22 proprietary and open-source LLMs using an LLM-as-a-judge framework, measuring clinical completeness, factual accuracy, and web-search integration. Models showed substantial variation in rubric-based completeness, with scores ranging from 46.5% to 82.3%. Factual errors were common, with hallucination rates (the percentages of responses containing at least one factual error) ranging from 6.0% for Gemini-2.5 Pro and GPT-4o to 53.8% for Llama-3.1-8B. Importantly, newer reasoning-optimized models did not consistently improve factuality: although o3 achieved the highest rubric score, it produced inaccuracies more frequently than other GPT-family models. Web-search integration did not inherently guarantee better responses. The average score changed from 66.8% to 63.9% for Gemini-2.5 Pro and from 73.8% to 72.8% for GPT-5 when web search was enabled. Synthetic AI-generated rubrics inflated absolute scores by 17.9 points on average while generally maintaining similar relative ranking.

---

### SubstratumGraphEnv: Reinforcement Learning Environment (RLE) for Modeling System Attack Paths

Authors: Bahirah Adewunmi, Edward Raff, Sanjay Purushotham
Published: 2026-03-02
Categories: cs.CR, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2603.01340v1

Automating network security analysis, particularly the identification of potential attack paths, presents significant challenges. Due in part to the sequential, interconnected, and evolutionary nature of system events which most artificial intelligence (AI) techniques struggle to model effectively. This paper proposes a Reinforcement Learning (RL) environment generation framework that simulates the sequence of processes executed on a Windows operating system, enabling dynamic modeling of malicious processes on a system. This methodology models operating system state and transitions using a graph representation. This graph is derived from open-source System Monitor (Sysmon) logs. To address the variety in system event types, fields, and log formats, a mechanism was developed to capture and model parent-child processes from Sysmon logs. A Gymnasium environment (SubstratumGraphEnv) was constructed to establish the perceptible basis for an RL environment, and a customized PyTorch interface was also built (SubstratumBridge) to translate Gymnasium graphs into Deep Reinforcement Learning (DRL) observations and discrete actions. Graph Convolutional Networks (GCNs) concretize the graph's local and global state, which feed the distinct policy and critic heads of an Advantage Actor-Critic (A2C) model. This work's central contribution lies in the design of a novel deep graphical RL environment that automates translation of sequential user and system events, furnishing crucial context for cybersecurity analysis. This work provides a foundation for future research into shaping training parameters and advanced reward shaping, while also offering insight into which system events attributes are critical to training autonomous RL agents.

---

### Causal Effects with Unobserved Unit Types in Interacting Human-AI Systems

Authors: William Overman, Sadegh Shirani, Mohsen Bayati
Published: 2026-03-02
Categories: stat.ML, cs.LG
arXiv: https://arxiv.org/abs/2603.01339v1

We study experiments on interacting populations of humans and AI agents, where both unit types and the interaction network remain unobserved. Although causal effects propagate throughout the system, the goal is to estimate effects on humans. Examples include online platforms where human users interact alongside AI-driven accounts. We assume a human-AI prior that gives each unit a probability of being human. While humans cannot be distinguished at the unit level, the prior allows us to compute the average human composition within large subpopulations. We then model outcome dynamics through a causal message passing (CMP) framework and analyze sample-mean outcomes across subpopulations. We show that by constructing subpopulations that vary in expected human composition and treatment exposure, one can consistently recover human-specific causal effects. Our results characterize when distributional knowledge of population composition (without observing unit types or the interaction network) is sufficient for identification. We validate the approach on a simulated human-AI platform driven by behaviorally differentiated LLM agents. Together, these results provide a theoretical and practical framework for experimentation in emerging human-AI systems.

---

### Adaptive Estimation and Inference in Conditional Moment Models via the Discrepancy Principle

Authors: Jiyuan Tan, Vasilis Syrgkanis
Published: 2026-03-02
Categories: stat.ML, cs.LG
arXiv: https://arxiv.org/abs/2603.01337v1

We study adaptive estimation and inference in ill-posed linear inverse problems defined by conditional moment restrictions. Existing regularized estimators such as Regularized DeepIV (RDIV) require prior knowledge of the smoothness of the nuisance function, typically encoded by a beta source condition to tune their regularization parameters. In practice, this smoothness is unknown, and misspecified hyperparameters can lead to suboptimal convergence or instability. We introduce a discrepancy-principle-based framework for adaptive hyperparameter selection that automatically balances bias and variance without relying on the unknown smoothness parameter. Our framework applies to both RDIV (Li et al. (2024)) and the Tikhonov Regularized Adversarial Estimator (TRAE) (Bennett et al. (2023a)) and achieves the same rates in both weak and strong metrics. Building on this, we construct a fully adaptive doubly robust estimator for linear functionals that attains the optimal rate of the better-conditioned primal or dual problem, providing a practical, theoretically grounded approach for adaptive inference in ill-posed econometric models.

---

### Provable and Practical In-Context Policy Optimization for Self-Improvement

Authors: Tianrun Yu, Yuxiao Yang, Zhaoyang Wang, Kaixiang Zhao, Porter Jenkins, Xuchao Zhang, Chetan Bansal, Huaxiu Yao, Weitong Zhang
Published: 2026-03-02
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2603.01335v1

We study test-time scaling, where a model improves its answer through multi-round self-reflection at inference. We introduce In-Context Policy Optimization (ICPO), in which an agent optimizes its response in context using self-assessed or externally observed rewards without modifying its parameters. To explain this ICPO process, we theoretically show that with sufficient pretraining under a novel Fisher-weighted logit-matching objective, a single-layer linear self-attention model can provably imitate policy-optimization algorithm for linear bandits. Building on this theory, we propose Minimum-Entropy ICPO (ME-ICPO), a practical algorithm that iteratively uses its response and self-assessed reward to refine its response in-context at inference time. By selecting the responses and their rewards with minimum entropy, ME-ICPO ensures the robustness of the self-assessed rewards via majority voting. Across standard mathematical reasoning tasks, ME-ICPO attains competitive, top-tier performance while keeping inference costs affordable compared with other inference-time algorithms. Overall, ICPO provides a principled understanding of self-reflection in LLMs and yields practical benefits for test-time scaling for mathematical reasoning.

---

### Perspective-Equivariant Fine-tuning for Multispectral Demosaicing without Ground Truth

Authors: Andrew Wang, Mike Davies
Published: 2026-03-02
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.01332v2

Multispectral demosaicing is crucial to reconstruct full-resolution spectral images from snapshot mosaiced measurements, enabling real-time imaging from neurosurgery to autonomous driving. Classical methods are blurry, while supervised learning requires costly ground truth (GT) obtained from slow line-scanning systems. We propose Perspective-Equivariant Fine-tuning for Demosaicing (PEFD), a framework that learns multispectral demosaicing from mosaiced measurements alone. PEFD a) exploits the projective geometry of camera-based imaging systems to leverage a richer group structure than previous demosaicing methods to recover more null-space information, and b) learns efficiently without GT by adapting pretrained foundation models designed for 1-3 channel imaging. On surgical and automotive datasets, PEFD recovers fine details such as blood vessels and preserves spectral fidelity, substantially outperforming recent approaches, nearing supervised performance. Furthermore, the performance of PEFD is demonstrated on raw, unprocessed data from a commercial multispectral sensor. Code is at https://github.com/Andrewwango/pefd.

---

### MetaState: Persistent Working Memory Enhances Reasoning in Discrete Diffusion Language Models

Authors: Kejing Xia, Mingzhe Li, Lixuan Wei, Zhenbang Du, Xiangchi Yuan, Dachuan Shi, Qirui Jin, Wenke Lee
Published: 2026-03-02
Categories: cs.CL, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2603.01331v3

Discrete diffusion language models (dLLMs) generate text by iteratively denoising a masked sequence. However, standard dLLMs condition each denoising step solely on the current hard-masked sequence, while intermediate continuous representations are discarded after sampling and remasking. We term this bottleneck the \textbf{Information Island} issue: continuous information remains isolated within individual denoising steps and fails to propagate across the trajectory. This bottleneck is especially harmful for reasoning, which requires intermediate reasoning state to be preserved and updated across many denoising steps. To address this limitation, we introduce \textbf{MetaState}, a lightweight recurrent augmentation that equips a frozen dLLM backbone with persistent, fixed-size working memory. MetaState comprises three modules with a shared time conditioner: a cross-attention \textbf{Mixer} that reads backbone activations into memory slots, a GRU-style \textbf{Updater} that integrates information across steps, and a cross-attention \textbf{Injector} that writes the updated memory back into the backbone. We train these modules with a dedicated $K$-step unrolling pipeline to learn multi-step dynamics. MetaState adds only ${\sim}0.6\%$ trainable parameters while keeping the backbone frozen, and consistently improves reasoning performance over frozen baselines on mathematical reasoning and code generation benchmarks, with an average gain of 4.5 percentage points across all evaluations.

---

### Mixture of Sequence: Theme-Aware Mixture-of-Experts for Long-Sequence Recommendation

Authors: Xiao Lin, Zhicheng Tang, Weilin Cong, Mengyue Hang, Kai Wang, Yajuan Wang, Zhichen Zeng, Ting-Wei Li, Hyunsik Yoo, Zhining Liu, Xuying Ning, Ruizhong Qiu et al.
Published: 2026-03-01
Categories: cs.IR, cs.AI
arXiv: https://arxiv.org/abs/2604.20858v1

Sequential recommendation has rapidly advanced in click-through rate prediction due to its ability to model dynamic user interests. A key challenge, however, lies in modeling long sequences: users often exhibit significant interest shifts, introducing substantial irrelevant or misleading information. Our empirical analysis corroborates this challenge and uncovers a recurring behavioral pattern in long sequences (\textit{session hopping}): user interests remain stable within short temporal spans (\textit{sessions}) but shift drastically across sessions and may reappear after multiple sessions. To address this challenge, we propose the Mixture of Sequence (MoS) framework, a model-agnostic MoE approach that achieves accurate predictions by extracting theme-specific and multi-scale subsequences from noisy raw user sequences. First, MoS employs a theme-aware routing mechanism to adaptively learn the latent themes of user sequences and organizes these sequences into multiple coherent subsequences. Each subsequence contains only sessions aligned with a specific theme, thereby effectively filtering out irrelevant or even misleading information introduced by user interest shifts in session hopping. In addition, to alleviate potential information loss, we introduce a multi-scale fusion mechanism, which leverages three types of experts to capture global sequence characteristics, short-term user behaviors, and theme-specific semantic patterns. Together, these two mechanisms endow MoS with the ability to deliver accurate recommendations from multi-faceted and multi-scale perspectives. Experimental results demonstrate that MoS consistently achieves the SOTA performance while introducing fewer FLOPs compared with other MoE counterparts, providing strong evidence of its excellent balance between utility and efficiency. The code is available at https://github.com/xiaolin-cs/MoS.

---

### Anthropomorphism and Trust in Human-Large Language Model interactions

Authors: Akila Kadambi, Ylenia D'Elia, Tanishka Shah, Iulia Comsa, Alison Lentz, Katie Siri-Ngammuang, Tara Buechler, Jonas Kaplan, Antonio Damasio, Srini Narayanan, Lisa Aziz-Zadeh
Published: 2026-03-01
Categories: cs.HC, cs.AI
arXiv: https://arxiv.org/abs/2604.15316v1

With large language models (LLMs) becoming increasingly prevalent in daily life, so too has the tendency to attribute to them human-like minds and emotions, or anthropomorphize them. Here, we investigate dimensions people use to anthropomorphize and attribute trust toward LLMs across more than 2,000 human-LLM interactions. Participants (N=115) engaged with LLM chatbots systematically varied in warmth (friendliness), competence (capability, coherence), and empathy (cognitive and affective). Warmth and cognitive empathy significantly predicted perceptions on all outcomes (perceived anthropomorphism, trust, similarity, relational closeness, frustration, usefulness), while competence predicted all outcomes except for anthropomorphism. Affective empathy primarily predicted perceived relational measures, but did not predict the epistemic outcomes. Topic sub-analyses showed that more subjective, personally relevant topics (e.g., relationship advice) amplified these effects, producing greater human-likeness and relational connection with the LLM than did objective topics. Together, these findings reveal that warmth, competence, and empathy are key dimensions through which people attribute relational and epistemic perceptions to artificial agents.

---

### Modeling and Controlling Deployment Reliability under Temporal Distribution Shift

Authors: Naimur Rahman, Naazreen Tabassum
Published: 2026-03-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2604.02351v1

Machine learning models deployed in non-stationary environments are exposed to temporal distribution shift, which can erode predictive reliability over time. While common mitigation strategies such as periodic retraining and recalibration aim to preserve performance, they typically focus on average metrics evaluated at isolated time points and do not explicitly model how reliability evolves during deployment. We propose a deployment-centric framework that treats reliability as a dynamic state composed of discrimination and calibration. The trajectory of this state across sequential evaluation windows induces a measurable notion of volatility, allowing deployment adaptation to be formulated as a multi-objective control problem that balances reliability stability against cumulative intervention cost. Within this framework, we define a family of state-dependent intervention policies and empirically characterize the resulting cost-volatility Pareto frontier. Experiments on a large-scale, temporally indexed credit-risk dataset (1.35M loans, 2007-2018) show that selective, drift-triggered interventions can achieve smoother reliability trajectories than continuous rolling retraining while substantially reducing operational cost. These findings position deployment reliability under temporal shift as a controllable multi-objective system and highlight the role of policy design in shaping stability-cost trade-offs in high-stakes tabular applications.

---

### Enhancing Safety of Large Language Models via Embedding Space Separation

Authors: Xu Zhao, Xiting Wang, Weiran Shen
Published: 2026-03-01
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2603.20206v1

Large language models (LLMs) have achieved impressive capabilities, yet ensuring their safety against harmful prompts remains a critical challenge. Recent work has revealed that the latent representations (embeddings) of harmful and safe queries in LLMs typically exhibit linear separability, a property that has been exploited to construct attacks by perturbing the embeddings of harmful queries towards the safe subspace. Motivated by this observation, we propose a representation-level fine-tuning approach, named Embedding Space Separation (ES2), which improves LLM safety by explicitly enlarging the distance between harmful and safe representations in the embedding space. To prevent degradation of model's general capabilities, we introduce a Kullback-Leibler (KL) divergence regularization term into the loss function, which constrains the logits of the fine-tuned model to align with those of the original base model on harmless inputs. We evaluate our method on several open-source LLMs using standard safety benchmarks. Extensive experimental results demonstrate that our approach substantially improves model safety while maintaining comparable general capabilities.

---

### From Feature-Based Models to Generative AI: Validity Evidence for Constructed Response Scoring

Authors: Jodi M. Casabianca, Daniel F. McCaffrey, Matthew S. Johnson, Naim Alper, Vladimir Zubenko
Published: 2026-03-01
Categories: cs.CL, cs.AI, cs.CY
arXiv: https://arxiv.org/abs/2603.19280v1

The rapid advancements in large language models and generative artificial intelligence (AI) capabilities are making their broad application in the high-stakes testing context more likely. Use of generative AI in the scoring of constructed responses is particularly appealing because it reduces the effort required for handcrafting features in traditional AI scoring and might even outperform those methods. The purpose of this paper is to highlight the differences in the feature-based and generative AI applications in constructed response scoring systems and propose a set of best practices for the collection of validity evidence to support the use and interpretation of constructed response scores from scoring systems using generative AI. We compare the validity evidence needed in scoring systems using human ratings, feature-based natural language processing AI scoring engines, and generative AI. The evidence needed in the generative AI context is more extensive than in the feature-based scoring context because of the lack of transparency and other concerns unique to generative AI such as consistency. Constructed response score data from a large corpus of independent argumentative essays written by 6-12th grade students demonstrate the collection of validity evidence for different types of scoring systems and highlight the numerous complexities and considerations when making a validity argument for these scores.

---

### Multilingual Hate Speech Detection and Counterspeech Generation: A Comprehensive Survey and Practical Guide

Authors: Zahra Safdari Fesaghandis, Suman Kalyan Maity
Published: 2026-03-01
Categories: cs.CL
arXiv: https://arxiv.org/abs/2603.19279v1

Combating online hate speech in multilingual settings requires approaches that go beyond English-centric models and capture the cultural and linguistic diversity of global online discourse. This paper presents a comprehensive survey and practical guide to multilingual hate speech detection and counterspeech generation, integrating recent advances in natural language processing. We analyze why monolingual systems often fail in non-English and code-mixed contexts, missing implicit hate and culturally specific expressions. To address these challenges, we outline a structured three-phase framework - task design, data curation, and evaluation - drawing on state-of-the-art datasets, models, and metrics. The survey consolidates progress in multilingual resources and techniques while highlighting persistent obstacles, including data scarcity in low-resource languages, fairness and bias in system development, and the need for multimodal solutions. By bridging technical progress with ethical and cultural considerations, we provide researchers, practitioners, and policymakers with scalable guidelines for building context-aware, inclusive systems. Our roadmap contributes to advancing online safety through fairer, more effective detection and counterspeech generation across diverse linguistic environments.

---

### HypeLoRA: Hyper-Network-Generated LoRA Adapters for Calibrated Language Model Fine-Tuning

Authors: Bartosz Trojan, Filip Gębala
Published: 2026-03-01
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2603.19278v2

Modern Transformer-based models frequently suffer from miscalibration, producing overconfident predictions that do not reflect true empirical frequencies. This work investigates the calibration dynamics of LoRA: Low-Rank Adaptation and a novel hyper-network-based adaptation framework as parameter-efficient alternatives to full fine-tuning for RoBERTa. Evaluating across the GLUE benchmark, we demonstrate that LoRA-based adaptation consistently achieves calibration parity with (and in specific tasks exceeds) full fine-tuning, while maintaining significantly higher parameter efficiency. We further explore a dynamic approach where a shared hyper-network generates LoRA factors (A and B matrices) to induce structural coupling across layers. This approach produced results similar to standard LoRA fine-tuning, even achieving better MCC on CoLA dataset. Our study also reveal a critical trade-off: constraining the adaptation space (e.g., freezing matrices A) acts as a powerful regularizer that enhances Expected Calibration Error (ECE), but necessitates a carefully balanced sacrifice in downstream task accuracy. To support future research, we provide a unified and reproducible implementation of contemporary calibration metrics, including ECE, MCE, and ACE. Our findings clarify the relationship between parameter efficiency and probabilistic reliability, positioning structured low-rank updates as a viable foundation for uncertainty-aware Transformer architectures. Code available at: https://github.com/btrojan-official/HypeLoRA

---

### MOSAIC: Modular Opinion Summarization using Aspect Identification and Clustering

Authors: Piyush Kumar Singh, Jayesh Choudhari
Published: 2026-03-01
Categories: cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2603.19277v1

Reviews are central to how travelers evaluate products on online marketplaces, yet existing summarization research often emphasizes end-to-end quality while overlooking benchmark reliability and the practical utility of granular insights. To address this, we propose MOSAIC, a scalable, modular framework designed for industrial deployment that decomposes summarization into interpretable components, including theme discovery, structured opinion extraction, and grounded summary generation. We validate the practical impact of our approach through online A/B tests on live product pages, showing that surfacing intermediate outputs improves customer experience and delivers measurable value even prior to full summarization deployment. We further conduct extensive offline experiments to demonstrate that MOSAIC achieves superior aspect coverage and faithfulness compared to strong baselines for summarization. Crucially, we introduce opinion clustering as a system-level component and show that it significantly enhances faithfulness, particularly under the noisy and redundant conditions typical of user reviews. Finally, we identify reliability limitations in the standard SPACE dataset and release a new open-source tour experience dataset (TRECS) to enable more robust evaluation.

---

### ICPRL: Acquiring Physical Intuition from Interactive Control

Authors: Xinrun Xu, Pi Bu, Ye Wang, Börje F. Karlsson, Ziming Wang, Tengtao Song, Qi Zhu, Jun Song, Shuo Zhang, Zhiming Ding, Bo Zheng
Published: 2026-03-01
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2603.13295v1

VLMs excel at static perception but falter in interactive reasoning in dynamic physical environments, which demands planning and adaptation to dynamic outcomes. Existing physical reasoning methods often depend on abstract symbolic inputs or lack the ability to learn and adapt from direct, pixel-based visual interaction in novel scenarios. We introduce ICPRL (In-Context Physical Reinforcement Learning), a framework inspired by In-Context Reinforcement Learning (ICRL) that empowers VLMs to acquire physical intuition and adapt their policies in-context. Our approach trains a vision-grounded policy model via multi-turn Group Relative Policy Optimization (GRPO) over diverse multi-episode interaction histories. This enables the agent to adapt strategies by conditioning on past trial-and-error sequences, without requiring any weight updates. This adaptive policy works in concert with a separately trained world model that provides explicit physical reasoning by predicting the results of potential actions. At inference, the policy proposes candidate actions, while the world model predicts outcomes to guide a root-node PUCT search to select the most promising action. Evaluated on the diverse physics-based puzzle-solving tasks in the DeepPHY benchmark, ICPRL demonstrates significant improvements across both its I. policy-only, and II. world-model-augmented stages. Notably, these gains are retained in unseen physical environments, demonstrating that our framework facilitates genuine in-context acquisition of the environment's physical dynamics from interactive experience.

---

### A Principle-Driven Adaptive Policy for Group Cognitive Stimulation Dialogue for Elderly with Cognitive Impairment

Authors: Jiyue Jiang, Yanyu Chen, Pengan Chen, Kai Liu, Jingqi Zhou, Zheyong Zhu, He Hu, Fei Ma, Qi Tian, Chuan Wu
Published: 2026-03-01
Categories: cs.CL
arXiv: https://arxiv.org/abs/2603.10034v1

Cognitive impairment is becoming a major public health challenge. Cognitive Stimulation Therapy (CST) is an effective intervention for cognitive impairment, but traditional methods are difficult to scale, and existing digital systems struggle with group dialogues and cognitive stimulation principles. While Large Language Models (LLMs) are powerful, their application in this context faces key challenges: cognitive stimulation dialogue paradigms, a lack of therapeutic reasoning, and static-only user modeling. To address these issues, we propose a principle-driven adaptive policy actualized through a Group Cognitive Stimulation Dialogue (GCSD) system. We first construct a dataset with over 500 hours of real-world CST conversations and 10,000+ simulated dialogues generated via our Principle-Guided Scenario Simulation strategy. Our GCSD system then integrates four core modules to overcome LLM limitations: (i) a multi-speaker context controller to resolve role confusion; (ii) dynamic participant cognitive state modeling for personalized interaction; (iii) a cognitive stimulation-focused attention loss to instill cognitive stimulation reasoning; and (iv) a multi-dimensional reward strategy to enhance response value. Experimental results demonstrate that GCSD significantly outperforms baseline models across various evaluation metrics. Future work will focus on long-term clinical validation to bridge the gap between computational performance and clinical efficacy.

---

### Zipage: Maintain High Request Concurrency for LLM Reasoning through Compressed PagedAttention

Authors: Mengqi Liao, Lu Wang, Chaoyun Zhang, Bo Qiao, Si Qin, Qingwei Lin, Saravan Rajmohan, Dongmei Zhang, Huaiyu Wan
Published: 2026-03-01
Categories: cs.DC, cs.AI
arXiv: https://arxiv.org/abs/2603.08743v1

With reasoning becoming the generative paradigm for large language models (LLMs), the memory bottleneck caused by KV cache during the decoding phase has become a critical factor limiting high-concurrency service. Although existing KV cache eviction methods address the memory issue, most of them are impractical for industrial-grade applications. This paper introduces Compressed PagedAttention, a method that combines token-wise KV cache eviction with PagedAttention. We propose a comprehensive scheduling strategy and support prefix caching and asynchronous compression for Compressed PagedAttention. Based on this, we have developed a high-concurrency LLM inference engine, Zipage. On large-scale mathematical reasoning tasks, Zipage achieves around 95\% of the performance of Full KV inference engines while delivering over 2.1$\times$ speedup.

---

### Approximate Nearest Neighbor Search for Modern AI: A Projection-Augmented Graph Approach

Authors: Kejing Lu, Zhenpeng Pan, Jianbin Qin, Yoshiharu Ishikawa, Chuan Xiao
Published: 2026-03-01
Categories: cs.IR, cs.DB, cs.LG
arXiv: https://arxiv.org/abs/2603.06660v2

Approximate Nearest Neighbor Search (ANNS) is fundamental to modern AI applications. Most existing solutions optimize query efficiency but fail to align with the practical requirements of modern workloads. In this paper, we outline six critical demands of modern AI applications: high query efficiency, fast indexing, low memory footprint, scalability to high dimensionality, robustness across varying retrieval sizes, and support for online insertions. To satisfy all these demands, we introduce Projection-Augmented Graph (PAG), a new ANNS framework that integrates projection techniques into a graph index. PAG reduces unnecessary exact distance computations through asymmetric comparisons between exact and approximate distances as guided by projection-based statistical tests. Three key components are designed and integrated into the graph index to optimize indexing and searching. Experiments on six modern datasets demonstrate that PAG consistently achieves superior queries per second (QPS)-recall performance -- up to 5x faster than HNSW -- while offering fast indexing speed and moderate memory footprint. PAG remains robust as dimensionality and retrieval size increase and naturally supports online insertions.

---

### Science Literacy: Generative AI as Enabler of Coherence in the Teaching, Learning, and Assessment of Scientific Knowledge and Reasoning

Authors: Xiaoming Zhai, James W. Pellegrino, Matias Rojas, Jongchan Park, Matthew Nyaaba, Clayton Cohn, Gautam Biswas
Published: 2026-03-01
Categories: cs.CY, cs.AI
arXiv: https://arxiv.org/abs/2603.06659v1

This chapter examines the potential of generative AI in enhancing science literacy across the K-16+ grade span, including its benefits as well as the conceptual and practical challenges that doing so presents. It begins with a discussion of what defines science literacy in the era of AI, including how AI has changed science and the demand for future citizens to be scientifically literate when AI is applied in their careers and lives. The chapter further discusses why science literacy presents such a challenge in K-16+ educational settings. It then develops an argument for the type of architecture needed for AI to assist in solving the problem by bringing coherence to the teaching, learning, and assessment of science knowledge and reasoning. Components of this architecture are illustrated with respect to the AI tools and capabilities needed for design and implementation. The chapter concludes with a consideration of what has been learned regarding both science literacy and AI, as well as what remains to be learned, including the research and development (R&D) needed, and the generalizability of this science literacy case to other disciplinary learning and knowledge domains.

---

### ASMIL: Attention-Stabilized Multiple Instance Learning for Whole Slide Imaging

Authors: Linfeng Ye, Shayan Mohajer Hamidi, Zhixiang Chi, Guang Li, Mert Pilanci, Takahiro Ogawa, Miki Haseyama, Konstantinos N. Plataniotis
Published: 2026-03-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.06658v1

Attention-based multiple instance learning (MIL) has emerged as a powerful framework for whole slide image (WSI) diagnosis, leveraging attention to aggregate instance-level features into bag-level predictions. Despite this success, we find that such methods exhibit a new failure mode: unstable attention dynamics. Across four representative attention-based MIL methods and two public WSI datasets, we observe that attention distributions oscillate across epochs rather than converging to a consistent pattern, degrading performance. This instability adds to two previously reported challenges: overfitting and over-concentrated attention distribution. To simultaneously overcome these three limitations, we introduce attention-stabilized multiple instance learning (ASMIL), a novel unified framework. ASMIL uses an anchor model to stabilize attention, replaces softmax with a normalized sigmoid function in the anchor to prevent over-concentration, and applies token random dropping to mitigate overfitting. Extensive experiments demonstrate that ASMIL achieves up to a 6.49\% F1 score improvement over state-of-the-art methods. Moreover, integrating the anchor model and normalized sigmoid into existing attention-based MIL methods consistently boosts their performance, with F1 score gains up to 10.73\%. All code and data are publicly available at https://github.com/Linfeng-Ye/ASMIL.

---

### GameVerse: Can Vision-Language Models Learn from Video-based Reflection?

Authors: Kuan Zhang, Dongchen Liu, Qiyue Zhao, Jinkun Hou, Xinran Zhang, Qinlei Xie, Miao Liu, Yiming Li
Published: 2026-03-01
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2603.06656v2

Human gameplay is a visually grounded interaction loop in which players act, reflect on failures, and watch tutorials to refine strategies. Can Vision-Language Models (VLMs) also learn from video-based reflection? We present GameVerse, a comprehensive video game benchmark that enables a reflective visual interaction loop. Moving beyond traditional fire-and-forget evaluations, it uses a novel reflect-and-retry paradigm to assess how VLMs internalize visual experience and improve policies. To facilitate systematic and scalable evaluation, we also introduce a cognitive hierarchical taxonomy spanning 15 globally popular games, dual action space for both semantic and GUI control, and milestone evaluation using advanced VLMs to quantify progress. Our experiments show that VLMs benefit from video-based reflection in varied settings, and perform best by combining failure trajectories and expert tutorials-a training-free analogue to reinforcement learning (RL) plus supervised fine-tuning (SFT).Our project page is available at https://gameverse-bench.github.io/ . Our code is available at https://github.com/THUSI-Lab/GameVerse .

---

### ProFocus: Proactive Perception and Focused Reasoning in Vision-and-Language Navigation

Authors: Wei Xue, Mingcheng Li, Xuecheng Wu, Jingqun Tang, Dingkang Yang, Lihua Zhang
Published: 2026-03-01
Categories: cs.RO, cs.CV
arXiv: https://arxiv.org/abs/2603.05530v2

Vision-and-Language Navigation (VLN) requires agents to accurately perceive complex visual environments and reason over navigation instructions and histories. However, existing methods passively process redundant visual inputs and treat all historical contexts indiscriminately, resulting in inefficient perception and unfocused reasoning. To address these challenges, we propose \textbf{ProFocus}, a training-free progressive framework that unifies \underline{Pro}active Perception and \underline{Focus}ed Reasoning through collaboration between large language models (LLMs) and vision-language models (VLMs). For proactive perception, ProFocus transforms panoramic observations into structured ego-centric semantic maps, enabling the orchestration agent to identify missing visual information needed for reliable decision-making, and to generate targeted visual queries with corresponding focus regions that guide the perception agent to acquire the required observations. For focused reasoning, we propose Branch-Diverse Monte Carlo Tree Search (BD-MCTS) to identify top-$k$ high-value waypoints from extensive historical candidates. The decision agent focuses reasoning on the historical contexts associated with these waypoints, rather than considering all historical waypoints equally. Extensive experiments validate the effectiveness of ProFocus, achieving state-of-the-art performance among zero-shot methods on R2R and REVERIE benchmarks.

---

### Quantifying Frontier LLM Capabilities for Container Sandbox Escape

Authors: Rahul Marchand, Art O Cathain, Jerome Wynne, Philippos Maximos Giavridis, Stuart Jennings, Freddy Tuxworth, Tolga H. Dur, Sam Deverett, John Wilkinson, Jason Gwartz, Harry Coppock
Published: 2026-03-01
Categories: cs.CR, cs.AI
arXiv: https://arxiv.org/abs/2603.02277v3

Large language models (LLMs) increasingly act as autonomous agents, using tools to execute code, read and write files, and access networks, creating novel security risks. To mitigate these risks, agents are commonly deployed and evaluated in isolated "sandbox" environments, often implemented using Docker/OCI containers. We introduce SANDBOXESCAPEBENCH, an open benchmark that safely measures an LLM's capacity to break out of these sandboxes. The benchmark is implemented as an Inspect AI Capture the Flag (CTF) evaluation utilising a nested sandbox architecture with the outer layer containing the flag and no known vulnerabilities. Following a threat model of a motivated adversarial agent with shell access inside a container, SANDBOXESCAPEBENCH covers a spectrum of sandboxescape mechanisms spanning misconfiguration, privilege allocation mistakes, kernel flaws, and runtime/orchestration weaknesses. We find that, when vulnerabilities are added, LLMs are able to identify and exploit them, showing that use of evaluation like SANDBOXESCAPEBENCH is needed to ensure sandboxing continues to provide the encapsulation needed for highly-capable models.

---

### A Comparative Study of UMAP and Other Dimensionality Reduction Methods

Authors: Guanzhe Zhang, Shanshan Ding, Zhezhen Jin
Published: 2026-03-01
Categories: cs.LG, stat.AP, stat.ML
arXiv: https://arxiv.org/abs/2603.02275v2

Uniform Manifold Approximation and Projection (UMAP) is a widely used manifold learning technique for dimensionality reduction. This paper studies UMAP, supervised UMAP, and several competing dimensionality reduction methods, including Principal Component Analysis (PCA), Kernel PCA, Sliced Inverse Regression (SIR), Kernel SIR, and t-distributed Stochastic Neighbor Embedding, through a comprehensive comparative analysis. Although UMAP has attracted substantial attention for preserving local and global structures, its supervised extensions, particularly for regression settings, remain rather underexplored. We provide a systematic evaluation of supervised UMAP for both regression and classification using simulated and real datasets, with performance assessed via predictive accuracy on low-dimensional embeddings. Our results show that supervised UMAP performs well for classification but exhibits limitations in effectively incorporating response information for regression, highlighting an important direction for future development.

---

### Contextual Invertible World Models: A Neuro-Symbolic Agentic Framework for Colorectal Cancer Drug Response

Authors: Christopher Baker, Tianyu Ren, Karen Rafferty, Hui Wang
Published: 2026-03-01
Categories: q-bio.QM, cs.AI
arXiv: https://arxiv.org/abs/2603.02274v3

Precision oncology is currently limited by the small-N, large-P paradox, where high-dimensional genomic data is abundant but pharmacological response samples are sparse. While deep learning achieves predictive accuracy, it frequently fails to provide the mechanistic clarity required for clinical adoption. We present the Contextual Invertible World Model (CIWM), a Neuro-Symbolic Agentic Framework that bridges this gap by integrating a quantitative machine learning emulator with a Large Language Model reasoning layer. Utilising a stringently curated, high-fidelity data engineering pipeline on the Sanger GDSC dataset (\( N=83 \)), we isolate true biological signals from in vitro artifacts to establish a rigorous baseline predictive correlation for complex transcriptomics (\( r=0.268 \)). Through Inverse Reasoning, we perform in silico CRISPR perturbations across the colorectal landscape. The framework autonomously overturns classical mechanistic assumptions, identifying a hierarchical dominance of mutant KRAS over the APC/Wnt-axis in driving 5-fluorouracil resistance (\( Δ=-0.0469 \)) via a "KRAS Shield" mapped to MAPK/PI3K networks. Furthermore, the agentic layer identified a "PIK3CA Paradox", revealing that repairing PIK3CA inadvertently increases chemoresistance (\( Δ=+0.0085 \)) by triggering a compensatory feedback loop that hyperactivates the dominant MAPK survival pathway.

---

### Graph Attention Based Prioritization of Disease Responsible Genes from Multimodal Alzheimer's Network

Authors: Binon Teji, Subhajit Bandyopadhyay, Swarup Roy
Published: 2026-03-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.02273v1

Prioritizing disease-associated genes is central to understanding the molecular mechanisms of complex disorders such as Alzheimer's disease (AD). Traditional network-based approaches rely on static centrality measures and often fail to capture cross-modal biological heterogeneity. We propose NETRA (Node Evaluation through Transformer-based Representation and Attention), a multimodal graph transformer framework that replaces heuristic centrality metrics with attention-driven relevance scoring. Using AD as a case study, gene regulatory networks are independently constructed from microarray, single-cell RNA-seq, and single-nucleus RNA-seq data. Random-walk sequences derived from these networks are used to train a BERT-based model for learning global gene embeddings, while modality-specific gene expression profiles are compressed using variational autoencoders. These representations are integrated with auxiliary biological networks, including protein-protein interactions, Gene Ontology semantic similarity, and diffusion-based gene similarity, into a unified multimodal graph. A graph transformer assigns NETRA scores that quantify gene relevance in a disease-specific and context-aware manner. Gene set enrichment analysis shows that NETRA achieves a normalized enrichment score of about 3.9 for the Alzheimer's disease pathway, substantially outperforming classical centrality measures and diffusion models. Top-ranked genes enrich multiple neurodegenerative pathways, recover a known late-onset AD susceptibility locus at chr12q13, and reveal conserved cross-disease gene modules. The framework preserves biologically realistic heavy-tailed network topology and is readily extensible to other complex disorders.

---

### Characterizing VLA Models: Identifying the Action Generation Bottleneck for Edge AI Architectures

Authors: Manoj Vishwanathan, Suvinay Subramanian, Anand Raghunathan
Published: 2026-03-01
Categories: cs.PF, cs.AI, cs.AR, cs.RO
arXiv: https://arxiv.org/abs/2603.02271v1

Vision-Language-Action (VLA) models are an emerging class of workloads critical for robotics and embodied AI at the edge. As these models scale, they demonstrate significant capability gains, yet they must be deployed locally to meet the strict latency requirements of real-time applications. This paper characterizes VLA performance on two generations of edge hardware, viz. the Nvidia Jetson Orin and Thor platforms. Using MolmoAct-7B, a state-of-the-art VLA model, we identify a primary execution bottleneck: up to 75% of end-to-end latency is consumed by the memory-bound action-generation phase. Through analytical modeling and simulations, we project the hardware requirements for scaling to 100B parameter models. We also explore the impact of high-bandwidth memory technologies and processing-in-memory (PIM) as promising future pathways in edge systems for embodied AI.

---

### You Only Need One Stage: Novel-View Synthesis From A Single Blind Face Image

Authors: Taoyue Wang, Xiang Zhang, Xiaotian Li, Huiyuan Yang, Lijun Yin
Published: 2026-03-01
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2603.01328v1

We propose a novel one-stage method, NVB-Face, for generating consistent Novel-View images directly from a single Blind Face image. Existing approaches to novel-view synthesis for objects or faces typically require a high-resolution RGB image as input. When dealing with degraded images, the conventional pipeline follows a two-stage process: first restoring the image to high resolution, then synthesizing novel views from the restored result. However, this approach is highly dependent on the quality of the restored image, often leading to inaccuracies and inconsistencies in the final output. To address this limitation, we extract single-view features directly from the blind face image and introduce a feature manipulator that transforms these features into 3D-aware, multi-view latent representations. Leveraging the powerful generative capacity of a diffusion model, our framework synthesizes high-quality, consistent novel-view face images. Experimental results show that our method significantly outperforms traditional two-stage approaches in both consistency and fidelity.

---