C E Wu

Jung-Hoon Cho, Cathy Wu · mit

Transit network design depends not only on the optimization algorithm but also on who shows up to the public hearing. Current practice often collects one-directional comments from self-selected attendees, leaving participant mix as an uncontrolled source of outcome variation. We present AGORA, a framework that holds the network, demand, and solver fixed while systematically varying meeting composition through stakeholder agents, structured deliberation, and governance gates. Across two standard benchmark networks at different scales, we find that (i) aggregate outcomes vary little across compositions, but on tail risk and fairness disparity, representative sampling still tends to outperform skewed compositions; (ii) without deliberation, composition produces no variation at all, showing that deliberation is the mechanism through which who attends affects outcomes; and (iii) governance gates compress cross-profile variance without shifting the average outcome on Mandl, but low acceptance on Mumford0 shows thresholds require instance-specific calibration. These findings reframe participation bias from an uncontrollable input to a process-design problem: even without guaranteed representative attendance, well-structured deliberation and governance criteria can substantially reduce how much outcomes depend on who is in the room.

Haoqi Yuan, Zhixuan Liang, Anzhe Chen et al.

Foundation models in language and multimodality achieve strong generalization by aligning heterogeneous data under a unified formulation and training at scale. In this report, we investigate whether this scaling recipe can be applied to robotic manipulation to achieve genuine generalization. This is challenging because, unlike text, manipulation data is heterogeneous by nature, expensive to collect, and narrow in diversity, making alignment and scale simultaneously difficult. We present Qwen-RobotManip, a generalizable Vision-Language-Action foundation model built on Qwen-VL. Qwen-RobotManip introduces a unified alignment framework across the representation, motion, and behavioral dimensions of manipulation, making large-scale multi-source training coherent rather than conflicting. This alignment capability in turn enables Qwen-RobotManip to absorb manipulation data at a scale that prior training regimes could not sustain. A human-to-robot synthesis pipeline converts egocentric hand demonstrations into robot trajectories across 15 platforms, and a rigorous curation pipeline harmonizes heterogeneous datasets. Using only open-source datasets and human videos without proprietary data collection, Qwen-RobotManip constructs a ~38,100-hour pretraining corpus and exhibits emergent generalization capabilities, including zero-shot instruction following, robustness to perturbations, reactive error recovery, and cross-embodiment transfer. We find that standard benchmarks fail to capture pretraining quality and instead adopt OOD settings including RoboCasa365, LIBERO-Plus, EBench, RoboTwin-Clean2Rand, RoboTwin-IF, and RoboTwin-XE. Qwen-RobotManip substantially outperforms prior state-of-the-art models, including $π$0.5, across all OOD settings, ranks 1st in RoboChallenge with a 20% relative improvement, and is validated on real-robot platforms including AgileX ALOHA, Franka, UR, and ARX.

Aiwei Liu, Cheng Shi, Chuhan Wu et al.

Scaling Large Language Models (LLMs) has been driven mainly by enlarging the Transformer backbone, but for an already-strong model this requires another round of costly pretraining. We study whether an existing backbone can keep improving by allocating more computation to each token while leaving the Transformer backbone fixed. Depth-recurrent (looped) Transformers pursue this goal but are hard to scale, because looped computation does not fit naturally with the pipeline parallelism used to train the largest models. We add computation along the sequence-length dimension, where the extra computation is simply a longer input and stays compatible with standard large-model training. We propose Hidden Decoding, a sequence-length scaling method applied during continued pretraining (CPT). It expands each token into n streams with independent embedding tables and keeps the intermediate streams' key-value cache as context, so each token performs more internal computation without adding or widening Transformer layers. To keep this affordable at scale, we introduce Stream-Factorized Attention, in which most layers attend only within each stream and only a few layers mix across streams, reducing the attention cost from quadratic to roughly linear in n. Experiments support two scaling results. At frontier scale, we train WeLM-HD4-80B and WeLM-HD4-617B at n=4 and improve their matched non-HD baselines, making Hidden Decoding the first demonstrated sequence-length scaling method at the 100B+ MoE scale. Across expansion factors, the gains grow as n increases, showing that sequence-length expansion is a practical fixed-backbone scaling path for frontier-scale LLMs.

Yonggan Fu, Lexington Whalen, Abhinav Garg et al.

We introduce Nemotron-Labs-Diffusion, a tri-mode language model (LM) that unifies AR, diffusion, and self-speculation decoding within a single architecture. Trained with a joint AR-diffusion objective, Nemotron-Labs-Diffusion can switch modes to sustain high throughput across deployment settings and concurrency levels. Our study shows that (1) AR and diffusion objectives are complementary: diffusion improves lookahead planning, while AR provides left-to-right linguistic priors. (2) In self-speculation mode, diffusion drafts while AR verifies, outperforming multi-token prediction (MTP) methods in both acceptance rate and real-device efficiency. (3) A speed-of-light analysis further demonstrates diffusion's long-term potential, with up to 76.5% more tokens per forward pass than self-speculation under an optimal sampler. Scaling to 3B, 8B, and 14B parameters, our Nemotron-Labs-Diffusion family, including base, instruct, and vision-language models, consistently outperforms state-of-the-art open-source AR and diffusion LMs in both accuracy and speed. For example, Nemotron-Labs-Diffusion-8B decodes 6x more tokens per forward than Qwen3-8B with comparable accuracy, translating to 4x higher throughput on SPEED-Bench with SGLang on a GB200 GPU.

Dongfang Li, Xiaodong Luo, Ruoyu Sun et al.

Full-parameter post-training of trillion-parameter-scale MoE models introduces substantial system-level challenges for large-scale distributed training, including severe memory pressure, non-overlapped communication overhead, and inefficient kernel execution. While most large-scale LLM training systems are built around GPU-based clusters, this report presents an end-to-end optimization practice on the Ascend NPU SuperPOD. Using the DeepSeek-V4 model family as the target workload, we develop a hierarchical optimization framework spanning model-level parallelism, computation-communication orchestration, and low-level kernel execution. The resulting system achieves 34.22% Model FLOPs Utilization (MFU) with a 2.93x improvement over the open-source baseline recipe while maintaining training stability. Building on this optimized infrastructure, we further establish a CPT and SFT workflow for complex Operations Research (OR) tasks. We refer to the integrated framework as SLAI T-Rex. Using DeepSeek-V4-Flash, we develop OR-oriented CPT and SFT data pipelines that combine collected domain resources with solver-verified synthetic optimization documents. The resulting dataset contains 10K high-quality SFT samples spanning four task categories and three problem representations. The specialized model achieves the highest average zero-shot Pass@1 score among the evaluated models, reaching 71.81% and outperforming GPT-5.4-Mini and the base DeepSeek-V4-Flash model by 3.98 and 11.27 percentage points, respectively. Overall, this work demonstrates a full-stack pathway from efficient trillion-parameter model post-training on Ascend infra to domain-specialized Flash models for solver-grounded mathematical modeling, advancing frontier-model systems for complex reasoning.

Guli Zhu, Chenwei Wu, Liyue Shen

Model editing promises a fast, targeted way to correct post-deployment mistakes in medical vision-language models (VLMs) without costly retraining. However, existing multimodal model editing benchmarks focus on general-purpose tasks and do not reflect realistic clinical domain requirements and variability. To address this, we introduce M3Bench, a clinically grounded benchmark for multimodal model editing that evaluates whether an edit remains reliable, precise, and generalizable under the challenges of image and text variation, modality and protocol shifts, clinical knowledge composition, and temporal progression. M3Bench contains 16,276 questions spanning diverse anatomy, modalities, and specialties, and supports both single and sequential edits. By evaluating 4 representative editors across 6 medical and general VLMs, we find that no method excels across all criteria. Gradient-based editors achieve strong transfer but suffer from catastrophic locality violations, whereas memory-based methods preserve locality but lack compositional generality and exhibit high backbone-dependent hyperparameter sensitivity. We further attribute these failures to the latent space geometry of VLMs and how different editing methods shift its landscape. Overall, M3Bench establishes a rigorous clinical stress test for multimodal model editing and offers actionable guidance for safer post-deployment adaptation. The benchmark is publicly available at https://github.com/BioMed-AI-Lab-U-Michgan/M3Bench .

Chris Xing Tian, Chengkai Wu, Ziyu Wang et al.

Large language model (LLM)-based lossless image compression methods typically represent pixel data through the native text interface of a pretrained model, converting pixel values into token sequences that the LLM processes through its vocabulary head. This design shows that pretrained language models can provide probability estimates for image coding, but it also couples compression to tokenizer behavior, vocabulary-specific numeric tokens, and model-family-specific adaptation. In this paper, we present LUMI (LLM-based Unified Model-agnostic lossless Image compression), a tokenizer-agnostic framework for lossless RGB image compression with frozen LLM backbones. LUMI replaces pixel-as-text tokenization with a pixel embedding module that maps raw intensity and channel information into the continuous embedding space of the LLM. It further introduces intra-patch position encoding to retain two-dimensional spatial structure after flattening, and uses a 256-way prediction head to produce probabilities over the native pixel alphabet. Only the pixel embedding, position encoding, soft-prefix parameters, and prediction head are trained, while the LLM backbone remains fixed. Experiments on natural, medical, and remote-sensing image benchmarks with LLaMA, Qwen, and Gemma backbones show that LUMI provides a unified interface across tokenizer families, achieves competitive compression rates, and improves cross-domain robustness over tokenizer-based LLM compression baselines. These results formulate LLM-based lossless image compression as pixel-space adaptation of frozen foundation models rather than tokenizer-specific language-symbol modeling.

Drew Miller, Cathy Wu

Freeway simulation is often difficult to deploy at scale not only because of model formulation, but because preparing road network inputs remains a manual, corridor-specific, and difficult-to-reproduce task. This paper presents an open-source tool that extracts freeway networks from OpenStreetMap (OSM) and converts them into a compact, station-referenced representation suitable for downstream freeway simulation. Unlike existing tools that primarily support arterial or general network conversion tasks, the proposed workflow is designed around the specific requirements of freeway traffic studies. The tool supports not only OSM data cleaning and conversion, but also the broader workflow required in practice: corridor-specific querying, visual inspection of extracted segments, extraction validation against OSM, and source-data validation against aerial imagery. A locally hosted frontend allows users to define corridor-specific queries, select endpoints visually, and inspect extracted segments. The extraction logic is designed to address several recurring challenges in freeway OSM data, including inconsistent route references, ambiguous path selection through interchanges, managed-lane interference, incomplete corridor capture from naive bounding-box queries, and inconsistent ramp classifications. The workflow was first tested on two prototype corridors, where the extract-first-then-validate approach proposed here required roughly one-third the analyst effort of manual ramp encoding from scratch. It was then deployed across 359.6 miles of freeway in Orange County, California, with total processing and validation averaging about 41 seconds per mile. This deployment also suggests that, in a well-mapped region, OSM is sufficiently accurate for many freeway traffic studies. Overall, the tool provides a more scalable and reproducible foundation for freeway network preparation.

Chang Wu, Junfeng Fang, Houcheng Jiang et al.

Safety alignment of large language models (LLMs) typically depends on high-quality supervision data, such as safe demonstrations or preference pairs. However, in real-world deployment, emerging safety requirements are often specified as natural-language policies, while corresponding supervision data may be costly, delayed, or unavailable. This creates a mismatch between rapidly evolving safety policies and conventional data-driven alignment methods. To address this, we propose PolicyAlign, a simple yet effective framework for directly aligning LLMs with safety policies. Given a safety policy, PolicyAlign first synthesizes policy-violating instructions and then performs on-policy self-distillation to internalize policy-guided behavior. To improve training stability and data efficiency, we further introduce Policy-Sensitive Filtering, which selects instructions where the policy induces the largest behavioral shift. Experiments across multiple models show that PolicyAlign consistently improves safety while maintaining low over-refusal and preserving general capabilities. PolicyAlign also generalizes to medical, legal, and financial safety scenarios, highlighting its potential as a scalable and maintainable approach to policy-based LLM safety alignment. The code is released at https://github.com/Qwen-Applications/PolicyAlign.

Kun Ma, Qilong Han, Chengjing Song et al.

Modern trajectory predictors increasingly condition on external spatial context, such as map geometry, signed distance fields (SDFs), and nearby moving agents. While this context improves prediction quality, constructing it for every training anchor has become a hidden systems bottleneck. In a representative maritime AIS pipeline, spatial context construction requires roughly 17 CPU-days for a 5.48M-anchor corpus, dominating the cost of the downstream predictor. We present M-CTX, an exact and scalable spatial context-retrieval framework for trajectory analytics. M-CTX recasts context construction as an ingest-once, query-many spatial database workload and replaces three brute-force stages -- OSM range retrieval, SDF computation, and moving-vessel neighbour lookup -- with composable, index-backed operators. Its learned range-index backend, BR-LZ, provides recall-complete MBR-overlap range retrieval and reduces candidate amplification by 1.1x--2.7x relative to global-expansion one-curve baselines. Across four maritime regions, eight baseline systems, synthetic workloads with up to 40M spatial features, and 10^7-record AIS streams, M-CTX reproduces the reference context exactly. On the 5.48M-anchor corpus, it reduces context construction from about 17 CPU-days to 1.8 hours, a measured 226x end-to-end speed-up. An optional storage mode further compresses SDF context by 64x with only a 0.04 m ADE change. These results establish exact spatial context retrieval as a first-class database problem in modern trajectory analytics. Code and datasets are publicly available at https://github.com/mark000071/M-CTX-Traj.

Zekai Zhang, Jiahao Li, Jie Zhang et al.

While text-to-image (T2I) models have achieved remarkable progress, they struggle with real-world requests that are often underspecified, implicit, or dependent on up-to-date knowledge. We identify this challenge as the Context Gap: the mismatch between the user context and the sufficient generation context for T2I models. To bridge this gap, we propose Qwen-Image-Agent, a unified agentic framework that integrates plan, reason, search, memory and feedback in a context-centric manner. Qwen-Image-Agent treats user input as partial context and progressively constructs the generation context through Context-Aware Planning and Context Grounding. Specifically, Context-Aware Planning identifies missing context and plans how it should be acquired and used, while Context Grounding gathers this context from reason, search, memory, and feedback. To evaluate agentic image generation, we further introduce Image Agent Bench (IA-Bench), a benchmark covering four core image agent capabilities: Plan, Reason, Search, and Memory. Experiments on IA-Bench, Mindbench and WISE-Verified show that Qwen-Image-Agent outperforms strong baselines and achieves state-of-the-art performance.

Chenyu Wu

Credit-card fraud losses are monetary, but papers often judge models with transaction-level scores. We ask whether transaction amount should shape training weights or be used later to order alerts. To separate this question from ordinary class imbalance handling, we keep total fraud-case weight fixed and vary only its allocation across fraud cases. The experiments test two chronological card-fraud datasets with XGBoost under unweighted training, standard class weighting, matched log-amount weighting, stronger amount-weighted variants, and score times amount reranking. Metrics are average precision, dollar recall, and dollar precision at fixed alert budgets over five seeds, with 95 percent day-block bootstrap intervals for the main contrasts. Results are narrower than expected. Amount-derived ratio and velocity features carry much of the signal, while raw amount fields add little once those features are in the model. In the matched setting, amount-conditioned training gives only small gains over class weighting and does not consistently beat the plain unweighted model. Stronger amount weights recover more fraudulent dollars, but at lower ranking quality and dollar precision. Reranking alerts by score times amount after training gives the largest dollar-recall shift. A small SHAP audit finds larger month-to-month attribution movement for fraud cases than for aggregate traffic. In these tests, amount is useful as a feature and as an alert-ordering variable, not by itself as a better sample-weighting rule.

Xiao Han, Yao Xiao, Chenyu Wu et al.

How many days of early behavior suffice for subscription churn prediction? In the public KKBox dataset, the early indicator of churn is typically an indicator of someone's contract status; however, when looking in the heavily churned manual-renewal segment, having access to early behavior creates a substantial increase in prediction for that specific segment (PR +0.10 at 120 days). A nine-window sufficiency curve shows a diminishing-returns knee in a 45-90 day band. However, stress-testing over three cohort/task designs shows that this curve is singular to the design being tested; for example, in our test with a moving target, the curve inverts and can shift depending on the feature set used. Therefore, any window-sufficiency claim should state its cohort construction, target definition, and feature families. All evidence is from one music-streaming dataset; the mechanism should generalize but the magnitudes may not.

Maolin Che, Yimin Wei, Chong Wu

High-dimensional hyperinterpolation is severely hampered by the curse of dimensionality, as its coefficient tensors grow exponentially with the ambient dimension. Existing research predominantly focuses on heuristic algorithmic optimizations, often overlooking the inherent structural properties of these tensors. This paper establishes a rigorous theory of intrinsic low-$ε$-Tucker-rank for hyperinterpolation coefficient tensors, delivering near-optimal low-rank approximations with error bounds that are nearly independent of the dimension. We further construct a unified, Tucker-compatible theoretical framework that integrates both Chidori-type and Fiber-type tensor CUR (TCUR) decompositions, deriving tight and stable Frobenius-norm error estimates that depend exclusively on tensor spectral properties and index set geometry. We mathematically verify the convergence and numerical stability of greedy adaptive index selection schemes and prove their near-optimality, enabling a fully tensor-free hyperinterpolation workflow that avoids constructing the full coefficient array. Three practical greedy TCUR algorithms and a lightweight TCUR-to-Tucker recompression pipeline are proposed as direct corollaries of our structural theory. Numerical experiments across three distinct families of high-dimensional test functions validate all theoretical predictions and confirm the intrinsic low-rank compressibility of hyperinterpolation coefficients. In contrast to prior algorithm-centric studies, this work prioritizes rigorous theoretical characterization over implementation tricks, establishing a unified structural and mathematical foundation for high-dimensional hyperinterpolation.

Zuhao Ge, Yuchen Zhou, Weitao Zhou et al.

Robust embodied robots should be able to recover from failures and retry tasks in order to operate reliably in unstructured and noisy real-world environments. Achieving this capability requires training policies on data that captures recovery behaviors. However, collecting such data through robot teleoperation is difficult to scale, as it is time-consuming to induce diverse failure states, perform corrective actions, and reset the environment. This challenge is further exacerbated by the high diversity of failure modes, which demands substantially more recovery data than success demonstrations. In this work, we show that egocentric human data capturing failure recovery processes provides a scalable alternative. By efficiently arranging task-level failure configurations and recording short recovery segments, human operators can generate more than 10x as much valid recovery data per hour compared to robot teleoperation under our protocol. To address the embodiment gap between human and robot, we propose EgoRecovery, a co-training framework for learning recovery behavior, where human recovery demonstrations are aligned to a compact corrective-intent space shared with robot data, which captures the timing and magnitude of correction. Only a small number of robot recovery demonstrations are required to connect this intent to executable robot actions. At deployment, a learned recovery gate predicts when correction is needed from robot observations and activates the corrective intent only in recovery states. Experiments on real-world recovery tasks show that EgoRecovery improves success from failure starts over robot-only recovery, direct co-training with human recovery data, and direct intent-transfer baselines.

Chunming Wu, Dafei Qiu, Congde Yuan et al.

Production customer-service bots must improve answer quality across iterative releases, yet large language models must not bypass evidence boundaries, policy rules, or human-handoff safeguards. We present an \textbf{Evidence-Grounded Customer-Service Agent Workflow} deployed in a real-world customer-service setting. BM25 recall, issue-title-vector recall, issue-description-vector recall, weighted RRF fusion, and cross-encoder reranking construct grounded FAQ evidence for controlled LLM decisions. Policy-guided orchestration then combines this RAG evidence with scenario-specific rule evidence, conversation memory, and clarification state inside a fixed LangGraph DAG~\cite{langgraph2024}. The paper contributes three reusable deployment patterns: \textbf{hybrid RAG evidence construction}, where multi-channel retrieval and reranking produce auditable FAQ candidates; \textbf{evidence-grounded issue/action decision}, where an Evidence-Grounded Decision Module selects an issue/action from typed FAQ evidence and scenario-specific rule evidence; and \textbf{trace-driven RAG and reranker improvement}, where traces diagnose whether failures come from recall, ranking, final candidate selection, clarification, rule-derived evidence, or action policy, and where reranker fine-tuning is evaluated not only for in-domain gain but also for forgetting risk.

Zhuohang Fan, Beichen Zhang, Yuanfa Li et al.

Graphical User Interface (GUI) agents powered by vision-language models hold promise for automating real-world mobile tasks. However, progress is limited by the lack of high-coverage, long-horizon interaction trajectories collected from element-rich and rapidly evolving apps. Existing pipelines often rely on costly human demonstrations or on-policy framework, which tends to over-sample common flows while missing rare transitions and complex multi-step procedures. To address this problem, we propose SEE, a two-stage data synthesis framework consisting of (i) an efficient exploration stage that builds an explicit UI transition graph over screens and elements, and (ii) a graph-based synthesis stage that composes diverse multi-step trajectories via planning and controlled sampling. This design yields reproducible and explainable data generation, while explicitly preventing spurious cycles and enabling long-horizon composition. Across multiple real-world apps, SEE produces trajectories with an average length of 14.8 steps while avoiding spurious loops, and agents fine-tuned on SEE achieve improved task success and generalization to unseen screens. We will publicly release our synthesis code and dataset.

Zefang Mao, Yunjia Li, Victoria Bamgboye et al.

Hydraulic soft fingers offer compliant and gentle manipulation, but their practical deployment is limited by bulky fluidic hardware, fabrication complexity, and insufficient design validation. This paper presents a compact 3D-printed soft hydraulic finger driven by a miniature cyclic peristaltic loop. The finger integrates compliant bellows, rigid connectors, and fluidic ports, while an Abaqus fluid-structure model is used to guide selection of wall thickness, pitch angle, and bellows length. The selected design is validated through baseline-corrected chamber-pressure measurements and vision-based angle tracking. Results show that the quasi-static finite-element model captures the main pressure-angle trends, with remaining offsets mainly attributed to bonding-induced stiffness and hydraulic losses. Vision-feedback control further enables repeatable angle tracking over a large bending range. Finally, grasping tests on fragile and deformable objects, including tofu and blueberries, demonstrate gentle, slip-free contact without visible damage. Overall, this work establishes a reproducible pipeline from FEA-guided design to closed-loop validation for compact 3D-printed hydraulic soft fingers.

Wujian Peng, Lingchen Meng, Yuxuan Cai et al.

Unified Multimodal Modeling aims to integrate visual understanding and generation within a single system. However, existing approaches typically rely on two disparate visual tokenizers, which splits the representation space and hinders truly unified modeling. We propose UniAR, a unified autoregressive framework where a single discrete visual tokenizer serves as the key bridge between understanding and generation, enabling a shared context in which the model can directly interpret its own generated visual tokens without additional re-encoding. UniAR adapts a pretrained vision encoder with multi-level feature fusion and a lookup-free bitwise quantization scheme, preserving both high-level semantics and low-level details while scaling the effective visual vocabulary at minimal cost. Building on this, the unified autoregressive model adopts parallel-bitwise-prediction to jointly predict spatially grouped, multi-level visual codes, substantially reducing visual sequence length and accelerating generation. Finally, a diffusion-based visual decoder operates on discrete visual tokens to decode high-fidelity images. Through large-scale pre-training, followed by supervised fine-tuning and reinforcement learning, UniAR achieves state-of-the-art performance on image generation and image editing while remaining competitive on multimodal understanding benchmarks. The project page is available at https://sharelab-sii.github.io/uniar-web.

Maksym Zubkov, Carol Wu, Shiwei Yang et al.

We study the expressivity of shallow polynomial neural networks (PNNs) with monomial activation functions over finite fields. For a given architecture, we define a neuromanifold as the image of the map from all possible network weights into the product of polynomial rings. We quantify the expressivity by the cardinality of the neuromanifold, and derive a natural lower and upper bound. This leads to counting rational points over finite fields, a problem closely linked to the Weil conjectures. Finally, we present an architecture that exhibits a striking difference in the neuromanifolds when considered over a characteristic zero versus a finite-characteristic field, illustrating the critical role of field characteristic in the notion of expressivity.

Xiao Han, Jingjing Liu, Moxuan Zheng et al.

A total of seven categorical encoding methods were tested on the IEEE-CIS fraud benchmark dataset (590,540 records, 3.5% positives, 8 high-cardinality columns). The encoders were evaluated using a stratified 5-fold cross-validation (CV) with three repetitions. Five of the encoders had identical frozen LightGBM learners in the downstream phase, allowing for controlled comparisons of their performance to each other. CatBoost and TabNet were included as comparisons across paradigms using different learners. The entity embeddings produced the highest AUC-ROC (0.9612), with a statistically significant tie with that of CatBoost (0.9602) and statistically superior to tier group encoding (0.9548), whereas target encoding was only 0.0023 worse than tier group encoding and the auditor-friendly tier boundaries were maintained. Off-the-shelf TabNet did not outperform tree-based pipelines and collapsed under data scarcity. On AUC-PR, CatBoost leads (0.822 vs. 0.793); no encoder dominated both metrics. Per-column analysis confirmed the embedding advantage arises from joint multi-column representation.

Chen Zhao, Xiantao Hu, Song Wu et al.

Spiking neural networks (SNNs) have garnered significant interest in computer vision due to their potential for efficiency and biological inspiration. While spiking CNN-based methods have shown promise for image restoration (IR) tasks, their performance is constrained by the inherent receptive field limitations of CNN operations. In the paper, we explore the benefits of discrete wavelet transformation and propose a spiking pyramid wavelet-based model (SPWM) for high-efficient and low-energy target. Specifically, we develop a spiking dual pyramid wavelet (SDPW) block to model long-range dependency and exploit the properties of the degradation in the wavelet domain. Experimental results on several benchmarks demonstrate that SPWM significantly lowers computational costs and energy consumption while maintaining image quality. Our method showcases the potential of SNNs in the field of IR, offering new insights for future applications of resource-limited devices.

Gang Liao, Yujia He, Abdullah Ozturk et al.

The database community has repeatedly advanced the state of the art by recognizing that new workloads demand new system architectures. We argue that long-horizon agentic tasks -- code generation, scientific discovery, hardware design -- are such a workload. These agents explore: they generate artifacts, execute tools, observe failures, branch, and repair over hundreds of steps. This search produces a structured object we call an experience graph: executable artifacts, tool outputs, rewards, sibling comparisons, and causal lineage. Yet existing agent frameworks treat this experience as disposable state -- JSON checkpoints and session logs that cannot be recovered after a crash, queried across users, or materialized into training data. We propose Trellis: a data foundation that treats the experience graph as first-class, governed, queryable database state. The core insight is that search over experience graphs is a database access pattern. Frontier selection is a query, cross-session reuse is vector-seeded graph retrieval, training-data extraction is a materialized view, and reconstructing what an agent knew at any past step is a time-travel query. When the database owns the experience graph, agents become stateless compute, and crash recovery, horizontal scaling, and a closed-loop training flywheel emerge as architectural byproducts. We ground the design in KernelEvolve, a production accelerator-kernel optimizer at Meta, where cross-session reuse reaches a target speedup roughly 10x faster at 52% lower token cost. More broadly, Trellis turns inference-time search from disposable computation into a durable institutional asset: logs made databases reliable; experience graphs may make agents cumulative.

Ali Larian, Qian Lin, Chang Zong Wu et al.

As autonomous agents are increasingly deployed across diverse operational contexts, aligning their behavior with human intent demands reward functions that remain robust to such changes rather than overfitting to any single environment. Inverse reinforcement learning (IRL) provides a principled way to infer such objectives from human feedback. However, existing analyses of optimal teaching approaches for IRL focus on single-environment, demonstration-only settings, leaving underexplored how heterogeneous feedback modalities and environment dynamics jointly constrain reward functions that generalize across multiple environments. Because demonstrations in one MDP entangle reward information with that environments specific structure, the resulting rewards frequently fail to generalize when the agent is deployed in a new setting. We first analyze how different feedback modalities constrain rewards, showing that, in the unlimited-data regime, comparisons impose strictly stronger global constraints than other modalities. Beyond this theoretical analysis, we introduce a hierarchical machine teaching algorithm for reward learning that operates across multiple MDPs. The algorithm first greedily selects informative environments that expose complementary reward constraints, then strategically queries low-cost feedback within those environments. Empirically, our method achieves substantially lower regret and stronger generalization to held-out environments than uniform teaching baselines under identical feedback budgets, demonstrating the importance of multi-environment, multi-modal teaching for learning dynamics-robust reward functions.

Zhuoyun Li, Boxuan Wang, Changshun Wu et al.

Rejecting inputs outside the defined in-distribution (IND) service scope is critical for large language model (LLM) services, where unsupported requests should be filtered before full generation. Existing out-of-distribution (OOD) detectors often rely on final outputs or final-layer representations, leaving unclear where service-boundary signals are most clearly encoded inside the model; they also lack a theoretical guarantee for held-out inputs. In this paper, we introduce SCOPE (Sequential Conformal OOD Probing and Evaluation), a framework that selects a readable hidden layer, constructs a conformal gate with IND calibration, and uses a supermartingale e-process to certify persistent service-boundary evidence. Experiments across multiple LLM backbones and six carefully designed boundary conditions show that SCOPE improves gate-level rejection over standard final-layer detectors, while revealing how different OOD boundaries take different geometric forms in hidden space.

Yang Liu, Zhaokai Luo, Huayi Jin et al.

Recent LLM-based agent systems continuously accumulate context across multi-turn interactions, tool invocations, and cross-session workflows. Replaying the full history for every request quickly becomes impractical: long contexts increase prefill cost, may exceed context limits, and often bury task-relevant evidence in irrelevant content, degrading both serving efficiency and output quality. We propose Akashic, a low-overhead memory system built around MemAttention, which organizes context into bounded chunks and models semantic relationships across chunks, preserving cross-chunk evidence without repeatedly rewriting the full history. Akashic further applies hardware-software co-designed memory placement to co-locate likely co-retrieved chunks, reducing retrieval fragmentation and I/O overhead. Across four representative workloads and three model sizes, Akashic improves task accuracy by up to 10.2 points, throughput by up to 1.21x, and sustainable request rate by up to 1.88x over strong prior memory baselines.

Haili Xiao, Can Wu, Shasha Lu et al.

Runtime estimates affect reservation quality, backfilling opportunities, and queue delay in HPC schedulers. Under heavy tailed workloads, however, averaging over jobs can misrepresent scheduling impact because a small fraction of jobs dominates resource usage. This paper presents an empirical evaluation methodology for HPC job runtime prediction that focuses on the tail, combining GeoAccuracy weighted by resource usage with decile and split analyses. Using production traces from NREL Eagle and ALCF Mira/Intrepid, we compare XGBoost and Last2 against the user provided walltime estimate at submission (UserReq). Across all three datasets, evaluation focused on the tail changes the offline conclusion: MeanAccuracy keeps the methods relatively close, whereas GeoAccuracy reveals clearer separation and makes UserReq's strength in the upper tail visible. In the top decile, UserReq achieves the highest GeoAccuracy and lowest underestimation rate on all three datasets, and this pattern remains stable across rolling splits. We then translate this signal into a simple hybrid scheduling policy that keeps XGBoost for most jobs and routes the top decile by proxy_cost at submission to UserReq. Online replay on four production queues reduces mean wait time by up to 8% and increases backfilled jobs by 50% to 115%. These results show that offline evaluation focused on the tail better characterizes prediction quality relevant to scheduling and informs scheduling policy design.

Gengtian Shi, Jinze Yu, Chenhao Wu et al.

The exponential growth of video traffic demands novel semantic communication paradigms that transmit meaning rather than raw bits. We present a generative AI-enabled framework for semantic video communication addressing two critical challenges: efficient hierarchical temporal modeling for bandwidth-constrained transmission and robust semantic alignment between video content and natural language queries at network edge devices. Our approach introduces a multi-scale temporal convolutional encoder that captures motion patterns across different temporal granularities with O(T) complexity suitable for resource-constrained IoT deployments. We further propose a capsule-based dynamic routing mechanism that iteratively refines segment-query associations, enabling flexible modeling of non-monotonic semantic alignments essential for goal-oriented communication. These components are unified through a multi-task learning objective optimizing temporal boundary regression, cross-modal alignment, and capsule diversity. Experiments on ActivityNet Captions demonstrate significant improvements, achieving 42.9% Recall@0.5 and 41.1% mean IoU while maintaining computational efficiency critical for edge deployment.

Fang Gao, Jiongkai Qin, Jiabao Wang et al.

Low-light degradation reduces image visibility and weakens structural cues that are important for visual representation and scene understanding. Existing low-light image enhancement methods mainly focus on appearance restoration, while insufficiently exploiting scene geometry to preserve structural consistency. To address this limitation, this paper proposes a Depth-guided Multi-scale Attention Network (DMSA-Net) for geometry-aware low-light image enhancement. DMSA-Net introduces depth-related structural priors into low-light representation learning through reflectance-geometry interaction. A Retinex-based decomposition module is first used to obtain illumination-invariant reflectance representations, from which depth cues are inferred to characterize scene structure under degraded illumination. A multi-scale depth-guided fusion strategy is then embedded into a hierarchical encoder-decoder architecture, where depth-aware attention adaptively integrates geometric and appearance features. Experiments on several benchmark datasets show that DMSA-Net achieves effective low-light restoration while improving structural preservation. Moreover, we construct LOL-D, a depth-augmented low-light dataset, to facilitate research on geometry-aware low-light vision.

Ziqi Yuan, Wenhao Lu, Hao Wu et al.

Vibe coding democratizes software development by allowing users to generate code via natural-language (NL) interaction with large language models (LLMs). However, the code is reliable only when it faithfully implements the user's intent, which is difficult and labor-intensive for users to validate. Existing validation methods either rely on LLM-assisted automated testing, which suffers from prompt ambiguity and model fallibility, or involve users only in partial software artifacts such as prompts and test cases, which may overlook corner cases and program details. Motivated by a bug study of LLM-generated code, we find that detailed human feedback is essential, as failures often stem from underspecified requirements or subtle semantic deviations. This paper presents verifiable literate programming (VLP), a human-in-the-loop framework designed to make the review/validation process of LLM-generated code accessible to users at all programming levels. At its core, VLP proposes unambiguous NL-based documentation as a readable intermediate layer between prompts and code. The documentation demonstrates concrete program semantics and enables users to provide feedback on potential intent-code mismatches. It supports human-involved, end-to-end repair and validation via three techniques: (i) an NL-style literate language with unambiguous syntax and mostly deterministic code-to-documentation translation, (ii) LLM-based fine-grained mismatch detection that uses trace links between prompts and documentation to focus users' review effort on suspicious documentation lines, and (iii) a verification module that leverages user-validated documentation to derive API-usage checks and formal properties, which are then verified against the generated code using model checking. Our evaluation shows that VLP improves code pass@1 from 28.7%-73.2% to 65.4%-93.5% with reasonable user effort.

Jingwei Song, Haofeng Xu, Jie Xiao et al.

High-throughput RLHF systems often decouple rollout generation from policy optimization, leading to the use of stale rollouts during learner updates. In this work, we study the effect of such staleness in asynchronous GRPO. We make the behavior policy explicit in the GRPO surrogate objective and distinguish between the surrogate-gradient mapping used by the learner and the true total derivative of a distribution-dependent population objective. Under assumptions of local boundedness, distributional smoothness, and behavior-policy smoothness, we show that stale rollouts introduce a per-step surrogate-gradient bias of order O(S * eta), where S denotes the maximum rollout lag and eta denotes the learning rate. We further derive a conditional collapse-time scaling law: when within-cycle drift remains below a batch-level clipping radius, collapse is governed primarily by cumulative learner drift T * eta; when the stale-rollout constraint is active, stability instead depends explicitly on S * eta. This yields a two-constraint stability condition eta << min{R_batch / (S * G_upd), R_crit / (T * G_upd)}, explaining why the maximum stable learning rate may appear weakly dependent on staleness in the horizon-limited regime.

Tao Feng, Xinke Jiang, Chao Wu

Agentic search equips large language models with dynamic retrieval abilities, but existing reinforcement learning methods remain limited by reward sparsity in knowledge boundary calibration -- deciding when to trust parametric memory, when to rely on retrieved evidence, and when to abstain. Binary rewards can penalize undesirable outcomes, but provide little guidance on the reasoning process required to make calibrated decisions across different knowledge states. To address this, we propose KbSD (Knowledge boundary Self-Distillation), a framework that tackles this limitation through dense token-level supervision, outcome-level sparse rewards, and quadrant-adaptive optimization. KbSD constructs a hint-augmented teacher, architecturally identical to the student, that receives explicit knowledge boundary signals -- including parametric certainty, retrieval quality, and ground-truth answers -- to generate calibrated reasoning demonstrations. This information-asymmetric self-distillation enables dense supervision without requiring a larger external model. To further account for the heterogeneous reasoning distributions across knowledge states, we introduce a quadrant-adaptive distillation objective: reverse KL for concentrated integration, forward KL for diverse refusal, and Pareto-optimal bidirectional KL for asymmetric quadrants requiring both precision and coverage. Experiments on multiple benchmarks show that KbSD consistently improves both task accuracy and hallucination mitigation over strong baselines, with the largest gains appearing in the challenging quadrants where sparse rewards are least informative.

Zetian Song, Chenming Wu, Junnan Liu et al.

High-fidelity reconstruction of dynamic urban environments is a cornerstone of autonomous driving simulation and large-scale world modeling. While 3D Gaussian Splatting (3DGS) has established a new standard for real-time rendering, its reliance on expensive per-scene optimization limits scalability. Conversely, recent feedforward methods that infer Gaussian parameters offer faster speed but face fundamental bottlenecks: they are memory-prohibitive at high resolutions and struggle to fuse dense multi-view observations consistently. This paper presents L2D2-GS, a unified framework that reformulates generalizable reconstruction not as a one-shot regression, but as a robust iterative process of optimization and densification. To resolve the ambiguity of supervision in primitive generation, we propose a self-supervised densification policy that derives explicit reward signals from global reconstruction gains to guide local densification. Furthermore, we mitigate irreversible early-stage artifacts through a geometric regularization mechanism, utilizing reparameterization to constrain the optimization manifold and prevent convergence to poor local optima. Extensive experiments on the PandaSet and Waymo datasets demonstrate that our method achieves state-of-the-art reconstruction fidelity and strong zero-shot generalization, while using fewer primitives than competing baselines.

Chao Wu, Marek Chrobak

We study the problem of gossiping (all-to-all information exchange) in ad-hoc radio networks. Such a network is represented by a strongly-connected directed graph with \(n\) vertices, whose topology is initially unknown to the protocol. In 2004, Gasieniec, Radzik, and Xin gave a \(\tilde O(n^{4/3})\)-time deterministic protocol for this problem, and closing the gap between their upper bound and the \(\tildeΩ(n)\) lower bound on the time complexity of gossiping remains a central open problem. We develop a deterministic protocol for gossiping in ad-hoc radio networks that achieves running time \(\tilde O((mn)^{3/5})\) for directed graphs with at most \(m\) edges. Our protocol improves on the \(\tilde O(n^{4/3})\) bound when \(m = O(n^c)\), for \(c < 11/9\). We also present a \(\tilde O(Δ^{1/2} n)\)-time gossiping protocol for \(Δ\)-regular graphs.

Yixian Xu, Kaiyuan Gao, Yuxiang Chen et al.

We present Qwen-Image-2.0-RL, a post-training pipeline that applies reinforcement learning from human feedback (RLHF) and on-policy distillation (OPD) to improve both the visual quality and instruction-following capability of the Qwen-Image-2.0 diffusion model. To provide reliable reward signals, we construct task-specific composite reward models by fine-tuning vision-language models with a pointwise scoring paradigm and chain-of-thought reasoning. For text-to-image generation, the reward models cover alignment, aesthetics, and portrait fidelity dimensions. For image editing tasks, the reward system addresses instruction-following accuracy and face identity preservation. Building on this reward system, we develop a scalable GRPO-based RL training framework, incorporating a hybrid classifier-free guidance (CFG) strategy to preserve pre-trained knowledge, prompt curation via intra-group reward range filtering, and per-category reward weight calibration. To merge the task-specialized RL policies for T2I and editing, we propose on-policy distillation as the final training stage, which consolidates multiple teachers into a single student model through trajectory-level velocity matching. Extensive evaluation shows that Qwen-Image-2.0-RL achieves 57.84 overall score on Qwen-Image-Bench (+2.61 over the base model), Elo ratings of 1193 in text-to-image arena (+78) and 1349 in image edit arena (+93), demonstrating consistent gains in aesthetic quality, prompt adherence, and editing accuracy.

ZunDong Zhang, Yanan Cheng, Zhaoxin Zhang et al.

As personal data privacy becomes increasingly critical in Internet of Things (IoT) environments, secure DNS protocols such as DNS over HTTPS (DoH) and DNS over TLS (DoT) have been widely adopted to protect device communications. However, without effective obfuscation, these protocols remain vulnerable to Website Fingerprinting (WF) attacks that can reveal user activity. With the ongoing deployment of DNS over HTTP/3 (DoH/3) in modern networked systems, padding strategies have been increasingly applied in practice. It is therefore essential to investigate whether DoH/3 can effectively mitigate WF attacks in realistic IoT and edge-network scenarios. To address this, we first collect and publicly release the first real-world benchmark dataset of DoH/3 traffic, generated from domain resolution processes in practical network environments. We further propose DoHFuse, a dual-branch learning framework that integrates inter-arrival time sequences and refined statistical features through an improved DMAG-LSTM, specifically designed to capture burst-aligned temporal patterns. Experimental results show that DoHFuse achieves an accuracy of 88.05% (precision 88.56, recall 87.96, F1 87.83) in a closed-world setting of 449 classes, and an AUPRC of 0.975 with an F1 score of 0.951 (precision 0.906, recall 1.0) in open-world detection. These findings demonstrate that DoH/3 traffic remains susceptible to WF attacks, suggesting that commonly deployed padding mechanisms alone are insufficient to ensure privacy protection in IoT-scale encrypted DNS communications.

Diana Guadalupe Gomez, Chenwei Wu, Zhiyi Wang et al.

Epidemic forecasting models typically rely on surveillance data reported over administrative regions, treating them as atomic units, thereby obscuring sub-regional spatial structure that shapes disease dynamics. We introduce a spatially structured multimodal epidemic forecasting setting that integrates region-level temporal surveillance data with spatially localized auxiliary signals that are misaligned in resolution and structure, reflecting realistic public health reporting constraints. Building on this formulation, we propose M-SPICE (Multimodal SPatIal Context for Epidemic Forecasting), a structure-aware spatiotemporal forecasting framework that performs joint reasoning over temporal disease dynamics and spatial context via attention-based multimodal fusion, allowing spatial signals to selectively condition temporal representations across forecast horizons. We evaluate our approach on real-world COVID-19, influenza, and influenza-like illness (ILI) forecasting tasks under realistic real-time evaluation protocols. Across all forecasting settings, our method consistently outperforms state-of-the-art multivariate time-series, multimodal, and epidemiological forecasting baselines while maintaining strong probabilistic forecasting performance. Finally, interpretability analyses reveal when, where, and how spatial signals are leveraged, highlighting settings in which purely temporal, region-aggregated models are most likely to fail.

Mani Hamidi, Sina Khajehabdollahi, Charley M. Wu et al.

The intricate structures of biological neural networks largely emerge during development, guided by a comparatively compressed blueprint encoded in the genome. The connectivity that emerges from this decoding process is rich in structure, and already equips the organism with functional modules upon birth. This initial structure serves as a scaffold that can be gradually refined and fine-tuned through lifelong experience, via a variety of plasticity mechanisms. Drawing inspiration from this interaction between evolutionary and developmental modes of learning, we use hypernetworks to learn a compressed generative process that generates the connectivity of a modular reservoir. We show that this marriage between curriculum-based meta-learning and modular reservoir computing can generate sparse recurrent networks that solve difficult temporal tasks with minimal training and without concessions to robustness.

Joshua Engels, Callum McDougall, Bilal Chughtai et al.

LLM reasoning transparency is a critical affordance for understanding model decisions, mitigating misuse and misalignment, and debugging surprising model behaviors. However, DiffusionGemma performs a larger fraction of its computation in a continuous latent space; does this make its reasoning less transparent? We study this question by decomposing transparency into two components: variable transparency, whether we understand intermediate snapshots of a model's computational state; and algorithmic transparency, whether we can use these snapshots to reconstruct the process by which the model arrived at its outputs. Naively, DiffusionGemma has poor variable transparency: its opaque serial depth, the amount of serial computation that occurs in between interpretable model states, seems at first 28.6X higher than the corresponding autoregressive Gemma 4 model. However, we show that we can map the information flowing between denoising steps through an interpretable token bottleneck with no decrease in downstream performance. Treating these intermediate states as interpretable reduces the opaque serial depth to just 1.1X that of Gemma 4. Algorithmic transparency is harder for diffusion models than for autoregressive models because all token predictions in the canvas can change at every denoising step, giving the model the power to implement complicated distributed algorithms during the denoising process. To begin bridging this gap, we conduct a suite of interpretability case studies, uncovering initial evidence of novel diffusion-specific phenomena such as non-chronological reasoning, token and sequence smearing, and intermediate-context reasoning. Finally, we test monitorability, a key application of transparency that measures whether model outputs are useful for downstream tasks. We find that DiffusionGemma is similarly monitorable to Gemma 4.

Haocheng Duan, Yuxin Guo, Jieyi Bi et al.

Neural Combinatorial Optimization (NCO) achieves strong performance, yet its black-box nature remains a key roadblock to deployment and scientific diagnosis. Standard interpretability tools, such as Concept Bottleneck Models (CBMs), are ill-equipped for NCO, whose decisions are dynamic, state-dependent, and lack proper concept vocabulary definition. To close this gap, we introduce Evolving Programmatic Bottlenecks (EPB), to our knowledge, the first framework for interpreting NCO policies by distilling black-box NCO models into human-readable program portfolios. EPB employs an LLM to autonomously evolve a bank of programs, where each program's per-step action distribution serves as the bottleneck. EPB works through an iterative framework: Block I fixes program bank capacity and introduces a hybrid textual-numerical gradient descent scheme that couples numerical gradients for student router updates and textual gradients for LLM-based program revision; Block II dynamically adapts bank capacity via fault-targeted expansion and redundancy pruning. Extensive experiments demonstrate EPB's effectiveness and broad applicability, where the distilled program portfolios largely match original performance. EPB also reveals that NCO behavior shifts across optimization stages and can be approximated as a composition of classic heuristic variants. Our work advances interpretable NCO and establishes EPB as a promising tool for interpreting sequential decision-making models.

Jung-Hoon Cho, Heling Zhang, Siqi Du et al.

Policies must operate across diverse conditions, yet a single policy is often conservative while fully adaptive schemes can be complex. We study zero-shot generalization in contextual dynamical systems and introduce a performance-centric, directional task dissimilarity--the signed divergence--that upper bounds the generalization gap from a source context to a target context. The signed divergence induces $\varepsilon$-tolerance sets that certify when a source policy class generalizes, and it yields a concrete notion of task-space complexity: the minimum number of source contexts needed so that every target context incurs at most $\varepsilon$ generalization gap. Under a mild local smoothness assumption on performance, the induced tolerance sets admit certified inner/outer balls and instance-dependent volume bounds on task-space complexity. In the finite-oracle setting, source selection reduces to set cover; a greedy strategy inherits the standard $H(n)$ approximation guarantee. Using a Mass-Spring-Damper system with linear-quadratic regulator (LQR) controllers and a nonlinear CartPole system with deep reinforcement learning controllers, we show that greedy selection achieves the same $\varepsilon$-coverage with fewer policies than uniform or random baselines. Our approach delivers a performance-based task similarity measure and practical certificates for building generalizable control with simple policies.

Jingdong Shen, Fu Wang*, Qifeng Lu et al.

Hyperspectral infrared observations are an important data source for numerical weather prediction (NWP) because they provide rich information on the vertical structure of atmospheric temperature and humidity. However, most existing deep learning methods mainly focus on one-way retrieval from radiances to atmospheric profiles, while the reverse radiance simulation process and the consistency between atmospheric state space and radiance observation space are insufficiently considered. In this study, we propose SIMBA, a unified bidirectional retrieval-forward simulation framework for FY-4A GIIRS hyperspectral infrared radiance modeling toward NWP applications. The framework jointly performs atmospheric profile retrieval and radiance reconstruction, introduces a cycle-consistency constraint to strengthen the coupling between the two processes, and employs a bidirectional Mamba state-space module to capture long-range dependencies along pressure levels. Using collocated FY-4A GIIRS observations and ERA5 reanalysis data, the proposed method is evaluated for temperature retrieval, specific humidity retrieval, long-wave radiance reconstruction, and medium-wave radiance reconstruction. Experimental results show that SIMBA outperforms several representative deep learning baselines across both retrieval and reconstruction tasks, while ablation experiments confirm the contribution of the bidirectional design and cycle-consistency mechanism. These results demonstrate that the proposed framework is effective for joint atmospheric profile retrieval and hyperspectral infrared radiance modeling, and suggest potential for future Jacobian-related analysis and NWP-oriented extensions.

Chenrui Wu, Zexi Li, Jiajun Bu et al.

Reinforcement Learning with Verifiable Reward (RLVR) has emerged as a powerful post-training paradigm that surpasses Supervised Fine-Tuning (SFT) in eliciting reasoning intelligence and resisting catastrophic forgetting. Recent studies further reveal that RLVR induces highly sparse and off-principal parameter updates compared to SFT. This naturally raises the question: does such sparsity make RLVR models more amenable to model merging? If so, model merging would offer a scalable, training-free path to aggregate diverse reasoning capabilities from independently trained RLVR models. Surprisingly, we find the opposite, uncovering a sparsity curse: the sparse RLVR updates are spread farther apart in parameter space, forming near-orthogonal shortcuts that make aggregation inherently fragile. This is likely rooted in the stochasticity of RL optimization and the diversity of emergent reasoning patterns. Unlike SFT models that converge to shared, flat basins and merge naturally, RLVR models suffer severe degradation under standard merging methods. Through systematic empirical analysis of the update geometry, we characterize the mechanisms behind this failure and propose Sensitivity-aware Resolving Merging (SAR-Merging), a merging recipe tailored for the unique structure of RLVR parameter spaces. SAR-Merging resolves conflicts in overlapping update regions via Fisher Information-based sensitivity arbitration, followed by magnitude-aware sparsification and rescaling to preserve fragile reasoning pathways. Experiments on mathematical and coding benchmarks demonstrate that SAR-Merging substantially outperforms existing merging methods on RLVR models, enabling both single-task enhancement and multi-capability fusion.

Longlong Zhu, Jiashuo Yu, Zedi Chen et al.

Network planning optimization is a fundamental problem across diverse domains, including transportation systems, communication networks, and power grids. It requires simultaneous optimization of multiple competing objectives under complex constraints. Existing network planning optimization frameworks rely on mixed integer programming (MIP) solvers, heuristics, and deep reinforcement learning (DRL) models to compute planning decisions. However, they lack effective adaptability to diverse and dynamic user intents, thus leading to the trade-off between execution time and optimality. In this paper, we propose OmniPlan, an adaptive framework that achieves both timeliness and near-optimality in network planning optimization. To achieve the adaptability lacking in existing solutions, OmniPlan employs a large language model (LLM)-based interpreter to convert heterogeneous natural-language intents into a unified and quantifiable user-preference vector. Then it employs a mixture-of-experts architecture that integrates MIP solvers, heuristics, and DRL models as specialized experts, where OmniPlan adapts to diverse intents by dynamically selecting timely and near-optimal experts. Finally, it incorporates a DRL-based expert configuration module that fine-tunes optimization objective weights to align planning decisions with user-specific preferences. We evaluate OmniPlan with a representative real-world workload, i.e., distributed machine learning (ML), where we leverage OmniPlan to offload a wide spectrum of ML inference tasks, e.g., decision trees, SVM, naive Bayes, XGBoost, and random forests, onto a network of hardware devices. Our experiments on a real-world testbed indicate that OmniPlan achieves near-optimal and low-execution-time offloading for real-world ML inference tasks, reducing latency by up to 97.8\% and network device resource consumption by up to 11.5\%.

Vincent T. Lee, Bilge Acun, Zachary Lewis et al.

As the carbon cost of manufacturing and operating semiconductor devices has come into sharper focus, sustainability has gradually emerged as a new system architecture design metric. Like power and performance modeling tools, enabling sustainability-aware silicon systems design and optimizations will require a new generation of electronic design automation and architectural modeling tools. Towards this end, we present an update to the Architecture Carbon Tool v3 (ACT3) which aims to provide an extensible and customizable modeling platform for research and advanced development to pave the path towards sustainability-aware architectural design space exploration. Compared to previous versions of ACT, ACT3 provides significantly richer modeling capabilities, enhanced collateral and analysis telemetry, and first order design space exploration capabilities. This technical brief provides an overview of these expanded capabilities and illustrates ACT3's basic utility across several case studies. Finally, we identify opportunities for research and development where we expect the research community can contribute towards continuing to improve sustainability modeling and design methodology for silicon systems.

Kun Ma, Qilong Han, Chengjing Song et al.

Vessel trajectory prediction is important for intelligent shipping, maritime surveillance, and navigation safety. However, existing public maritime AIS resources are often limited by inconsistent forecasting protocols, uneven data quality, and the lack of benchmark-ready contextual annotations, which hinder fair comparison and context-aware modeling. To address this gap, we present EnvShip-Bench, a unified benchmark for short-term vessel trajectory prediction built from large-scale raw AIS data from the Danish Maritime Authority (DMA) and NOAA through a common processing pipeline. EnvShip-Bench adopts a standardized forecasting protocol with 10 minutes of observation, 10 minutes of prediction, and 20-second sampling in vessel-centric local metric coordinates. Beyond the large-scale core benchmark, it provides a quality-first compact subset for efficient and reproducible experimentation, together with synchronized environmental and nearby-vessel context extensions. As a result, EnvShip-Bench supports trajectory-only, environment-aware, and interaction-aware forecasting under a unified evaluation framework. Extensive benchmark statistics and analysis demonstrate that EnvShip-Bench offers a standardized, extensible, and context-aware foundation for maritime trajectory forecasting research.

Hanqi Zhou, David G. Nagy, Peter Dayan et al.

How do people build abstract, reusable knowledge from sequential experience under bounded cognitive resources? To answer this question, we integrate rate-distortion theory with recent advances in program induction to describe how prior knowledge shapes which future structures are cheap to encode and easy to discover. We formalize this in a hierarchical Adaptor Grammar (HAG) with distinct local (within-task) and global (across-task) libraries, governed jointly by constraints on memory and computation. In simulations, HAG achieves better rate-distortion trade-offs and stronger generalization than fixed grammars or shallow chunking methods. In an online melodic sequence-learning experiment, participants' recall errors reflected systematic simplifications and reaction times increased at inferred program boundaries. Trial-by-trial fits further showed that hierarchical libraries best explained individual differences in both recall and out-of-sample continuation choices, outperforming all alternative models. These findings cast structured learning as bounded program induction in which the order of experience shapes future abstractions a learner builds.