Computer Science & Mathematics

In minimally invasive surgery (MIS), maintaining an appropriate field of view (FoV) is essential for surgical safety and efficiency. However, conventional autonomous laparoscope control methods often treat visual tracking and surgical phase recognition independently, limiting robustness under occlusion, entity overlap, and camera motion. This paper proposes a closed-loop autonomous laparoscope control framework that couples real-time surgical phase recognition with camera-motion-compensated visual perception. By exploiting phase-dependent relationships among surgical entities, the framework improves target state estimation under challenging visual conditions, supporting task-aware autonomous FoV control. Experiments on benchmark datasets evaluate the proposed framework in surgical phase recognition and target tracking, including scenarios involving occlusion and entity overlap. Phantom experiments further demonstrate the feasibility of autonomous FoV control, showing improved target visibility and task execution performance compared with fixed-FoV control. These results highlight the potential of incorporating task-level semantics into visual state estimation for autonomous laparoscope guidance.

Indoor IoT cameras can support assisted living and activity-aware spaces, but exporting raw RGB frames raises privacy concerns by exposing facial, clothing, and household details that body-motion tasks often do not require. We present StickFigure, a pose-only IoT camera system that, during real-time monitoring, keeps raw frames local and ephemeral while exporting only compact skeleton streams sufficient for downstream human activity recognition (HAR). While pose estimation has matured on GPUs, extracting accurate pose streams on resource-constrained IoT cameras remains difficult. To overcome these challenges and enable on-device pose streams on (almost) any IoT camera, we present two core designs in StickFigure: an activation-aware pose backbone that allocates channel capacity after spatial downsampling and a lightweight hierarchical MLP refiner that selectively corrects elbow and wrist coordinates using intermediate visual features and kinematic context. We fully implement StickFigure and evaluate it on public benchmarks and real-world deployment. Compared with state-of-the-art MCU pose models, StickFigure achieves the highest pose accuracy and downstream HAR utility on both NTU RGB+D and Penn Action, while using 96.5% fewer parameters and 99.3% fewer MACs per frame than RTMPose-S-based baselines, achieving 0.107 s p95 latency on a representative STM32H753 MCU. We believe StickFigure establishes a new operating point for privacy-aware edge vision on massive IoT cameras.

The emergence of diverse intelligent services is driving 6G networks toward an intelligent communication ecosystem with increasingly complex operating scenarios. However, existing AI-based communication methods mainly focus on isolated tasks or individual physical-layer mod- ules, which is insufficient for AI-native 6G systems that must jointly perceive dynamic environments, understand specific user intents, coordinate multiple coupled PHY-layer modules, and remain compatible with existing communication infrastructures. To address this gap, this paper, for the first time, explores auto- matic environment- and intent-aware full-link construction, which aims to generate executable PHY-layer configurations according to channel conditions and service requirements. We propose AGENS, an agentic communication decision-making framework for AI-native 6G link construction. Rather than replacing the existing PHY-layer protocol stack with a black-box end-to-end model, AGENS operates on top of the modular communication framework and jointly optimizes end-to-end PHY-layer configu- rations. Specifically, AGENS encodes CSI and natural-language user instructions, aligns communication states with user intents through cross-modal fusion, and leverages a pretrained LLM for intent-aware reasoning. An LSTM-based sequential action generator further captures dependencies among coupled sub- actions and produces structured full-link strategies. To enable stable training in the high-dimensional discrete action space, AGENS adopts a two-stage learning framework, where heuristic supervised initialization is followed by multi-objective reinforce- ment learning to jointly optimize reliability, throughput, and power consumption.

Multipath parameter prediction provides a structured and physically interpretable representation of wireless propagation for environment-aware communication systems. However, existing Transformer-based methods repeatedly encode the same environment when predicting multiple BS--UE links, resulting in redundant computation. To address this issue, we propose an environment-conditioned dynamic KV caching framework for multipath parameter prediction. Specifically, we decouple environment-dependent computation from BS--UE geometry, enabling environment-side representations and intermediate KV states to be reused across multiple link queries while only the geometry branch is updated. Building on this reusable cache, we further introduce a dynamic compression mechanism that uses environment-side attention scores to retain informative tokens and adaptively allocates a fixed cache budget across Transformer layers. Experiments on the WAIR-D dataset show that the proposed method reduces KV memory from 0.2500 MB to 0.1406 MB, while the delay NMSE increases only slightly from 0.0673 to 0.0678. Code is available at https://github.com/LZhaoyang/kv.

Reinforcement learning with verifiable rewards (RLVR) enables large language models (LLMs) to continuously self-improve their reasoning capabilities. Compared to RLVR's scalar rewards, recent self-supervised on-policy distillation methods leverage denser supervision signals. However, both paradigms suffer from exploration bottlenecks, struggling to escape local optima when the policy fails to sample valid solutions. Relying on external demonstrations can bypass this issue; however, the off-policy nature hinders genuine self-improvement and the mismatched-distribution reasoning can not be fully exploited. In this work, we propose Self-Evolution Policy Optimization (SEPO), a framework that drives continuous reasoning capability evolution via guided exploration and in-distribution exploitation. Specifically, SEPO utilizes labels to elicit correct reasoning from the policy, unlocking the exploration potential. By treating current policy conditioned on this in-distribution reasoning as a self-teacher, SEPO distills the newly discovered solution back into the policy. Through this iterative cycle of policy-intrinsic exploration and exploitation, SEPO achieves robust and continuous capability evolution. Across scientific, tool-use, and complex logical reasoning domains, SEPO consistently achieves the best performance under identical training time or steps, demonstrating superior efficiency and convergence ceilings. Notably, SEPO maintains stable improvements even in scenarios where baselines severely underperform or fail entirely, such as on complex tasks or with small-scale models. Our code is available at https://github.com/Eric8932/SEPO.

On million-node graphs, the memory wall arrives before the compute wall: a GNN that never fits in VRAM cannot deliver the all-node inference its accuracy promises. The standard remedy, pruning, has always been an accuracy tax - paid once and never refunded. DPR refuses this framing. Its central insight is that the mask that shrinks training can also scaffold recovery: one compact node set bounds both computations of graph learning, serving as the encoder's training graph and, at inference, as the anchor from which a sparse structure-aware decoder rebuilds every pruned node (double-layer sparsification). The order is inverted - prune for efficiency, then reconstruct for accuracy - and the tax is repaid: the deployed 30%-budget model trains in 1,450 MiB of peak memory on one 6 GB GPU and recovers 96.9% of its dense-reconstruction reference, beating a same-GPU rerun of the edge-sparsification baseline DSpar by +1.36 pp at 50.1% lower peak training memory; against the same-protocol full-batch no-prune encoder the deployed point pays a measured 2.64 pp accuracy tax for a 2.8x smaller training footprint. Pruning without reconstruction leaves the encoder at 0.378 within the same 40-epoch window (Fig. 4), and all-node serving runs 46x faster than sampling-based GraphSAGE, giving up approx. 5.4 pp of test accuracy. The framework scales to a 2.45M-node production graph at 6,285 MiB of peak training memory; extended variants (long-range context, external text features, a CAN-bus application) are reported as extensions and do not compete with the deployed method, and the 41.8%-budget M4 point, scored on a coarsened super-node graph with its own split, is reported separately rather than compared with the original-node baselines. DPR thus turns aggressive node pruning from an accuracy tax into a safe training accelerator for VRAM-limited all-node inference.

Legacy C codebases remain central to system software and performance-critical infrastructure, but their weak memory safety imposes security and maintenance risks. Recent C-to-Rust translation systems, including LLM-based translators, make automated migration increasingly feasible, yet they still struggle to recover repository-level intent, avoid raw pointers and unsafe constructs, maintain cross-file interfaces, and repair compilation and validation failures efficiently. We propose SpeC2Rust, a specification-driven pipeline for repository-level C-to-Rust translation. It has three stages: (1) Project-Intent Structuring, which decomposes a C repository into semantically coherent modules and generates persistent repository- and module-level specifications; (2) Spec-Guided Generation, which translates each Rust module against these specifications with a live symbol registry to preserve cross-file consistency; and (3) Layered-Feedback Repair, which batches repairs and injects only failure-relevant diagnostics, specification context, and source evidence for efficient compilation and test-driven correction. We evaluate SpeC2Rust on eight repository-level C projects. It generates Rust projects that compile and pass the original end-to-end tests, with no raw-pointer declarations, raw-pointer dereferences, or unsafe lines. On projects shared with an LLM-based baseline, SpeC2Rust reduces cost by 66.1% and input-token usage by 84.7%.

Parallel ensembles of successive cancellation (SC)-style decoders improve error-rate performance by decoding multiple representations of the received vector. Conventional ensembles, however, keep every member active until completion. To reduce computational cost while retaining the diversity gain of ensemble decoding, we propose early ensemble selection (EES), which ranks members using intermediate decoding states and retains a prescribed numbers of members at each predefined selection stage. Within the EES framework, we propose two selection rules, namely path-metric-based selection (PM-EES) and neural-network-based selection (NN-EES). We analytically characterize the computational cost of EES through stage-dependent decoder log-likelihood ratio (LLR) update counts and provide an information-theoretically motivated criterion for choosing selection stages. Simulations on the \((128,64)\) extended BCH code and the \(\operatorname{RM}(4,8)\) code show that suitable selection schedules approach full-ensemble error-rate performance with fewer decoder LLR updates.

Fine-tuning adapts pretrained models to new tasks, but identifying which weight changes support the resulting behavior remains difficult. Weight-space interpretability assigns semantic descriptions to these changes, called fine-tuning deltas. Such descriptions leave open how much task performance depends on the identified updates, particularly when other updates can support the same predictions. We address this gap by pairing reversion, which restores selected parameters to their pretrained values, with retention, which keeps only selected fine-tuning updates. We evaluate Contrastive Language-Image Pretraining (CLIP) models on synthetic financial charts, Flowers102, EuroSAT, and FGVCAircraft with ViT-L/14, and on Flowers102 with ViT-B/16, using a shared recipe that tunes the final six visual blocks. Our contributions are threefold. First, attention value-output (VO) updates alone recover 43-87% of the accuracy improvement, although reverting them leaves chart accuracy unchanged. Cross-entropy loss preserves this contrast on Charts and EuroSAT, while correct-class score margins reveal additional contributions. Second, VO weights recover more loss improvement than all ten equally sized, layer-matched random subsets in each setting. VO and multilayer perceptron (MLP) retentions jointly correct 25-81% of pretrained errors, demonstrating overlapping recovered behavior, which we call operational redundancy. Third, removing semantically selected directions reduces VO-retained recovery more than matched-random removal in all five settings. Together, these results distinguish semantic alignment, removal cost, and retained recovery as complementary evidence about the functional role of fine-tuning updates within the evaluated CLIP recipe.

Long-context language models can increasingly process entire documents, making full-context prompting a practical alternative to passage retrieval. However, greater capacity determines how much information a model can receive, but not which information best supports a query. This distinction motivates rethinking retrieval: when the full document already fits, can selecting and organizing its evidence improve answering? The challenge is to retain supporting facts and relations while keeping the context focused on the question. To reconcile these goals, we introduce \ours, a training-free framework that uses semantic abstraction to guide selection while retaining original text for answering. \ours constructs a reusable semantic hierarchy from the full document in a single generation, linking node descriptions to source spans. Given a query, structural--semantic scoring and fine-grained lexical attribution nominate shared source nodes; their union with ancestor context preserves structural relations and complete local text. Across four frontier answer models, \ours improves average QA-F1 by 1.72 points over full-context prompting on three LongBench tasks, surpassing the evaluated baselines on the four-model average while reducing answer-context length by 58.55\%. Our code is available at the $\href{https://anonymous.4open.science/r/treefuse-ECFE}{repository}$.

Deformable linear objects (DLO) pose a coupled perception and control challenge: their slender geometry makes precise localization difficult, while their deformations depend on object-specific physical properties. To address this challenge, we propose PS-DLO, a property- and spatially aware DLO manipulation policy. First, contrastive image-text learning maps the initial observation to categorical descriptions of DLO properties, providing explicit physical and structural priors for action generation. Second, SAM3 extracts mask-guided features from future observations, which are distilled into the policy's action expert to encode task-relevant DLO regions and their subsequent configurations. We evaluate PS-DLO on nine real-world tasks across single-arm and bimanual platforms, training a shared multi-task policy for each embodiment. PS-DLO achieves an average success rate of 68.9%, exceeding the strongest baseline by 20.0 percentage points. In generalization experiments, PS-DLO achieves higher success rates than the baseline in six of seven visual-shift conditions on butterfly-knot untying. These results demonstrate that combining object-property priors with future spatial supervision improves multi-task DLO manipulation and robustness to visual variation.

Optical character recognition (OCR) systems depend strongly on the scale and diversity of their training data. One approach to document OCR first analyzes the page layout and extracts text-line images, which are then transcribed by a recognition model. Multilingual line-level recognition involving Persian, Arabic, and English presents additional challenges because different writing directions, fonts, and styles may coexist within a single line. This paper introduces MuSTEL, a large-scale synthetic multilingual OCR dataset containing more than 14.5 million annotated text-line images, organized into a training partition and predefined small and large test sets. Samples are generated from natural-language corpora, including multilingual content already present in the source texts. The generation framework incorporates capability-aware font selection, within-line font and style changes, mixed-direction text rendering, highlighting, and underlining. A validated repository of 2,009 font files from 711 families supports broad typographic coverage, while varied backgrounds, blur, geometric distortion, skew, stroke width, and image save quality provide diverse document-like appearances. Statistical analysis characterizes the dataset's language composition, transcription lengths, lexical and character-sequence diversity, image dimensions, typography, and rendering conditions. MuSTEL provides a resource for training and evaluating multilingual line-level OCR systems and investigating recognition robustness across languages, writing directions, and visual conditions.

While Multimodal Large Language Models (MLLMs) excel at zero-shot reasoning, their spatial capabilities are primarily evaluated through static, passive observation. In complex, human-centric scenes, occlusion often hides crucial geometric evidence, and passive benchmarks force models to guess rather than recognize their own visual uncertainty. To bridge this gap, we introduce SATE-Bench, a novel interactive benchmark comprising 316 real-world scenes and 2,525 multi-turn QA pairs to evaluate visual self-awareness and active exocentric view-exploration. It combines a Passive multi-view control with an Active exploration setting, where models start from a severely occluded view and must iteratively query adjacent viewpoints until they gather enough evidence to fully ``sate'' the question, scored by a step-penalized efficiency reward. Evaluating state-of-the-art MLLMs reveals that they terminate exploration prematurely while reporting high confidence, and struggle with fine-grained 3D geometric tasks; matched-view controls further attribute the Active--Passive gap primarily to insufficient evidence acquisition. Moreover, fine-tuning with Group Relative Policy Optimization (GRPO) shows that rewarding view-seeking increases active exploration but lowers overall accuracy, as models over-explore and fail to fuse expanding multi-view contexts across turns. These findings indicate that scaling passive multi-view perception does not intrinsically yield active spatial reasoning, highlighting the need for models that know where to look, how to integrate what they see, and when to stop. Code and data are available at https://gtc-gh.github.io/SATE-Bench/

Vision-language models now grade robot videos, and text placed beside the frames can change their verdicts. On clips where the robot is losing progress, we add one log line saying the robot is making progress. Across Gemini, Qwen and GPT judges, this flips correct verdicts to progress 10 to 35 percentage points more often than a neutral line does. A one-line warning that the log may be wrong removes most of this effect. Selected fake "task complete" claims flip half of Qwen's correct verdicts, against a tenth for unrelated log lines. They have no detectable effect on Gemini.

Within the broader landscape of artificial intelligence architectures, this paper introduces a compact, edge-centric neural framework that resolves the computational bottlenecks of modern edge-activated paradigms like Kolmogorov-Arnold Networks (KANs) while retaining their expressiveness over traditional Multi-Layer Perceptrons (MLPs). While KANs demonstrate remarkable representational capacity by placing learnable univariate functions on connection edges rather than hidden nodes, their reliance on non-local B-splines introduces severe optimization hurdles, high memory overhead, and poor hardware scaling. To bridge this gap, we propose the Parametric Tanh Connection Function, a smooth, bounded architectural primitive defined as $a_{\tanh}(x;δ,α,β,γ) = δ+ α\tanh(βx + γ)$. This formulation unifies offset, amplitude, local slope, and horizontal displacement into a lightweight, 4-parameter differentiable mapping optimized edge-by-edge via backpropagation. Evaluated on dense architectures using MNIST and CIFAR-10, our parametric tanh connection model achieves significant absolute accuracy gains over traditional affine baselines-reaching 95.63% $\pm$ 0.08% on MNIST and 44.74% $\pm$ 0.18% on CIFAR-10. Crucially, it matches or exceeds the expressiveness of unconstrained 7-parameter exponential connection baselines while requiring up to 85% less training runtime. By replacing complex splines or hyper-flexible exponents with a bounded, analytically smooth hyperbolic tangent kernel, this work delivers a highly scalable, hardware-friendly compromise for heterogeneous connection-level deep learning.