The paper introduces an autonomous AI workflow that uses specialized large language model agents to translate legacy Fortran code into performance‑portable Kokkos C++ programs. The pipeline handles translation, validation, compilation, execution, testing, debugging, and optimization, and has been shown to modernize benchmark kernels across heterogeneous GPU‑accelerated hardware. Experiments demonstrate that paid OpenAI models can produce optimized Kokkos code for only a few dollars, outperforming Fortran baselines, while open‑source models often fail to generate functional code.
By Sparsh Gupta, Kamalavasan Kamalakkannan, Maxim Moraru, Galen Shipman, Patrick Diehl
HDVS introduces a heterogeneous dual-branch voting supervision framework for semi-supervised semantic segmentation. It enhances pseudo-label reliability by correlating labels from diverse branches and uses a feature communication module to inject perturbations, promoting prediction diversity while preserving convergence. Experiments on PASCAL VOC 2012 and Cityscapes show HDVS outperforms current state‑of‑the‑art methods.
By Yongqi Shan, Yunzhi Zhuge, Huchuan Lu
V-LUMEN introduces a visual lookup memory for vision‑language models by discretizing visual features with a frozen codebook, creating spatially aware memory keys via hashing, and modulating them with text‑conditioned relevance signals. The retrieved embeddings are fused into a frozen Qwen3‑VL‑2B backbone through a gated residual pathway, yielding substantial improvements on five reasoning‑intensive benchmarks and zero‑shot transfer to three disjoint image datasets. At a matched parameter budget, reallocating capacity from LoRA or MoE adapters to V‑LUMEN preserves accuracy on two backbones, demonstrating that external visual memory offers an independent capacity axis.
whyItMatters":"The paper shows that adding a non‑parametric visual lookup memory can significantly boost vision‑language model performance without increasing per‑token compute, offering a new way to scale such models."
By Miso Choi, Daekeun Kim, Eunji Kim, Jungbeom Lee
The paper introduces DCKV, a differentiable codec designed to compress the key-value (KV) cache used in hybrid language models. DCKV jointly optimizes its transform, quantizer, and entropy models—conditioned on prompt tokens and neighboring positions—directly against a frozen language model, balancing compression rate with minimal impact on output quality. Experiments on Qwen3.5-2B demonstrate that DCKV can reduce the storage requirement for a compressed token by over 55× compared to bf16, with a loss of no more than 1.8 points on long‑context benchmarks, and by over 100× with a loss of no more than 3.6 points.
By Egor Maximov, Aleksei Merkelov, Vasiliy Alekseev, Ivan Krivorotov, Valentin Beliaev, Nikita Arsenin, Dmitry Bocharov, Aleksei Goncharov
LearnPrunerV2 is a fully learnable token‑pruning framework for Vision‑Language Models that replaces heuristic pruning with lightweight modules in every backbone layer. These modules estimate token importance by combining visual saliency and query relevance via a shadow token, revealing a consistent staircase redundancy pattern across layers. The method introduces Budget‑Deterministic Pruning to set fixed per‑layer budgets, achieving state‑of‑the‑art efficiency while preserving 99.8% of performance even after pruning 77.8% of visual tokens on LLaVA‑v1.5.
By Rinyoichi Takezoe, Yaqian Li, Weiren Wang, Zihao Bo, Huihong Shi, Anzhou Hou, Mo Guang, Jiale Yan, Kaiwen Long
VideoLVR introduces a method for multimodal large language models to maintain visual grounding during chain-of-thought reasoning by constructing latent visual evidence from local video regions at relevant moments. It uses a visual token prediction objective that unifies visual evidence reconstruction and text generation, allowing models to interleave text tokens with latent visual states. The approach is further extended with reinforcement learning, treating visual tokens as actions and optimizing policies with final-answer rewards, thereby improving video understanding without expanding the multimodal context.
By Shilong Li, Xingyuan Bu, Hao Shi, Yun Xing, Weixin Feng, Qiangpeng Yang, Shilei Wen
ALADDIN is a one‑step face‑swapping method that distills a pre‑trained multi‑step diffusion model into a fast generator trained adversarially with identity and target‑preservation constraints. It offers a controllable trade‑off between identity transfer and attribute preservation via triplet pseudo‑label supervision, and it outperforms other one‑step methods in realism and target preservation while matching the best multi‑step model with far fewer evaluations and training pairs. The authors also uncover a bias in standard evaluation protocols and re‑evaluate baselines with an independent identity encoder.
By Ivan Shchekotov, Timofei Gritsaev, Viacheslav Pirogov, Dmitry Vetrov
ReferStream is a new framework for interactive subject injection in streaming video generation that does not require additional training. It constructs reference features using the current scene and injects them via a transient cache, then restores the global cache to preserve existing context. The authors also introduce ReferStream Bench, a benchmark with scheduled reference arrivals and evolving instructions, and show that their method outperforms baselines in text adherence, subject similarity, and count accuracy.
By Shuolin Xu, Guanyuan Li, Boyu Li, Kelu Yao, Xiaogang Xu, Hongchuan Yu, Shuicheng Yan, Xiaobin Hu
Difficulty-Estimated Policy Optimization (DEPO) is a new framework that improves the efficiency and robustness of reasoning alignment in Large Reasoning Models. It introduces an online Difficulty Estimator that filters training data before rollouts, ensuring computational resources focus on high‑learning‑potential samples. Empirical results show DEPO achieves about 1.86× faster rollouts than DAPO and outperforms GRPO on benchmarks at similar training costs.
By Yu Zhao, Fan Jiang, Tianle Liu, Bo Zeng, Xintong Wang, Jiaxuan Teng, Yu Liu, Yichao Du
PRISM is a tile‑level benchmark for pathology foundation models that evaluates reliability across four dimensions: model choice, label fraction, calibration, and out‑of‑distribution transfer. It tests eight models on six datasets at six label fractions (1%–100%) with 864 linear probe runs, plus a five‑hospital Camelyon17 transfer study. The benchmark reveals four key phenomena, including the changing relationship between AUROC and calibration at low labels, collapse of linear probes on harder datasets, the impact of source‑domain labels on target calibration, and limitations of temperature scaling.
By Alicem Koyun
The paper introduces QuadPose, a method that estimates the full 18-DoF pose of a quadruped robot—including its 6-DoF base pose and 12-DoF joint angles—from sequential monocular images. It also presents QuadSyn, the first benchmark dataset for camera-to-robot pose estimation in quadruped systems. On this benchmark, QuadPose achieves high accuracy metrics such as 92.18% PCK@2.5px and 3.65° mean joint‑angle error while running at about 33 FPS.
By Seobin Hwang, Ji-Seong Beom, Yeong-Jun Cho
EvoIn is a fine‑tuning framework that improves agent decision‑making by evolving new procedures from execution traces, validating them in the harness, and then rewriting the traces to internalize the logic into the model. The approach removes reliance on the evolved harness at inference time, enabling stronger decision‑making that generalizes to unseen tasks. Experiments show consistent gains, raising in‑domain pass rates by 10.9 points and out‑of‑domain by 9.2 points, and the method works across different model families.
By Shihan Dou, Shaofan Liu, Zhonghang Lu, Jiahang Lin, Shichun Liu, Binghai Wang, Jiajie Jin, Guanting Dong, Tao Gui, Qi Zhang, Xuanjing Huang
Ctrl-A (ControlAugment) is an automated data‑augmentation algorithm that uses control‑theory principles to adjust augmentation‑strength distributions online during training. It removes the need to pre‑set augmentation strengths by dynamically adapting each distribution based on a control‑loop and relative operation‑response curves. Experiments on CIFAR‑10, CIFAR‑100, SVHN‑core, and Tiny ImageNet with WideResNet‑28‑10 and ResNet show that Ctrl‑A achieves competitive performance across these image‑classification benchmarks.
By Jesper B. Christensen, Ciaran Bench, Spencer A. Thomas, H\"usn\"u Aslan, David Balslev-Harder, Nadia A. S. Smith, Alessandra Manzin
The paper introduces a decoupled memory control framework for spatially consistent long‑horizon video generation. By separating memory conditioning from generation, it uses a lightweight memory branch to learn spatial consistency from past observations and a per‑frame cross‑attention mechanism to condition each frame on the most relevant historical data. A camera‑aware gating mechanism further ensures memory is used only when meaningful references exist, resulting in lower training costs, higher spatial consistency, and state‑of‑the‑art visual quality.
By Yanjun Guo, Zhengqiang Zhang, Pengfei Wang, Xinyue Liang, Zhiyuan Ma, Lei Zhang
MixSIS introduces a mixed-supervised instance segmentation framework that permits a flexible combination of image, point, box, and mask annotations within a dataset, addressing the limitation of existing weakly-supervised methods that focus on a single annotation level. The authors present AnnsMixer, a baseline system that generates pseudo mask annotations from all annotation levels, even within a single image, and demonstrate its effectiveness on both WSIS and MixSIS tasks. Experiments and case studies show that MixSIS and AnnsMixer can reduce annotation time by up to 90% or improve segmentation results by 34% using simple annotation rules or heterogeneous data.
By Christian Wilms, Emre G\"ulsoylu, Sven Schreiber, Simone Frintrop
ReGuLaR introduces a relation‑grounded latent reasoning framework for large vision‑language models, explicitly anchoring latent states to key objects and their inter‑object relations. It employs a training‑time ReGFormer to focus reasoning on question‑relevant visual evidence, while inference proceeds without the ReGFormer. The authors also create the RGROUNDING‑351K dataset and demonstrate that ReGuLaR outperforms existing methods across diverse benchmarks, achieving state‑of‑the‑art results.
By Zihu Wang, Karthik Somayaji N. S, Hejia Geng, Peng Li
TALON introduces Token‑Aligned Lightweight Adapters for 6‑DoF spacecraft pose estimation, inserting spatiotemporal 3D adapters before self‑attention layers of a frozen Vision Transformer. A patch‑token alignment loss grounds adapted features to keypoint structure, enabling the frozen backbone to reason over temporally enriched tokens with minimal added parameters. On the SPADES and SwissCube datasets, TALON cuts pose error by 50% and improves ADD‑0.1d accuracy by 21.8%, respectively, while also achieving significant gains in zero‑shot sim‑to‑real cross‑domain evaluation.
By Abid Ali, Nidhal Eddine Chenni, Arunkumar Rathinam, Djamila Aouada
Goal2Pixel introduces a pixel‑based approach for vision‑language navigation in continuous environments, replacing low‑level action prediction with navigable pixel grounding. The model predicts a visible pixel on the image plane, which is back‑projected into a 3D waypoint for forward motion, while auxiliary regions handle turning and stopping. Coupled with a visibility‑aware keyframe memory (ViKeyMem) and semantic embeddings, Goal2Pixel achieves state‑of‑the‑art performance on R2R‑CE and RxR‑CE with significantly fewer VLM calls.
By Muyi Bao, Yuxin Cai, Hang Xu, Zongtai Li, Jinxi He, Jingfan Tang, Chen Lv, Ji Zhang, Yaqi Xie, Wenshan Wang
LiveSearch is a new financial benchmark that evaluates agents on up‑to‑date facts by storing each question as executable verifier code that calls a professional financial data service and extracts the answer at evaluation time. The gold answer is recomputed daily without human or language‑model intervention, ensuring that models must retrieve current data rather than rely on memorized, stale answers. The benchmark automatically generates these verifier programs from real tool responses, and experiments show that most questions change their gold value daily, revealing dynamic correctness that aggregate metrics can miss.
By Sinuo Wang, Wentao Zhao, Jiacheng Lu, Rui Sun, Beidi Luan, Qiming Shi, Jing Li, Daxin Jiang, Zuo Bai
The paper presents a reliability‑aware benchmarking framework for short‑term glycemic risk prediction, using 5,503,441 prediction anchors from 736 type‑1 diabetes patients. Three neural models—a Hybrid LSTM‑Transformer, a Multimodal Digital Twin, and a State‑Updating Digital Twin—were evaluated on 30‑, 60‑, and 120‑minute horizons for hypoglycemia and hyperglycemia, with identical patient‑level validation. While no significant differences in predictive accuracy were found, the models displayed distinct reliability traits: the Multimodal Digital Twin had the lowest Brier score, the State‑Updating Digital Twin the lowest expected calibration error, and the Hybrid LSTM‑Transformer the strongest explanation faithfulness.
By Pir Bakhsh Khokhar, Carmine Gravino, Fabio Palomba