UMER is a Unified Multimodal Embedding and Ranking framework that combines contrastive embeddings with Pair‑Aware Discriminative Reasoning to improve universal multimodal retrieval. It replaces item‑wise reflection with pair‑wise comparison of query–candidate pairs, enabling explicit identification of matching and discrepancy evidence. A mutual distillation strategy transfers reliable pairwise preferences between the embedding and ranking components, and UMER achieves state‑of‑the‑art performance on the MMEB‑V2 benchmark while supporting budget‑adjustable inference.
By Libiao Chen, Xiyang Liu, Yanheng Wei, Tao Wang, Zhenyu Tang
Learning-based global point cloud registration has achieved remarkable progress, yet its reliance on geometric representations makes existing methods sensitive to variations in point density, scan pattern, viewpoint, and sensor characteristics. We propose CVSD-Reg, a robust global LiDAR registration framework that distills visual semantic priors from a vision foundation model into LiDAR representations.
The article describes a real‑world case of scaling an enterprise integration pipeline from 500 to 8,000 events per second. It emphasizes that during this throughput increase, two correctness guarantees were strictly maintained and never compromised. The post illustrates how to achieve high performance while preserving essential data integrity constraints.
By Yuelin Ou
The paper introduces Group-Calibrated On-Policy Distillation (GC‑OPD), a method that aligns token‑level teacher guidance with task‑level verifier rewards for long‑context reasoning. GC‑OPD normalizes verifier and OPD scores within rollout groups, uses their difference as a signed disagreement residual, and redistributes this residual across tokens via Relative‑Advantage‑Based Credit Assignment (RACA). Experiments on five long‑context benchmarks show that GC‑OPD improves Qwen3‑4B and Qwen3‑8B checkpoints from 29.08/35.12 to 40.47/44.65, outperforming vanilla OPD and demonstrating the effectiveness of group‑relative residual calibration.
The paper presents a super‑resolution generative adversarial network (SRGAN) that boosts the throughput of electron backscatter diffraction (EBSD) for Li‑ion battery electrode materials. By training on LiNixMnyCozO2 cathode data, the SRGAN outperforms classical interpolation across upscaling factors of 2× to 12×, especially preserving small grains and realistic boundaries. A 5× upscaling yields a 25× speed‑up or larger field of view with acceptable accuracy, reducing grain‑size errors to within ±15%.
GrabVG is a visual grounding framework for UAV imagery that tackles the challenges of small, densely packed, and visually similar objects by separating the task into preattentive hypothesis search and graph-attentive feature binding. It first generates a compact set of reliable object hypotheses using distillation-guided proposal induction and text-aware filtering, then constructs a sparse graph where language-guided visual cues and inter-instance topological relationships are jointly bound and propagated via graph attention. Experiments on AerialVG and AerialSense demonstrate that GrabVG achieves a strong accuracy–speed trade‑off, reaching 67.31% and 80.34% Acc@0.5 and outperforming baselines by 10.55 and 8.76 percentage points.
SkillForge is a self‑distillation framework that proactively builds project‑specific knowledge for large language model agents by synthesizing and resolving artificial issues derived from a repository’s test‑covered core functionalities. Rather than waiting for real issues to reveal knowledge gaps, SkillForge generates these synthetic problems, learns reusable entity‑grounded skills, and associates them with relevant repository entities. Experiments with both open‑source and closed‑source models show that this proactive knowledge acquisition consistently outperforms strong baselines in issue resolution tasks.
The paper investigates how sparsity impacts the expressivity of graph neural networks, focusing on relational and temporal variants. It extends the Strong Expressive Lottery Ticket Hypothesis to multi-relational and temporal domains, proving that sufficiently large RGNNs contain sparse subnetworks that preserve 1‑RWL expressivity and providing a probabilistic bound for random pruning. Experiments validate the theoretical bounds, compare them to empirical results on synthetic data, and explore the relationship between pre‑training expressivity, optimization behavior, and prediction quality on temporal and molecular benchmarks.
The paper presents a cost‑effective approach for industrial explainable‑recommendation systems by decoupling explanation generation from selection. Explanations are pre‑generated using six prompt styles and two commodity LLMs, then a lightweight CPU‑resident selector (e.g., LambdaRank) chooses the best one at request time, achieving sub‑100 ms latency without GPUs. Experiments on a 2,958‑pair Google Local subset and a 300‑pair MovieLens‑1M split show that pairwise ranking outperforms single‑action RL methods, while KG‑path selectors achieve near‑perfect unique‑output rates, and the overall end‑to‑end build cost is around $15 on commodity hardware.
DART‑SD introduces a diamond‑topology aware framework for training large language models to perform multi‑turn tool‑calling. It models the execution process as an Interaction‑State Transition Graph, identifies critical topological breakpoints, and retrieves recovery references to guide localized self‑distillation. Experiments show that this approach outperforms traditional full‑trajectory baselines on complex multi‑turn tool‑calling benchmarks.
arXiv:2608. 16947v1 Announce Type: cross Abstract: Huang, Lou, and Xiao introduced Dynamic Mixture-of-Experts Serving and gave an O(sqrt(log k))-competitive randomized algorithm for its integral primal problem, where k is the number of replica GPUs beyond the mandatory copy of each expert.
By Ian D'Ambrosio (Nth Research Collective)
MotoSafety is an edge-AI system designed to assess collision risk for two‑wheeler riders under varying time pressure. It is trained on a large dataset of 129,000 labeled time‑series sequences from 153 simulator rides, capturing 64 features related to vehicle dynamics, control inputs, proximity, and behavioral violations. The model achieves 94.97% accuracy and 99.33% ROC AUC, with only 1.15 M parameters and 0.135 ms latency, making it suitable for low‑cost CPU deployment and demonstrating strong transferability to other domains.
By Sumit S. Shevtekar, Chandresh K. Maurya, Gourab Sil, Subasish Das
The paper investigates why self‑distillation can sometimes worsen the reasoning abilities of large language models (LLMs). It finds that the process suppresses the model’s epistemic verbalization—its expression of uncertainty—leading to shorter but less accurate responses in mathematical reasoning tasks. Experiments on several LLMs show performance drops of up to 40%, especially on out‑of‑distribution problems where uncertainty expression is beneficial.
By Jeonghye Kim, Xufang Luo, Minbeom Kim, Sangmook Lee, Dohyung Kim, Jiwon Jeon, Dongsheng Li, Yuqing Yang
ImplicitTerrainV2 introduces a wavelet-guided, spatially adaptive neural representation for digital elevation models (DEMs). It uses a wavelet complexity field to localize high-frequency capacity to complex terrain, adaptive sampling to focus training, and gradient matching to preserve smooth manifold structure. After mixed-precision quantization and entropy coding, the model achieves 1.23 bpp with only a 0.28 dB PSNR loss, outperforming prior work by 5.70 dB while using 3.2× fewer parameters and training in 55 s per tile on a single GPU.
By Haoan Feng, Xin Xu, Leila De Floriani
The paper establishes a bounded‑differences inequality for functions of non‑isotropic Gaussian vectors, showing that the concentration bound depends on the condition number of the covariance matrix. It applies this result to demonstrate the subgaussianity of sign‑quantized linear maps, addressing a question posed by Simone Bombari. The authors also correct attribution to earlier work by Barber and Kolar and highlight the role of AI assistance in the discovery process.
By Guangyi Zou, Roman Vershynin
arXiv:2608. 17802v1 Announce Type: new Abstract: Let $\varepsilon_1,\ldots,\varepsilon_n$ be independent Rademacher signs and let $a=(a_1,\ldots,a_n)\in\R^n$ satisfy the normalization below.
By Peigan Gao, Jian Qian
LoopVLA introduces a recurrent Vision‑Language‑Action architecture that learns to refine multimodal representations, predict actions, and estimate when further refinement is unnecessary. By iteratively applying a shared Transformer block and producing a sufficiency score at each step, it decouples refinement from fixed layer indices and aligns confidence scores with action quality through a self‑supervised objective. Experiments on LIBERO, LIBERO‑Plus, and VLA‑Arena demonstrate that LoopVLA reduces model parameters by 45% and boosts inference throughput up to 1.7× while matching or surpassing strong baselines in task success.
By Boyang Shen, Kaixiang Yang, Hao Wang, Qiuyu Yu, Qiang Xie, Qiang Li, Zhiwei Wang
The paper introduces DNBNet, a Debiased Neural Basis-Function Network designed for irregular time series forecasting. It addresses two main limitations of existing methods: asymptotic bias from ignoring timestamp sampling density and limited adaptability of predefined basis functions. DNBNet employs importance sampling to correct bias, neural‑network parameterized basis functions for flexibility, a multi‑scale decomposition with mass‑aware fusion for sparse data, and a dual‑branch decoder, achieving strong performance across diverse real‑world datasets.
By Rongwen Li, Changjian Chen
The paper explores energy-aware knowledge distillation for large language models (LLMs) used in software engineering tasks such as clone detection, vulnerability prediction, and code summarization. It shows that the commonly used FLOPs metric does not reliably reflect actual energy consumption, and that using energy-surrogate models during distillation can reduce inference energy by up to 90% and memory usage by 86% with only modest accuracy loss. The study demonstrates that guiding distillation with direct energy estimates improves the sustainability and deployability of LLMs on consumer hardware.
By Enrique Barba Roque, Lu\'is Cruz, Annibale Panichella
SOD: Step-wise On-policy Distillation for Small Language Model Agents proposes a new framework that adaptively reweights distillation strength at each reasoning step based on step-level divergence. This approach mitigates cascading errors in tool-integrated reasoning by attenuating misleading teacher signals in high-divergence regions while preserving dense guidance where student and teacher align. Experiments on math, science, and code benchmarks show up to 20.86% improvement over the second-best baseline, with a 0.6B student scoring 26.13% on AIME 2025.
By Qiyong Zhong, Mao Zheng, Mingyang Song, Xin Lin, Jie Sun, Houcheng Jiang, Xiang Wang, Junfeng Fang