The paper introduces GQ-FSL, a green quantized federated split learning framework designed for wireless edge networks. It uses stochastic quantization for both local training and wireless transmissions, allowing asymmetric precision between client and server submodels to balance device energy limits with global convergence. The authors develop energy models and a convergence bound for heterogeneous data, then formulate an optimization problem to set the DNN split point and precision levels, achieving lower energy consumption while meeting latency and accuracy targets.
By Idan Roth, Lutz Lampe
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
The study evaluates how different graph topologies influence the localization of the epileptogenic zone (EZ) from resting‑state stereo‑EEG recordings in 40 patients. Using a consistent learning model and leave‑one‑patient‑out validation, the authors compare dense graphs, anatomy‑ and geometry‑informed priors, budgeted sparsification, learned sparsification, and a new Region‑Bridge‑c topology. At about 30% edge retention, Region‑Bridge‑c achieves the highest mean PR‑AUC and ROC‑AUC while using roughly 69% fewer edges than a dense graph, indicating that graph construction significantly impacts EZ localization performance.
By Daniel Wendelken (University of Cincinnati, Cincinnati, USA), Brian Ervin (Cincinnati Children's Hospital Medical Center, Cincinnati, USA), Ravindra Arya (Cincinnati Children's Hospital Medical Center, Cincinnati, USA), Ali A. Minai (University of Cincinnati, Cincinnati, USA)
The paper introduces HB‑SJD, a batched Speculative Jacobi Decoding backend that accelerates visual on‑policy distillation (OPD) by allowing images to advance independently and processing multiple tokens in parallel without a draft model. HB‑SJD switches between Full and Compact execution as images finish, reducing rollout and overall training time while maintaining generation quality. Experiments with LlamaGen demonstrate significant speedups without altering the teacher, distillation objective, or optimization procedure.
By Bingqi Shan, Zhehao Yu, Kenhong Lin, Baoquan Zhang
The paper introduces geometric iterative retrieval, a new approach for resynthesizing high‑quality audio from coarse Residual Vector Quantization (RVQ) codec tokens. Instead of choosing between discrete token prediction or continuous regression, the method performs contrastive retrieval within the continuous codebook space, leveraging the RVQ hierarchy as an iterative decomposition. Experiments on speech and music codec restoration tasks demonstrate that this technique outperforms both single‑pass token prediction and one‑step regression baselines.
By Leo Schmidt-Traub, Fr\'ed\'eric Berdoz, Luca A. Lanzend\"orfer, Roger Wattenhofer
The paper investigates whether large, instruction‑following Vision‑Language Models (VLMs) can reliably perform zero‑shot image privacy classification. It compares three open‑source VLMs to specialized privacy models on two public benchmarks, evaluating accuracy, robustness to image degradations (compression, lighting changes, noise), inference speed, and parameter count. The findings show that while VLMs remain robust to perturbations, they are less accurate and significantly slower than smaller, purpose‑built privacy models, indicating that scaling alone does not guarantee effective privacy classification.
By Alina Elena Baia, Alessio Xompero, Andrea Cavallaro
The paper introduces DART‑SD, a framework for improving multi‑turn tool‑calling agents by respecting the diamond‑shaped topology of task sub‑goals. It models the execution as an Interaction‑State Transition Graph, identifies critical topological breakpoints during rollouts, and uses these to retrieve recovery references. A progressive self‑distillation process then applies localized supervision only on recovery steps, preserving valid reasoning prefixes and enhancing policy diversity.
By Hangrui Xu, Jiarui Wang, Yang Yang, Chuanbo Zhu, Fangda Chen, Ziqi Wu, Jingming Cai, Yan Song
The study investigates how post‑training quantization (PTQ) affects proactive interference (PI) in large language models. Using bitsandbytes, the authors compare FP16, INT8, and INT4/NF4 precision across three instruction‑tuned models and find that INT4 quantization markedly degrades accuracy under high interference, with INT8 also incurring a smaller penalty in two of the three models. The degradation is linked to increased same‑key intrusion errors and originates in the quantized transformer backbone rather than the output layer.
By Shayan Shahrabi-Farahani (Shahid Beheshti University, Tehran, Iran), Dara Rahmati (Shahid Beheshti University, Tehran, Iran)
arXiv:2608. 18147v1 Announce Type: cross Abstract: Adaptive stochastic quantization (ASQ) is a recently introduced quantization approach that optimizes the Mean Squared Error (MSE) for a given input while preserving unbiasedness.
By Ran Ben Basat, Yaniv Ben-Itzhak, Michael Mitzenmacher, Shay Vargaftik
FlashAttention-V is a blocked FlashAttention implementation optimized for scalable vector architectures, designed to reduce the memory bandwidth bottleneck of transformer attention on CPUs. By fusing operations, exploiting parallelism across attention heads, and inter‑head packing, it improves vector register utilization and memory locality, enabling efficient scaling from short to very long vectors. Benchmarks on TinyLlama, Llama 3.2, Qwen2.5, and Pythia‑410M show 22×–42× speedups over scalar FlashAttention in prefill and 8×–11× in decode on a Banana Pi BPI‑F3, while also revealing quantization‑related bottlenecks that limit long‑vector scalability.
By Sonia Rani Gupta, Nikela Papadopoulou, Miquel Peric\`as
The paper introduces Group‑Calibrated On‑Policy Distillation (GC‑OPD), a method that aligns token‑level teacher guidance with trajectory‑level verifier rewards for long‑context reasoning tasks. GC‑OPD normalizes rewards within rollout groups, uses the signed teacher‑verifier disagreement as a residual, and distributes 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.
By Zhu Zhang, Jixun Wang, Xiaoang Xu, Xiaorong Wang, Zihan Zhou, Zhiyuan Wang, Shuo Wang, Chaojun Xiao, Yuezhi Zhou
The paper presents a method for distributing large language model inference across multiple Intel AI PCs by splitting the model into pipeline shards, each pre‑compiled into an OpenVINO graph. Three key techniques—beam_idx Gather to enable GPU optimizations, speculative decoding on stateful models, and interleaved micro‑batching—allow a two‑node Llama 3.1 8B INT4 pipeline to serve two users at 1.79× the throughput of a single‑node model, while a four‑node deployment can run a 70B model that no single PC can hold. The authors provide code, benchmark logs, and reproduction scripts on GitHub.
By Tate Berenbaum, Muthaiah Venkatachalam
Online Conformal Anomaly Detection with Prediction-Powered Data Acquisition introduces C-PP-COAD, a framework that uses synthetic calibration data to reduce reliance on real-world calibration while maintaining assumption-free false discovery rate control. The method wraps any anomaly detection algorithm, converting its scores into conformal p-values for online testing. Experiments on synthetic and real datasets—including thyroid dysfunction, O‑RAN conflict, 5G intrusion, and UE throughput degradation—show that C-PP-COAD preserves FDR guarantees while significantly cutting the need for real calibration data.
By Amirmohammad Farzaneh, Osvaldo Simeone
The paper introduces ERASE, a training schedule that detaches subgraphs in a neural network to launch their backward passes early on separate CUDA streams, thereby overlapping backward and forward computations. By applying this technique to a lightweight transformer and a large‑scale click‑through‑rate model, the authors demonstrate up to a 9.51% increase in training throughput while maintaining performance metrics close to the baseline.
By Ergan Shang, Flavio Sales Truzzi
The paper introduces a Tsetlin Machine (TM)-based framework for Non‑Intrusive Load Monitoring (NILM) that can run in real time on resource‑constrained microcontrollers. By reformulating NILM as a classification problem, the authors achieve high accuracy—90 % precision and 96 % recall for two appliances, and 77 % precision and 80 % recall for four appliances on the REDD dataset—while keeping the trained model to only 18 KB of flash memory. On an ESP32, the system delivers inference in 0.43 ms, demonstrating its feasibility for privacy‑preserving edge deployment.
By Tianhang Tan, Han Wu, Tousif Rahman, Shengyu Duan, Alex Yakovlev, Rishad Shafik
The paper introduces DCGCNet, a dual-codebook graph collaborative network that jointly reconstructs ECG signals and classifies atrial fibrillation. It incorporates a local‑global contrastive module for noise‑invariant feature learning and an adaptive codebook vector quantizer to prevent codebook collapse. The model achieves state‑of‑the‑art intra‑dataset performance and consistently attains AUC > 0.98 across seven cross‑dataset settings, even under realistic noisy conditions.
By Hongtao Li, Jia Wei, Guoyao Li, Yuchen Lei, Guangnian Ma, Jia Xiao, Yuanjun Lai, Shuzhen Lv, Xueqiang Ouyang
ADEPT is a reinforcement‑learning framework that first pre‑trains a dexterous policy on a generic object reposing task and then post‑trains downstream policies using this pretrained behavior as a prior. The approach avoids relearning basic skills for each new task, and employs a stable post‑training recipe—behavior‑cloning distillation, critic warm‑up, and conservative on‑policy updates—to preserve the pretrained capabilities. ADEPT’s joint‑space Geometric Fabric mediates between the policy and the robot, enabling zero‑shot sim‑to‑real transfer on a 23‑DoF Kuka‑Allegro and a 29‑DoF Flexiv‑Sharpa, where the robots solve long‑horizon tasks from challenging initial states at human‑level speed.
By Jayjun Lee, Jessica Yin, Asif Rana, Nicholas Blauch, Sam Mady, Mohak Bhardwaj, Nima Fazeli, Nathan Ratliff, Karl Van Wyk, Ankur Handa
The paper presents a super‑resolution generative adversarial network (SRGAN) that boosts electron backscatter diffraction (EBSD) throughput for lithium‑ion battery electrode materials. Trained on LiNixMnyCozO2 cathode data, the SRGAN outperforms classical interpolation across 2×–12× upscaling, especially preserving small grains and realistic boundaries. A 5× upscaling yields a 25× speed‑up or larger field of view with acceptable errors in grain size and shape metrics.
By John Mangum, Andrew Glaws, Francois Usseglio-Viretta, Steven Spurgeon, Donal Finegan
The paper argues that catastrophic forgetting and loss of plasticity alone cannot explain why naive sequential training underperforms offline joint training. It introduces data co-observation as a third factor, showing that observing training data together consistently improves performance across supervised and self-supervised settings. The study also reinterprets common continual learning methods, suggesting that memory replay’s success stems from restoring co-observation benefits rather than merely mitigating forgetting.
By Timm Hess, Abhishek Jha, Gido M. van de Ven, Tinne Tuytelaars
GrabVG is a visual grounding framework for UAV imagery that tackles the challenges of small, densely packed, and visually similar objects. It splits the task into preattentive hypothesis search and graph‑attentive feature binding, using distillation‑guided proposals and a sparse graph to capture intra‑ and inter‑instance relationships. Experiments on AerialVG and AerialSense show that GrabVG achieves higher accuracy and speed, outperforming baselines by significant margins.
By Chaowei Wang, Yan Di, Jingjun Sun, Baozhe Liu, Jiaxu Tian, Yuheng Li, Guangqian Guo, Shan Gao