UniLayDiff is a Unified Diffusion Transformer that tackles content‑aware layout generation across a wide range of tasks—unconditional, element‑type, size, and relationship‑conditioned generation—using a single end‑to‑end trainable model. It treats layout constraints as a distinct modality within a Multi‑Modal Diffusion Transformer framework, capturing interactions among background images, layout elements, and constraints. The model is further refined with LoRA fine‑tuning to incorporate relation constraints, achieving state‑of‑the‑art performance and, according to the authors, the first unified solution for all content‑aware layout generation tasks.
By Zeyang Liu, Le Wang, Sanping Zhou, Yuxuan Wu, Xiaolong Sun, Gang Hua, Haoxiang Li
The paper introduces a holistic framework that jointly optimizes architecture, token, and bit-width for Vision Transformers (ViTs). It uses Neural Architecture Search (AutoFormer) to find compact backbones, token merging (ToMe) to reduce information processing, and fp16 mixed‑precision inference to accelerate operations. Experiments on ImageNet‑1K and a real‑world 3D X‑ray semiconductor defect dataset show over 10× improvements in throughput and reductions in parameters, FLOPs, and energy while preserving accuracy.
By Phat Nguyen, Xue Geng, Kaixin Xu, Wang Zhe, Xulei Yang, Ngai-Man Cheung
arXiv:2609.12397v1 Announce Type: new
Abstract: Multi-modal image generation, particularly subject-driven customization, has garnered growing attention in recent years. Despite the rapid advancement...
By Danning Zhang, Yijing Lin, Shuhan Zhuang, Mengqi Huang, Shaojin Wu, Shancheng Fang, Zhendong Mao
GenOR‑Twin is a neuro‑symbolic middleware that translates unstructured operational logs into formal constraints for mathematical optimization. It uses Large Language Models as semantic translators, not direct solvers, preserving the feasibility guarantees of exact combinatorial methods. The system dynamically injects constraints in real time, couples operational observations with a virtual model, and adapts its decision policy between schedule repair and full re‑optimization, demonstrating applicability across six optimization domains.
By Rahimeh Neamatian Monemi, Shahin Gelareh, Lubin Cui, Nelson Maculan
The paper introduces BQ-LoRA, a low‑rank adaptation framework for large language model agents that consolidates multiple LoRA adapters into a single one. It uses a behavior quotient manifold to balance trajectory updates and a decision‑preserving compression module to keep updates within a fixed rank budget while minimizing distortion of decision distributions. Experiments on AppWorld and BrowseComp‑Plus demonstrate that BQ‑LoRA outperforms standard LoRA and other low‑rank methods, with ablations confirming the benefits of both components.
By Pengyang Zhou, Xiaobin Tu, Zhengxi Liu, Rongkun Xue, Haochen Li, Miancan Liu, Ziyuan Chen, Yinggui Wang, Jinkui Ren, Xiantao Zhang
The paper introduces a quantization technique for tabular foundation models that focuses on converting queries, keys, and values to FP8 and employing explicit FP8 matrix multiplication to accelerate attention calculations. It emphasizes aligning quantization errors between training and test rows to avoid accuracy loss, and demonstrates up to 1.7× speedup over 16‑bit kernels with no significant accuracy degradation on TabPFN‑v3 and TabICLv2 across TabArena and BeyondArena.
By Jonas M. K\"ubler, Benjamin J\"ager, Klemens Fl\"oge, Noah Hollmann, Frank Hutter
The paper investigates the environmental impact of running large language models (LLMs) on mobile devices. It evaluates 18 different LLM configurations on two smartphones and a server, measuring energy per token, latency, accuracy, and battery-cycle consumption. Findings reveal that on-device inference is about three times less energy‑efficient than batched server inference, that energy consumption varies non‑monotonically with quantization bit‑width, and that most models are not on the Pareto front of accuracy and energy efficiency. The study concludes that local AI is not inherently more sustainable than cloud inference, with the majority of environmental impact stemming from device embodied carbon.
By \'Edouard Gu\'egain, Tristan Coignion
R2VC is a modular fact‑checking system that separates retrieval, reasoning, verification, and confidence calibration. It uses hybrid sparse‑plus‑dense Wikipedia retrieval, a fine‑tuned generator for structured verdicts, an NLI cross‑encoder for selecting evidence‑based candidates, and a lightweight calibrator for confidence and abstention. On the FEVER benchmark, R2VC improves accuracy by 13.74% over a baseline and shows that verifier‑based candidate selection and calibration are key contributors to performance.
By Dhruv Dixit, Paritosh Pandey
Robion is a new serving and management system designed to run Vision‑Language‑Action (VLA) models on multi‑GPU edge servers for robot factories. It splits the VLM and ADiT stages within a single GPU, shares streams across multiple models, and prioritizes requests by remaining SLO time, enabling high robot load while meeting strict latency requirements. In experiments, Robion achieves 6.7× higher robot load than vLLM‑Omni and 1.5× higher than a monolithic pipeline, and can serve 64 robots on a 4‑GPU server with 98% SLO attainment.
By Dionysios Adamopoulos, Nattapol Chanpaisit, Basel Fakhri, Christina Giannoula
The article presents the first demonstration of an agentic TCAD calibration workflow for a fabricated bottom‑gate In–W–O transistor. Starting from measured transfer curves, the workflow uses measurement–TCAD residuals and local sensitivity tests to guide parameter corrections or additional physical models, accepting only updates that improve device metrics. For a 2%‑W reference device, five agent‑suggested updates reduced the multi‑metric device objective by 14.3× and achieved low errors in threshold voltage and on‑current across varying bias and geometry, showing model transferability and process sensitivity.
By Gyujun Jeong, Junmo Lee, Sungwon Cho, Woohyun Hwang, Kwangyou Seo, Suhwan Lim, Wanki Kim, Daewon Ha, Rishi Ranade, Kihang Youn, Ram Cherukuri, Yiyi Wang, Asif Khan, Shimeng Yu
GAUGE is a new offline protocol that evaluates whether the common practice of using an LLM-as-a-judge to rank task‑oriented agents actually aligns with a verifiable reward. Across 25 agents from six providers on two benchmarks, GAUGE finds that user satisfaction scores are largely uncorrelated with task success, and that the judge’s ranking loses precision when agents are closely matched in performance. The study highlights a gap between ranking validity and construct validity in current evaluation practices.
By Umesh Bodhwani, Thanh Tran, Kai Wei
The paper investigates whether small models distilled from larger ones behave similarly when using byte versus token tokenization. It introduces two methods—Marginalize‑It (approximate) and End‑Of‑Token (exact)—to convert token logits to byte logits, and conducts a large‑scale study on decoder‑only dense transformers ranging from 1 billion to 1 trillion bytes of data. Results show that while token‑based models excel early, byte‑based models eventually surpass them with more compute, achieving higher performance ceilings, greater data efficiency, and lower logit storage costs.
By Kalyani Marathe, Artidoro Pagnoni, Tomasz Limisiewicz, Margaret Li, Mike Lewis, Luke Zettlemoyer, Srinivasan Iyer
The paper reports six submissions by the ESTS team to the WMT26 Model Compression Shared Task for English–Simplified Chinese and English–Egyptian Arabic. Each submission offers three compression operating points derived from GPT‑OSS‑20B, using routing‑informed expert pruning, cross‑lingual routing divergence for capacity allocation, and MXFP4 quantization of retained expert projection weights. The resulting models, ranging from 4.186 B to 7.770 B parameters, are fine‑tuned on GPT‑5.1 synthetic data and evaluated internally with xCOMET‑XL.
By Liu O. Martin, Lucas Bandarkar, Nanyun Peng
BEAST is the first Bayesian Swin Transformer for atmospheric forecasting at 0.25° global resolution, capable of quantifying both aleatoric and epistemic uncertainty. The authors introduce a 4D‑parallelization scheme with domain‑tensor‑parallelism and a novel uncertainty parallel method, allowing a 2.4‑billion‑parameter model to reach 3.96 EFLOP/s on 20,480 NVIDIA GH200 GPUs. A 700‑million‑parameter BEAST trained on 40 years of data achieves predictive skill comparable to state‑of‑the‑art probabilistic AI models, predicts extreme events with high accuracy, and generates large ensembles 3–4× faster than the current best AI model.
By Deifilia Kieckhefen, Juan Pedro Guti\'errez Hermosillo Muriedas, Lars Helge Heyen, Mathis Bode, Iida Hakulinen, Andreas Herten, Chelsea Maria John, Thorsten Kurth, Anni Moisala, Asena Karolin \"Ozdemir, Kaleb Phipps, Oskar Taubert, Arvid Weyrauch, Markus G\"otz, Charlotte Debus
The paper introduces Affective Agent, a three‑layer reference architecture designed for on‑device personalized intervention reasoning in wearable systems. It integrates a compact sub‑billion‑parameter language model with physiological data, context, and user history to determine when and how to intervene, all without cloud support or per‑user retraining. The architecture’s perception, personalization, and reasoning layers adapt through host‑managed structured memory evolution, and evaluation on simulated indoor environmental quality scenarios shows that memory‑driven personalization and two‑pass reasoning enhance intervention decisions.
By Reina Mun, Zishen Wan, Vijay Janapa Reddi
The paper introduces PA-CDM, a new position‑aware character detection matching metric for handwritten mathematical expression recognition that improves on existing render‑based and tree‑edit metrics by incorporating position‑forest encoding and divergence‑level weighting. It also presents StructPerturb v2.0, a benchmark of 1,340 controlled perturbation pairs, and a cross‑metric consistency protocol that includes a sensitivity matrix, a human study, and LLM‑judge calibration. In a six‑annotator study, PA‑CDM achieves the highest correlation with human judgments (Spearman rho = 0.9535) among seven automatic metrics, approaching the performance of a costly LLM judge while remaining deterministic and cost‑free.
By Shiliang Luo (East China Normal University)
The paper investigates how a single SM utilization metric can misrepresent the true workload of large language model (LLM) inference on Nvidia Hopper GPUs. By profiling vLLM with FlashAttention‑3 and cuBLASLt on an H100 NVL across various phases (cold prefill, warm prefill, and decode) and varying sequence length and batch size, the authors replace the single utilization figure with eight detailed counter‑validated views. These views, tied to specific Nsight Compute counters or formulas, reveal how factors such as fragment fill, occupancy limits, stall signatures, wave quantization, and kernel selection create utilization gaps across four production models and six per‑layer kernel roles.
By Mohammad Siavashi, Gerald Q. Maguire Jr., Dejan Kostic, Marco Chiesa
The paper discusses how large language models (LLMs) can be fine‑tuned with observational data to improve alignment with human preferences and business goals. It highlights that directly using such data can cause models to learn spurious correlations, and introduces DeconfoundLM, a method that removes known confounders from reward signals. Experiments show that DeconfoundLM better recovers causal relationships and outperforms baseline methods by over 16% in objective score when confounding is present.
By Erfan Loghmani
The paper introduces BenQ, a data‑free post‑training quantization (PTQ) method that leverages Benford’s Law to guide the construction of a log‑spaced codebook for transformer weights. BenQ selectively applies this codebook to linear layers while preserving higher precision for LayerNorm parameters, achieving consistent 4‑bit group‑wise PTQ improvements over uniform RTN and competitive results with NF4 across various models and tasks. The authors also explore dynamic activation quantization, noting that log‑spaced grids can mitigate RTN failures but that handling outliers remains crucial for reliable low‑bit activation PTQ.
By Arthur Negr\~ao, Pedro Silva, Vander L. S. Freitas, Gladston Moreira, Eduardo Luz
The paper introduces Spatial‑FAD, a few‑shot medical anomaly detection framework that fuses Vision‑Language Model (CLIP) semantics with spatial priors from Vision Foundation Models (DINO). A VFM‑enhanced adapter injects structural affinity into CLIP features, while a sliding‑window aggregation produces high‑resolution embeddings for finer lesion localization. Prototype‑enhanced support memory further improves efficiency and performance, yielding significant gains on Liver CT, Retinal OCT, and Brain MRI datasets, notably an 11.4% Dice improvement in 4‑shot scenarios.
By Juzheng Miao, Yuchen Yuan, Cheng Chen, Pheng-Ann Heng