The paper introduces Musical Attention, a Transformer-based music generation model that incorporates meta-information such as bar numbers, key, signatures, and tempos into its attention mechanism. By representing each note with five events (pitch, bar number, onset, duration, velocity) plus three metadata elements, the model captures correlations among eight features, improving musical coherence and reducing repetition. Experiments show that Musical Attention outperforms prior methods like Full Attention and Strided Attention in coherence, variation, and overall quality, producing more diverse and harmonically consistent melodies.
By Shinnosuke Takasuka, Hideo Mukai
The paper introduces HypoKG, a unified biochemical knowledge graph built from KEGG, Rhea, and UniProt, and uses it to benchmark 13,200 biomedical hypotheses generated by six large language models (LLMs). By varying the biological information provided—source enzyme only, full biological path, or source and disease endpoint—the study finds that LLMs produce higher-scoring hypotheses when given minimal information, but these are less evidence‑grounded. When supplied with the full biological path, the models generate hypotheses that align more closely with known mechanistic relationships, a phenomenon the authors term evidence‑disciplined reasoning, which is confirmed by shuffling intermediate path steps.
"whyItMatters":"The study demonstrates that knowledge graphs can both uncover novel disease–enzyme pairs and guide LLMs to reason more accurately from evidence, improving the reliability of AI‑generated biomedical hypotheses."
By Dominic Okonkwo, Adetayo Okunoye, Ismailcem Budak Arpinar
GraphProfiler is an auditable LLM-based profiler that infers sensitive attributes such as age, income, and occupation from user-generated content by aggregating indirect cues across many posts. It represents each user's post history as a source-linked personal knowledge graph, allowing predictions to be traced back to specific posts, concepts, and relationships. The system achieves an 86.7% success rate on the SynthPAI benchmark and 84.6% on PANDORA, citing supporting evidence for over 98% of predictions and demonstrating that cited posts significantly contribute to attack success.
By Ahmed Sohair Khan, Estrid He, Chenglong Ma, Monica Wachowicz, Elham Naghizade
AMDKernelVault is an open HIP and Triton kernel corpus and training framework designed for AMD CDNA GPUs. It includes 62,153 verified HIP kernels, 39,893 Triton kernels, and 2,377 ROCm library QA entries, and introduces agent-driven pipelines (HIPKernelGen and TritonKernelGen) that convert PyTorch references into GPU kernels, compile, validate, and profile them on AMD hardware. The corpus was used to fine‑tune Qwen3-8B, achieving the highest correctness on several benchmarks such as PyTorch-to-HIP, TritonBench‑G, and ROCmBench under fixed evaluation budgets.
By Ji Liu, Saptarshi Majumder, Yiqing Huang, Wenwen Ouyang, Umang Pandey, Zeping Li, Chushi Chen, Zihao An, Puyuan Yang, Zekai Li, Sina Rafati, Ziqiong Liu, Pratik Prabhanjan Brahma, Dong Li, Zicheng Liu, Sharon Zhou, Emad Barsoum
The paper introduces PRISMA-LLM, a reporting framework for AI-assisted systematic reviews. It is based on an analysis of 888 review-automation papers, showing a shift toward LLM- and software-driven workflows and inconsistent reporting of evaluation and limitations. The framework separates implementation details from consequence-sensitive evaluation and limitation reporting.
By Miguel Zabaleta, Baihan Lin
The paper introduces the Atomic User Model (AUM), a structured representation of a user’s personality that separates a stable identity nucleus from four interpretable shells—psychological, cognitive & experiential, behavioural, and social—along with cross-shell entries for conflict and authenticity. It proposes using AUM as a retrieval index rather than a prompt prefix, enabling a task‑specific, budgeted retrieval of relevant fields at generation time. Experiments with simulated participants show that retrieving eight AUM fields improves style fidelity, preference accuracy, and user voice identification compared to flat preference notes, especially benefiting users whose default assistant performs poorly.
By B. Sankar, Deepthika S, Pawni Yadav, Amogh A S
DRS‑VPT is a feed‑forward transformer that performs image‑to‑scan registration by predicting the scan pose, point maps, and a coarse‑to‑fine feature pyramid for direct reprojective alignment. It unifies tasks like camera‑LiDAR calibration and indoor camera‑to‑map relocalization, achieving state‑of‑the‑art results in autonomous driving and competitive performance indoors without map‑specific training. The model also learns complex scan‑to‑image projection behaviors, such as occlusion of back‑facing points.
By Lanke Frank Tarimo Fu, Maurice Fallon
GRACE is a new framework for concept erasure in text-to-image diffusion models that uses a semantically weighted sensitive subspace to guide localized interventions and lightweight subspace-constrained adapters to avoid global semantic disruption. It replaces manual counterfactual prompts with an automatically decoupled safe-anchor mechanism and controls intervention strength through an energy-driven dynamic gating system. Experiments show GRACE improves NSFW reduction by 17.86% over five state-of-the-art methods while also reducing target CLIP Score and FID, indicating stronger concept suppression with better preservation of generative quality.
By Qinghui Gong, Yihuai Liang, Yuanlun Xie, Deepak Kumar Jain, Vitomir \v{S}truc, Zhengchun Zhou
Uni-HOI is a unified framework that learns the joint distribution among text, human motion, and object motion for 4D human‑object interaction (HOI). It uses large language models and two motion‑specific VQ‑VAEs to convert heterogeneous motion data into token sequences, enabling seamless integration of all three modalities. A two‑stage training strategy first captures correlations on a large‑scale HOI dataset and then fine‑tunes for specific tasks, achieving strong performance on text‑driven HOI generation, object‑motion‑driven human motion generation, and human‑motion‑driven object motion prediction.
By Mengfei Zhang, Jinlu Zhang, Zhigang Tu
CanvasAnneal is a curriculum‑guided reinforcement learning framework designed to improve Diffusion Language Models (DLMs) on complex reasoning and tool‑use tasks. It starts training by injecting reasoning traces from a stronger teacher model into the diffusion canvas, then gradually reduces this guidance so the model learns to generate reasoning independently. Experiments on mathematical reasoning and tool‑use benchmarks show that CanvasAnneal outperforms standard diffusion RL methods such as diffu‑GRPO on tasks like MATH500, Countdown, and Tau2, and accelerates reward improvement, though the gains vary by task.
By Blake Olson, Yuhang Song, Emmett McQuinn, Yuan Shangguan
MAxBench is a geometry‑agnostic benchmark for evaluating how well language models recover multinomial concept representations. The study compares ten localization methods across five geometry types, six concepts, and four models, finding that affine subspaces generally steer more reliably and recall more instances than rank‑one or linear subspaces. The results also show that manifold steering can match the best methods when applicable, and that no method consistently outperforms prompting for these complex concepts.
By Divya Appapogu, Freya Behrens, Yonatan Belinkov, Aaron Mueller
The article argues that large language models (LLMs) are inadequate for consequential quantitative tasks such as pricing, risk assessment, and medical triage because language is a lossy representation of quantitative data that cannot be reversed. It formalizes this limitation as a property of the training representation rather than model capacity and identifies three essential properties—reproducibility, traceable lineage to source records, and calibrated uncertainty—that language substrates cannot provide. The authors propose a new class of models, Large Quantitative Models (LQMs), designed to meet these requirements.
By Reuben Vandeventer, David Imrem, David J. Wild
The paper introduces AIM, a privacy‑aware memory framework that lets multi‑agent, multi‑user large language models manage both private and shared memory. AIM classifies data as private (user‑specific) or public (shared) and enforces index‑level access controls to protect sensitive information while enabling shared knowledge to improve coordination. The authors also present MUMBench, a new dataset for evaluating memory operations in multi‑user settings, and report high accuracy metrics for AIM on this benchmark.
By Zachary Johnson, Nigel Boachie Kumankumah, Somya Chatterjee, Tejas Sathyamurthi, Min Chen, Xinyi Alice Li, Xiao Wang, Emily Morgan Gelchie, Jessica Lin, Sadid A. Hasan, Sulaiman Vesal
The study develops a 1‑D ResNet classifier that uses a fixed 17‑channel hemodynamic representation derived from photoplethysmography (PPG) to predict in‑hospital stroke risk states up to six hours before clinical recognition. Using data from MIMIC‑III and MC‑MED, the model achieved F1‑scores ranging from 0.7956 to 0.9888 across 4‑, 5‑, and 6‑hour horizons, outperforming four non‑waveform clinical and structured‑EHR comparators in all cohort‑horizon settings. Retrospective analysis showed that the PPG model’s false‑positive rates on high‑risk non‑stroke controls could be reduced through persistence aggregation, though the study does not establish a calibrated bedside alarm or a clinically validated prediction lead time.
By Jiaming Liu, Cheng Ding, Jian Wu, Hongxia Xu, Daoqiang Zhang
HoliBench is a modular benchmarking and deployment toolkit that jointly measures accuracy, latency, and energy for foundation models across a wide range of devices, from single-board computers to GPU servers. It provides a platform abstraction layer that calibrates cross-device measurements and supports multiple model modalities, inference engines, and quantization levels. Using HoliBench, the authors evaluated 20 models on 7 device types, revealing tradeoffs such as limited latency gains from quantization on low‑precision hardware and diminishing accuracy returns relative to energy consumption, while also showing that single-model profiles can predict multi-model pipeline performance within a few percent.
By Inesh Chakrabarti, Zejun Xiong, Pragya Sharma, Mani Srivastava
The paper introduces LAST-CQ, a five-agent, training‑free, execution‑grounded framework for Text‑to‑Cypher that evaluates which components of an agentic pipeline contribute most to performance. Experiments on 2,471 live‑database queries across six backbones show that removing correction reduces execution‑BLEU by 3.1–12.3%, while substituting schema‑grounded feedback with raw error strings has negligible impact. Parallel sampling degrades quality by 10–11%, whereas failure detection and retry routing recover 91.7% of initially failed queries, highlighting that simple failure handling is more effective than sophisticated feedback or increased sampling.
By Ioannis Prokopiou, Athanasios Aidinis, Panagiotis-Christos Kyrmpatsos, Pantelis Vikatos
CluSTER is a cluster‑aware balanced sampling framework designed to improve the efficiency of instruction‑tuning for large language models. It reduces redundant computation by clustering data in gradient space and allocating samples across GPUs in a data‑parallel setting, while preserving the original distribution through weighted updates. Experiments show that CluSTER can cut training time by up to 69.6% with negligible loss in accuracy compared to existing sampling methods.
By Hyunjin Kim, Youngeun Nam, Jaemin Han, Wonhyeok Choi, Jae-Gil Lee
The paper introduces a weakly supervised framework for extracting dataset mentions from forced displacement and Fragile, Conflict, and Violence (FCV) documents. It uses a lightweight model trained on general research literature to generate candidate mentions, which are then refined by a large language model that validates or rejects them and corrects boundaries. The refined annotations are augmented with synthetic and contrastive examples to fine‑tune the model, achieving 74.1% precision and 70.5% recall on a benchmark of 1,706 passages, with higher precision (89.5%) on passages that contain dataset references.
By Rafael Macalaba, Aivin V. Solatorio, Patrick Michael Brock, Olivier Dupriez
arXiv:2609. 12366v1 Announce Type: new Abstract: We present ORQA, a method for testing occupation-level knowledge in large language models.
By Shreyas Krishnan, Serina Chang, Abhishek Nagaraj
The paper introduces EAR, an Entity‑Aware Partitioning approach that improves retrieval‑augmented generation for multiple‑choice question answering by extracting normalized surface anchors from questions, answers, and the corpus. EAR retrieves local windows around matching anchors and can attach a larger parent passage via an extractive summary, reducing retrieved words by 37.5‑40.2% compared to fixed‑size chunks. Experiments on a cleaned MMLU‑style subset with Mistral, Gemma, and DeepSeek show modest accuracy changes, none statistically significant, highlighting EAR’s methodological contribution of compact, inspectable retrieval units.
By Cenab Batu Bora, Oylum Alatl{\i}, Sebnem Bora, Oguz Dikenelli