EigenLI introduces a spectral approximation framework that compresses late‑interaction representations by identifying document‑specific low‑dimensional subspaces. By selecting dominant eigendirections, it constructs reduced interaction representations that outperform clustering‑based pooling methods on ColBERTv2 and AnswerAI‑ColBERT‑small. The framework also yields EigenLI‑SV, a single‑vector ANN‑compatible representation that consistently surpasses comparable surrogates such as MUVERA across multiple datasets and text models.
By Archish S, Sabyasachi Basu, Ankit Garg, Ravishankar Krishnaswamy, Kirankumar Shiragur
The paper introduces a neuro‑symbolic framework for constructing knowledge graphs (KGs) that are grounded in an ontology. It combines open‑domain extraction, embedding‑based canonicalization of types and predicates, and a post‑extraction LLM‑based correction step to fix ontology violations, thereby reducing token usage and improving KG consistency. The resulting KGs support symbolic querying, as evidenced by the prevalence of SPARQL graph patterns in the extracted data.
By Lorenzo Loconte, Timothy Hospedales, Cristina Cornelio
arXiv:2609.08442v1 Announce Type: cross
Abstract: Aerial Vision-and-Language Navigation requires drones to follow natural-language instructions and navigate through complex urban environments. Accura...
By Shanwei Fan, Bin Zhang, Zhiwei Xu, Yingxuan Teng, Siqi Dai, Lin Cheng, Guoliang Fan
arXiv:2609.08709v1 Announce Type: new
Abstract: Designing foundation models for graphs is challenging due to the irregular structure of graphs and the different sizes and characteristics of embedding...
By Jonathan Frank, David Richerby, Ansgar Scherp
FastE is a training‑free, plug‑and‑play method that compresses token prefixes in large language model (LLM) embedding inference. It uses a shared fixed threshold on batch‑mean readout‑prefix alignment to decide when to compress and ranks prefix states by readout attention scores to keep the most important ones. Experiments on Qwen3‑Embedding models show that FastE can cut decoder‑backbone FLOPs by over 40% while preserving more than 99% of the original ranking quality across multiple benchmarks and tasks.
By Jinsong Shu, Jinyong Wen, Baokun Wang, Zhongle Xie, Lidan Shou, Weiqiang Wang, Gang Chen
arXiv:2605.27156v2 Announce Type: replace-cross
Abstract: Retrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by incorporating external knowledge, particularly for long-tail do...
By Ruikang Zhang, Zhanni Chen, Yiqiao Cai, Qi Su
HoneyRoute is an inference‑serving layer that identifies malicious requests and redirects them to a honeypot model, protecting the production model while collecting adversarial interactions for intelligence. It combines a lightweight streaming router, a dual‑implementation honeypot, and an analysis loop that turns captured interactions into attacker fingerprints for retraining. In evaluations on production traces and a seven‑domain attack corpus, HoneyRoute achieves high detection F1 scores with minimal latency overhead, significantly reduces token consumption during flooding attacks, and improves fidelity‑traceability trade‑offs compared to naive bait injection.
By Han Jin
arXiv:2609.08200v1 Announce Type: new
Abstract: Token-level text anomaly detection, as an emerging trend of text anomaly detection, moves beyond coarse-grained document-level detection by localizing...
By Kehan Yan, Yue Tan, Qingfeng Chen, Shiyuan Li, Yu Zheng, Yixin Liu
arXiv:2609.07262v1 Announce Type: cross
Abstract: Late-interaction visual document retrievers preserve fine-grained page evidence by storing many token embeddings per page, but the resulting storage...
By PS Rishi, Rajeev Ranjan Dwivedi, Vinod K Kurmi
arXiv:2609.06127v1 Announce Type: cross
Abstract: Retrieval-augmented generation is usually formulated as a statistical information-retrieval problem. Graph-based variants add relational structure, b...
By Xingting Wang, Min Wu
The fusion of hyperspectral (HS) and Light Detection and Ranging (LiDAR) data plays a crucial role in enhancing land-cover classification by jointly exploiting spectral, spatial, and structural cues....
Data augmentation is fundamental to training modern deep vision and multimodal models. While individual methods, such as RandAug, CutMix, Mixup, RandErase, and DropPath, offer strong regularization ef...
Biomedical ontology normalization maps free-text expressions to standardized concepts, enabling consistent integration and analysis of biomedical data. This task remains challenging because lexical va...
Real-world detectors must often interpret functional or ambiguous prompts, yet conventional models such as YOLO remain restricted to fixed class lists. Even open-vocabulary models like YOLO-World freq...
Odor quality is commonly represented using high dimensional descriptor profiles, yet their low dimensional organization remains unclear. We investigated whether a two-dimensional hyperbolic embedding...
Q2D-Web is a new large‑scale benchmark for agentic Retrieval‑Augmented Generation (RAG) systems, featuring a 190 million‑document web corpus and 70 k machine‑reformulated search queries in ten languages. It supplies three sets of relevance judgments—agent citations, production rankings, and a combined set enriched with LLM‑based labels—to evaluate first‑stage retrievers. Experiments on 13 retrievers show consistent ranking across judgment sets but significant variation across domains, languages, and query types, and demonstrate that a carefully sampled sub‑corpus can approximate full‑corpus evaluation with minimal loss in Recall@1000.
The paper explores person identification using millimeter‑wave point clouds beyond traditional gait analysis, focusing on seven activities of daily living (ADLs). It introduces the mm‑ADL dataset of 11 subjects and proposes an activity‑conditioned framework that routes each clip to an activity‑specific identity expert via a supervised mixture of experts. Experiments show that hard routing improves closed‑set ID accuracy from 62.1% to 68.0% and significantly boosts re‑identification metrics, demonstrating the benefit of activity context under controlled indoor conditions.
The paper investigates zero‑shot out‑of‑distribution (OOD) detection in medical imaging using vision‑language models (VLMs). It shows that intermediate layers, rather than only the final layer, provide valuable OOD signals and that the best layer depends on the imaging modality. To overcome instability in entropy‑based layer selection, the authors introduce a multi‑resolution entropy estimation that aggregates histogram statistics across scales, achieving consistent improvements over state‑of‑the‑art methods on the MIDOG and OASIS benchmarks.
ReMoMask-2 is a retrieval‑augmented text‑to‑motion generation framework that improves on complex motion descriptions by addressing coarse retrieval and representation gaps. It introduces a structure‑aware RAG pipeline with Hierarchical Bidirectional Momentum contrastive learning, Semantic Spatial‑Temporal Attention, and Topology Structured Masking, and rebuilds the retrieval database in the generator’s latent space using a lightweight projector. Experiments on HumanML3D, KIT‑ML, and SnapMoGen show state‑of‑the‑art retrieval accuracy and the lowest FID scores, with a single mask‑transformer stage delivering faster inference than the previous two‑stage design.
HypLTSF introduces a hyperbolic geometric framework for long‑term time series forecasting by embedding multi‑scale representations into the Poincaré ball. The approach imposes radial and angular constraints to align the geometry with temporal hierarchies, ensuring that coarser scales abstract finer ones and that related fine‑scale patterns share common ancestors. Experiments on forecasting benchmarks demonstrate that this explicit geometric modeling of the hierarchy yields state‑of‑the‑art performance.