The paper introduces AWM, a framework that treats the terminal working memory of long‑document VQA agents as an answerable evidence artifact. It proposes a memory‑only answerability diagnostic and incorporates this signal into the GRPO reward, giving higher advantage to trajectories whose final memory can answer the question alone. Experiments on MMLongBench‑Doc and LongDocURL show that AWM‑GRPO boosts final‑answer accuracy by up to 11.9 points and reduces the rate of correct answers that cannot be supported by memory alone.
By Dongzhuoran Zhou, Yuqicheng Zhu, Yule Liu, Zhen Yang, Rui Lu, Yuxiao Dong, Jie Tang, Evgeny Kharlamov
Just Pass Twice (JPT) is a method that allows causal large language models to perform token classification for zero‑shot named entity recognition by concatenating the input with itself, giving each token full bidirectional context without architectural changes. The approach combines these representations with definition‑guided entity embeddings to enable flexible zero‑shot generalization. JPT achieves state‑of‑the‑art results, outperforming prior methods by an average of +7.9 F1 on CrossNER and MIT benchmarks and running over 20× faster than comparable generative approaches.
By Ahmed Ewais, Ahmed Hashish, Amr Ali
The paper introduces Efficient Retrieval Adapter (ERA), a query‑side adapter framework that enables dense retrieval systems to adapt to asymmetric query–document scenarios without re‑indexing. ERA first aligns the embedding spaces of a powerful query embedder and a lightweight document embedder using unlabeled documents, then fine‑tunes the aligned query representation with a small set of labeled query‑document pairs. In experiments on 126 MAIR retrieval tasks across six domains, ERA boosts average nDCG@10 by up to 8.2 points in symmetric settings and over 12 points in asymmetric settings while requiring far fewer labels than fully supervised adapter training.
By Seiji Maekawa, Moin Aminnaseri, Pouya Pezeshkpour, Estevam Hruschka
The paper introduces a three‑layer security framework designed to protect retrieval‑augmented generation (RAG) chatbots from both direct and indirect prompt injection attacks. Layer 1 filters user input with rule‑based patterns and a semantic anomaly classifier; Layer 2 enforces a provenance‑based instruction hierarchy during context assembly; Layer 3 audits model output with a policy rule engine and semantic drift detector. Evaluations on GPT‑4o, Llama 3, and Mistral 7B demonstrate a reduction in attack success rate from 71.4 % to 11.3 %, outperforming existing single‑layer defenses while keeping false positives low and latency acceptable.
By Gulshan Saleem, Nisar Ahmed, Muhammad Imran Zaman, Ali Hassan, Umar Mujahid
Corpus2Skill is a retrieval architecture that transforms an enterprise knowledge base into a hierarchical skill directory, enabling an LLM agent to navigate from high-level summaries to specific documents and backtrack when necessary. On an enterprise customer‑support benchmark, it outperforms single‑shot dense, hybrid, hierarchical‑retrieval, and agentic RAG baselines in answer quality and grounding, with a moderate cost tradeoff. An eleven‑dataset study shows that corpus navigation excels on single‑domain corpora with a recoverable topical taxonomy but is less effective on open‑domain factoid pools or homogeneous‑tabular corpora, providing a design guideline for knowledge‑grounded systems.
By Yiqun Sun, Pengfei Wei, Lawrence B. Hsieh
TAU-Agent is a retrieval‑augmented framework designed for traffic anomaly understanding in transportation videos. It uses a central retrieval agent to coordinate a Video Captioning Tool and an Open‑Vocabulary Tracking Tool, gathering captions, temporal intervals, and object trajectories relevant to a query. The collected evidence, along with sampled frames and the query, is fed into a fine‑tuned vision‑language model that reasons and generates an answer. TAU-Agent was evaluated on the AI City Challenge 2026 benchmarks, achieving notable scores across multiple tracks and ranking second, twelfth, and fifth respectively.
By Yuqiang Lin, Yan Shi, Sam Lockyer, Harish Tayyar Madabushi, Adrian Evans, Wenbin Li, Yinhai Wang, Nic Zhang
EXAONE Tabular 1.0 is a compact tabular foundation model family that performs classification and regression via in-context learning without dataset-specific gradient updates. It is pretrained exclusively on a synthetic structural‑causal‑model prior and introduces an architecture‑centered redesign that interleaves feature‑axis and item‑axis attention within each Transformer layer, mediated by summary tokens. Across four public benchmarks, its 20.81 M‑parameter classification model ranks first on TabArena, surpassing tuned ensembles and AutoML pipelines, while its regression model matches the performance of a 1.64 B‑parameter model at roughly one‑eleventh the inference cost, and it achieves top rankings on BCCO, TALENT, and ScoringBench.
By Moonjung Eo, Min-Kook Suh, Hye-Seung Cho, Jiwon Kim, Seoyoon Kim, Sangjun Nam, Soonyoung Lee
The paper introduces a retrieval‑augmented multi‑agent framework that automatically generates instance‑specific evaluation rubrics for medical language models. By retrieving authoritative medical evidence, decomposing it into atomic facts, and combining these with user interaction constraints, the system produces fine‑grained criteria that outperform GPT‑4o on HealthBench and LLMEval‑Med. The generated rubrics also guide response refinement, improving medical LLM output quality by 9.2%.
By Yinzhu Chen, Abdine Maiga, Hossein A. Rahmani, Emine Yilmaz
arXiv:2607.12176v2 Announce Type: replace
Abstract: Ambivalence and hesitancy (A/H) undermine digital behaviour-change interventions, and recognizing them automatically from video is the goal of the...
By Josep Cabacas-Maso, Ismael Benito-Altamirano, Carles Ventura
GreenLeaf Law Embed Tiny is a 0.6 B parameter embedding model designed for legal domain retrieval. It achieves 75.11 % on the Massive Legal Embedding Benchmark and 64.38 % on MTEB(Law, v1), outperforming other models under 1 B parameters. The model is trained via a two‑stage pipeline that distills knowledge from a larger teacher, fine‑tunes with hard negative mining, and uses a curated dataset of 3.4 million query‑passage pairs, including 150,000 human‑curated samples from diverse legal jurisdictions, while supporting efficient inference with multiple quantization levels.
By Surya Saka
The paper investigates evidence generation for biomedical claim verification, evaluating various large language models and retrieval strategies on the CARE-XAI benchmark. It finds that fine‑tuned LLMs excel at producing evidence, while biomedical classifiers still lead in verdict‑only prediction. PubMed retrieval helps on PubMed‑aligned datasets but can mislead on broader public‑health claims, prompting the authors to propose Bio‑GRACE, a diagnostic that normalizes gold references to assess retrieval utility.
By Pritam Deka, Prabhjot Singh
PANDA (Prototype‑Anchored Data Alignment) is a two‑stage framework that enables a primary‑modality model to benefit from auxiliary modalities even when those modalities are only partially paired or absent at inference. In Stage 1, a shared embedding is learned from the paired subset and class prototypes are estimated from the auxiliary data; in Stage 2, the primary encoder is trained on all subjects using cross‑entropy and alignment to the frozen prototypes. PANDA was evaluated on Alzheimer’s MRI and TCGA‑Lung pathology, achieving significant AUC gains and improved survival prediction while requiring no auxiliary inputs during deployment.
By Sheethal Bhat, Mahfuzur Rahman Chowdhury, Paula Andrea Perez-Toro, Stephan Wunderlich, Rose Dawn Bharat, Siming Bayer, Andreas Maier
The paper introduces TRACE, a lightweight framework for detecting corpus poisoning in Retrieval-Augmented Generation systems. TRACE works by tracing answer-related tokens through token influence attribution, first identifying recurrent high-influence keywords across retrieved documents and then verifying their impact on model predictions. Experiments on three QA benchmarks and six large language models show that TRACE achieves strong detection performance while also revealing attacker-specified target answers.
By Yan-Lun Chen, Pin-Yu Chen, Chia-Mu Yu, Ying-Dar Lin, Yu-Sung Wu, Wei-Bin Lee
The paper reports a new phenomenon in CLIP embeddings where human and AI‑generated paintings naturally separate along dominant principal directions without any supervised training. The authors investigate this separation by linking embedding directions back to image features using interpretable representations and gradient‑based inversion, finding that the separation is driven by distributed multiscale image structure rather than simple global or local statistics. They also show that small, imperceptible image perturbations can cause large displacements along these directions, highlighting a mismatch between CLIP’s visual evidence and human perception.
By Andrea Asperti
The paper investigates when frequency decomposition aids Physics-Informed Neural Networks (PINNs) by introducing a dual‑branch, spectrally‑gated architecture (DBSG‑PINN) that separates low‑ and high‑frequency components. Experiments on five one‑dimensional PDE benchmarks show that frequency decomposition significantly reduces error—up to 59.2% on a multimodal wave problem—when the target solution is spectrally complex, but offers little improvement on smoother problems and can even worsen performance on a simple 1D wave benchmark. The adaptive gate’s effectiveness scales with the spectral richness of the solution, suggesting it exploits frequency structure rather than adding noise.
By Shubham Rai
The paper investigates how medical vision‑language models (VLMs) behave when faced with distribution shifts such as changes in acquisition domain, supervision, or evaluation protocol. Using datasets like NIH ChestXray14, CheXpert, PadChest, and OpenI, the authors isolate cross‑dataset visual transfer, evaluate multimodal alignment, and quantify source‑proxy leakage in frozen embeddings. They find that self‑supervised visual initialization improves transfer, adversarial adaptation is only marginally helpful, and that multimodal retrieval performance drops under external stress tests while source‑proxy information remains recoverable, highlighting hidden failure modes in medical VLMs.
By Ayoub Louaye Bouaziz, Lokmane Chebouba, Yassine Himeur
EigenCL is a contrastive learning framework that stages crop stress by embedding Sentinel‑2 NDRE trajectories, producing four physiologically coherent clusters—Healthy, Mild, Moderate, and Severe—without retraining across different states. Trained on 10,000 maize NDRE patches from Iowa in 2020 and validated on Nebraska data in 2023, EigenCL outperformed baselines such as K‑Means, SimCLR, and ProtoCLR, achieving high silhouette, DBI, and CHI scores. The resulting clusters align with maize growth stages, correlate strongly with soil moisture and yield anomalies, and provide interpretable outputs like heatmaps and scouting priorities for decision‑support systems.
By Shafqaat Ahmad
MoganBert-TR is a 149‑million‑parameter Turkish encoder foundation model trained from scratch on a language‑specific corpus using a two‑stage CLM‑to‑MLM curriculum. The model, along with its embedding variant MoganBert‑Embed, achieves state‑of‑the‑art results on Turkish benchmarks such as TrGLUE and TabiBench, outperforming existing Turkish BERT models. Its tokenizer, comprising 50,048 tokens, also surpasses other Turkish tokenizers in compression and fertility metrics.
By Furkan Yilmaz, Habibe Aleyna Tasdemir, Muhammed Faruk Gozay
The paper introduces AdaptiveEmbed, a framework that learns sample‑adaptive multi‑vector representations for multimodal retrieval. It proposes the SAMVR setting, where each sample receives a content‑adaptive embedding set whose size is determined by its retrieval utility. The method uses Multi‑Group Contrastive Learning with set‑to‑set similarity and Utility Policy Optimization to allocate representation capacity, achieving better performance across image, text, video, and audio benchmarks.
By Xinze Liu, Lei Yang, Dayan Wu, Hengjie Zhu, Zihao Zhang, Hanqi Wu, Tianzhu Hu, Peng Fu, Zheng Lin, Weiping Wang
The paper introduces ReT-2, a unified retrieval model that handles multimodal queries containing both images and text and searches across multimodal document collections. It employs a recurrent Transformer architecture with LSTM-inspired gating to integrate information across layers and modalities, capturing fine-grained visual and textual details. Evaluations on M2KR and M-BEIR benchmarks show state‑of‑the‑art performance, faster inference, and lower memory usage, and the model also boosts downstream tasks in retrieval‑augmented generation pipelines.
By Davide Caffagni, Sara Sarto, Marcella Cornia, Lorenzo Baraldi, Rita Cucchiara