Enhancement
-
ImIR: Image-Instruction Tuning for All-in-One Image Restoration
2026-09-21 18:15:08S\"uleyman Aslan, G\"orkay Aydemir, M{\i}sra Yavuz, Yunus Bilge Kurt, Nasrin Rahimi, Ahmet Rasim Emirda\u{g}{\i}, Burak Can Biner, M. Ak{\i}n Y{\i}lmazAbstract
Degradations vary widely across images, so a practical restoration system has to handle many degradation types with one model. A recent and effective recipe adapts a large pretrained image-editing model to restoration using a small low-rank adapter with a text prompt. We replace that prompt with an instruction derived from the degraded image itself. The image reaches the editor through two paths: its structure comes from the model's VAE, and its semantic instruction comes from a lightweight token mapper that shifts the degraded image's vision-language embedding toward the embedding a clean image would produce. Because the instruction is a continuous vector, scaling it yields a family of valid restorations for tasks whose target is not unique, such as low-light enhancement. We adapt one Qwen-Image-Edit model to six tasks with a single adapter trained in about three hours on one GPU. The image instruction outperforms text conditioning under a matched comparison, and it supports task agnostic restoration without a degradation label, which the text variant does not.
Abstract (translated)
URL
https://arxiv.org/abs/2609.25267
PDF
-
INTCORT: Training-Free Spatial Reasoning Enhancement for Vision-Language Models via Input Transformations and Confidence Routing
2026-09-21 16:04:34Haoran Sun, Jingqi Xu, Yanhui Li, Enci Liu, Kaidi Xu, Yanwei LiuAbstract
Vision-Language Models (VLMs) have demonstrated remarkable capabilities in multimodal tasks, yet they still exhibit poor ability in spatial reasoning. Existing training-dependent and training-free enhancement methods suffer from high computational costs with catastrophic forgetting and internal mechanism interference that compromises general capabilities, respectively. In this work, we first verify two key hypotheses: appropriate geometric image transformation and query-reversal transformation can recover incorrect spatial predictions, and correct predictions exhibit higher relation-token confidence than incorrect ones. Based on these findings, we propose INTCORT, a training-free spatial reasoning enhancement framework that constructs multiple inference views through input transformations and aggregates their predictions via relation-token confidence routing, without modifying the VLM's internal mechanisms. Experimental results on several commonly-used benchmarks demonstrate that INTCORT substantially improves spatial reasoning accuracy across diverse VLMs, achieving an average improvement of 10.01% over all models and benchmarks. Compared with prior works, INTCORT achieves superior performance with improvements of up to 25.01%.
Abstract (translated)
URL
https://arxiv.org/abs/2609.24813
PDF
-
Corrective Forcing: Unified Post-Training for Diffusions and Flows in Generative Speech Enhancement
2026-09-21 14:23:15Qing Yao, Lijian Gao, Qirong MaoAbstract
Diffusion and flow models, as promising generative paradigms for speech enhancement, face a training--inference mismatch: training uses analytical path states, whereas inference recursively evaluates models on self-generated rollout states along discretized sampling trajectories. This mismatch causes prediction and discretization errors to accumulate. To address it, we introduce Corrective Forcing (CoF), a post-training paradigm that forces diffusion and flow models to learn from self-generated rollouts and correct their predictions. CoF corrects clean-speech predictions on rollout states toward the ground truth under dynamic sampling schedules, exposing the model to varying inference conditions. It further regularizes local evolution using locally corrected counterfactual transitions as references for factual transitions. By expressing model outputs through a shared clean-speech prediction parameterization, CoF applies the same post-training objective across diffusion and flow formulations. Experiments with SB-VE and OT-CFM demonstrate improvements in perceptual quality and reconstruction fidelity, together with robust performance across different numbers of sampling steps.
Abstract (translated)
URL
https://arxiv.org/abs/2609.24651
PDF
-
URA-NER: A Unified Retrieval-Augmented Framework with Retrieval Alignment and Uncertainty Reduction for Low-Resource NER
2026-09-21 10:04:43Jingyu Wang, Shijie Wu, Fusheng JinAbstract
In-context learning (ICL) based on large language models (LLMs) has shown promising potential in alleviating performance bottlenecks caused by the limited availability of annotated data in Named Entity Recognition (NER). However, existing methods still face issues of retrieval misalignment and generation uncertainty, making their performance heavily dependent on the LLM's capabilities. As the parameter scale of LLMs decreases, their performance in few-shot settings deteriorates significantly. In this paper, we propose a novel unified retrieval-augmented framework, URA-NER, including three key components: Progressive Granularity Retrieval (PGR), Model-aware Representation Enhancement (MaRE), and Reason-aware Knowledge Verification. PGR is a two-stage retrieval mechanism that achieves stage alignment. It first retrieves demonstrations for span detection based on the query's global semantics, and then for type classification based on the specific entity context, providing fine-grained local information. Moreover, MaRE employs entity pre-recognition to guide the construction of representations, ensuring the query and demonstrations are aligned within the LLM's semantic space and attention pattern. In addition, to mitigate generation uncertainty, we propose RaKV, a closed-loop "generation-retrieval-verification" process. It explicates the LLM's reasoning paths, leverages them for the retrieval of external knowledge, and reorganizes the knowledge into verification evidence aligned with the original reasoning paths. We conduct extensive experiments on multiple low-resource NER datasets. Results demonstrate that URA-NER significantly enhances the performance of LLMs under low-resource settings, with particularly pronounced gains for smaller LLMs, achieving new state-of-the-art results on several benchmarks.
Abstract (translated)
URL
https://arxiv.org/abs/2609.24372
PDF
-
Mitigating Entity Type Confusion in Cross-Domain NER via Multidimensional Quantification and Reasoning Enhancement
2026-09-21 09:46:59Jingyu Wang, Shijie Wu, Fusheng JinAbstract
Cross-domain Named Entity Recognition (CD-NER) aims to transfer the rich knowledge in the source domain to the target domain. Recent studies adopting decomposition or generation paradigms have achieved significant performance improvements, demonstrating high accuracy in entity span detection. However, during entity type classification, models severely suffer from entity type confusion, the erroneous tendency that models classify entities of one type in the text as another similar but incorrect type. To address this issue, we first propose a Multidimensional Confusion Quantification Model (MCQM) that quantifies a model's confusion extent between entity types from three dimensions: source-target hierarchy analysis, semantic similarity analysis, and explicit data evaluation. Moreover, we propose the Progressive Bidirectional Reasoning Chain (PBRC). PBRC leverages the source-target hierarchy and confusion analysis from the MCQM to prompt the LLM to generate two-stage reasoning information. The two-stage reasoning information is utilized to augment the knowledge of the model, significantly mitigating entity type confusion and improving the model's generalization performance. Experimental results demonstrate that our method achieves new state-of-the-art results on all domains of the CrossNER dataset.
Abstract (translated)
URL
https://arxiv.org/abs/2609.24357
PDF
-
ORION-CMR: On-scanner Reporting with Integrated Foundation Model for End-to-End Cardiac MRI Analysis and Interpretation
2026-09-20 23:46:46Omer Burak Demirel, Kelly K. Horst, Alessio Perazzolo, Elisa Bruno, Kenan Kaya, Rongzhen Ouyang, Enas Ahmed, Jouke Smink, Spencer L. Waddle, Zainudeen Kallumpurath, Tzu Cheng Chao, Dinghui Wang, Steve G. Langer, Timothy L. Kline, Panagiotis Korfiatis, Jacinta Browne, Ivana Isgum, Tim LeinerAbstract
Cardiovascular magnetic resonance (CMR) provides comprehensive cardiac assessment but remains underutilized because of the complexity of acquisition, post-processing, and interpretation. Existing artificial intelligence (AI) methods address isolated tasks, limiting clinical integration. We present ORION-CMR (On-scanner Reporting with Integrated fOunda-tioN Model), the first clinically evaluated scanner-native end-to-end CMR foundation model. Pretrained on 12,896,733 CMR images from 9,258 studies, ORION-CMR performs sequence classification, ventricular function assessment, late gadolinium enhancement (LGE) detection, binary and multiclass disease classification, and local large language model-based report generation in approximately 90 seconds. The framework. was evaluated on public benchmarks and clinically validated in a multi-vendor cohort of 68 subjects with normal examinations, congenital heart disease, dilated cardiomyopathy, and myocardial infarction. ORION-CMR outperformed supervised baselines and the previously published CMR foundation model (CMR-FM), achieving state-of-the-art performance for LGE classification and scar segmentation. Clinical evaluation achieved an AUC of 0.96 for normal-versus abnormal classification and 0.88 for multiclass disease classification, while generated reports demonstrated 81.4% agreement with expert interpretation. These results demonstrate the feasibility of real-time scanner-native AI-assisted CMR analysis and automated report generation.
Abstract (translated)
URL
https://arxiv.org/abs/2609.23950
PDF
-
TRACE: Deployable Tree-Relational Structure Enhancement for Oncology LLMs
2026-09-20 07:37:08Jizheng Lai, Yingyun Li, Ying Qin, Haiyang QianAbstract
Large language models are increasingly used in oncology applications, but their predictions are often weakly grounded in explicit medical structure. We present TRACE, a deployable tree-relational enhancement framework for oncology LLMs. TRACE separates expensive offline structure learning from lightweight online inference: oncology concepts and relations are organized into an updatable tree-relational structure, refined using LM-loss-derived evidence, and retrieved at inference time as compact prompt evidence. This design supports task-adaptive evidence selection without requiring supervised labels in the zero-shot setting. Across ten oncology classification tasks and one MedQuAD CancerGov QA benchmark, TRACE improves both label-free evaluation and supervised fine-tuning. Additional analyses show that TRACE improves over vanilla RAG and generic GraphRAG, remains useful under leakage-controlled METABRIC inputs, and produces interpretable evidence paths aligned with clinical reasoning. These results suggest that explicit, updatable medical structure is a practical path toward more accurate and auditable oncology LLM deployment.
Abstract (translated)
URL
https://arxiv.org/abs/2609.35810
PDF
-
BLINC: Blind Calibration For Training-Free Speech Enhancement Adaptation
2026-09-18 15:26:30Tobias Raichle, Ekaterina Gavrilko, Bin YangAbstract
Speech enhancement (SE) models degrade under domain shifts and have to adapt to unseen target domains during deployment. Most existing test-time adaptation (TTA) methods for SE do so by adapting a subset of the model weights using a self-supervised loss, which requires backpropagation at test-time and permanently alters the model. We instead recalibrate the prediction itself and propose BLINC, a training-free TTA method that remaps the predicted time-frequency mask onto a bimodal target distribution by histogram matching. At test-time, the target distribution is parameterized from blind features of the noisy recording, so neither a reference distribution from a classical algorithm nor online metric optimization is involved. BLINC improves the overall quality of both evaluated SE models on almost every target condition and matches or exceeds the loss-based TTA baselines at minimal overhead.
Abstract (translated)
URL
https://arxiv.org/abs/2609.21898
PDF
-
Multicentre Bi-atrial Segmentation from LGE-MRI for Atrial Fibrillation with a 2D and 3D Framework
2026-09-18 10:58:43Malitha Gunawardhana, Gregory B. Sands, Mark L. Trew, Jichao ZhaoAbstract
Accurate delineation of bi-atrial structures from late gadolinium enhancement MRI (LGE-MRI) is an important prerequisite for structural analysis and future fibrosis-quantification workflows in atrial fibrillation (AF). However, automated segmentation is challenging due to thin-walled anatomy, domain shifts across imaging centres, and limited benchmarking of existing methods. This study presents a two-stage segmentation framework and benchmarking platform for evaluating how ROI localisation, encoder design, 2D/3D dimensionality, and ensemble fusion affect bi-atrial wall and cavity segmentation across multicentre LGE-MRI datasets. The framework integrates 3D localisation and fine segmentation using 2D and 3D U-Net variants with ResNeXt encoders and compares them with convolutional, transformer-based, and state-space architectures. { Evaluation across three independent cohorts assessed accuracy and cross-domain transfer without target-domain fine-tuning. Cavity segmentation transferred more consistently across centres than atrial wall segmentation, while wall performance remained sensitive to domain shift, particularly in the Kobe cohort.} By quantifying how 2D, 3D, and ensemble architectures behave across centres and between walls and cavities, this work provides a reproducible benchmark for future methodological development and clinical validation.
Abstract (translated)
URL
https://arxiv.org/abs/2609.22398
PDF
-
Lesion-Gated Hybrid Synthesis for Virtual Contrast-Enhanced Breast MRI: A MAMA-SYNTH Challenge Solution
2026-09-18 09:53:32Shohei YoshimotoAbstract
Purpose: Contrast-enhanced breast MRI depends on intravenous gadolinium-based contrast agents, motivating methods that synthesise post-contrast appearance from pre-contrast images alone. The MAMA-SYNTH Challenge (MICCAI 2026 Deep-Breath Workshop) evaluates such synthesis across four metric groups: image fidelity, tumour region of interest, downstream classification, and downstream segmentation. The four group ranks are averaged, so optimising a single objective is insufficient. Materials and Methods: We developed a lesion-gated hybrid synthesis pipeline. A lesion probability map, estimated from the pre-contrast slice alone by an ensemble of pre-contrast-only segmentation networks, spatially coordinates a tumour-focused regression pathway and a background-focused Pix2PixHD synthesis pathway, followed by a region-dependent calibration of the predicted enhancement. Training used the public MAMA-MIA collection with a patient-level split. Inference consumes a single pre-contrast 2D slice, with no mask, no post-contrast image, and no auxiliary metadata beyond image geometry. Results: On internal validation (n = 120 patients) the method reached MSE 0.523, LPIPS 0.185, and tumour-region SSIM 0.495. It was submitted to the hidden external 300-case test cohort as a self-contained inference container. The method ranked sixth in the official MAMA-SYNTH Challenge leaderboard. Conclusion: A test-compatible lesion probability map derived from the pre-contrast image can coordinate complementary tumour-focused regression and background-focused perceptual synthesis, enabling balanced virtual contrast enhancement under a multi-metric challenge setting.
Abstract (translated)
URL
https://arxiv.org/abs/2609.22397
PDF
-
When Online Adaptation Hurts: Parameter-Frozen Test-Time Ensembling for Continual Medical Image Segmentation
2026-09-18 07:21:32Ruijie HuangAbstract
Medical image segmenters often get worse when sites, scanner vendors, or protocols change. Continual test-time adaptation (CTTA) addresses this problem without target labels, but it can be impossible to update a model on a non-stationary stream and can lead to a lot of errors. We examine a more reasonable and meaningful alternative: parameter-frozen inference enhancement(PIE). We use a source-trained segmenter that learns about anatomy-preserving scale and flip views, maps their predictions back to the native location, and averages the probabilities. We do not modify the weights of the model or the normalization statistics. On a cardiac MRI stream from M\&Ms, which is trained on vendor A and evaluated sequentially on vendors B, C, and D, PIE has 0.7786 mean Dice, compared to 0.7680 for source-only inference and 0.7388--0.7416 for five other online-adaptation baselines. The controlled ablations show that performance saturates at 28 views, and confidence weighting, class-prior correction, connected-component filtering, morphological refinement, and inter-slice smoothing have no effect or cause negative transfer. Qualitative results on cardiac MRI and fundus images are also consistent with the frozen ensemble keeping thinner and nested anatomical structures. These results provide a strong, stable baseline for medical CTTA and expose an important failure mode: adaptation and handcrafted refinement can be less reliable than carefully designed inference.
Abstract (translated)
URL
https://arxiv.org/abs/2609.21412
PDF
-
Adaptive Color Grading
2026-09-18 00:32:26Trevor D. Canham, Abhijith Punnappurath, Michael S. BrownAbstract
Independent control of tonescale regions (e.g., shadows, highlights) is essential for painters, photographers and cinematographers to bring 2D images to life. In image manipulation software this is most directly addressed by color grading modules, which use intensity thresholds to segment distinct illumination regions for local manipulation. In this work we develop an open source color grading tool and use it to annotate a large dataset of video frames with tonescale region thresholds. Using these thresholds we conduct modeling experiments with strategies based on both practitioners' conventional wisdom and machine learning. Results show that K-nearest neighbors is an effective prediction strategy, outperforming state-of-the-art end-to-end methods for image enhancement. This outcome demonstrates the benefit of focusing on a compact set of core parameters when modeling creative stylization processes. Our adaptive color grading interface and data are available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2609.21169
PDF
-
Forecasting Intrathecal Tracer Enhancement from Pre-Contrast Brain MRI: Direct Regression versus Flow Matching
2026-09-17 20:49:59Qinghui Liu, Jon Andr\'e Ottesen, Thu Nguyen, Geir Ringstad, Ingrid Mossige, Siri Fl{\o}gstad Svensson, Atle Bj{\o}rnerud, Per Kristian Eide, Kyrre Eeg EmblemAbstract
Intrathecal contrast-enhanced MRI tracks how a cerebrospinal-fluid tracer spreads through the brain, but requires repeated scans over 24--48\,h. Forecasting enhancement from a pre-contrast scan could one day help select patients for intrathecal drug treatment and plan their dose. We ask how much of the enhancement at a given time can be predicted from the pre-contrast scan and the elapsed time alone. Training on 104 patients, we compared direct image-to-image regression (I2I) with conditional flow matching (CFM) using the same 3D U-Net and protocol, and tested on 23 held-out patients and 51 external patients with normal pressure hydrocephalus. All models were measured against simply copying the pre-contrast scan. I2I removed about 60\% of this copying error internally and 25\% externally, outperformed CFM on both test sets (mean absolute error 0.036 vs.\ 0.058 and 0.056 vs.\ 0.063), and predicted each volume in a single forward pass, whereas CFM ran its network ten times. CFM uncertainty located errors but was poorly calibrated. Much of tracer enhancement is thus predictable from anatomy and timing, and direct regression is the more accurate and cheaper choice at this data scale.
Abstract (translated)
URL
https://arxiv.org/abs/2609.22377
PDF
-
A Dual-Stream Regulated Reconstruction and Segmentation Network with Hierarchical Artifact-Prior Modeling for Ultra-Low-Field Pediatric Neuroimaging
2026-09-17 15:26:57Bahram Jafrasteh, Leo Milecki, Qingyu ZhaoAbstract
Automated quality assessment, enhancement, and segmentation of multiple structures in $0.064\,\mathrm{T}$ ultra-low-field pediatric MRI are limited by a low signal-to-noise ratio, weak anatomical boundaries, and frequent artifacts. We present a unified framework for the LISA 2026 Challenge that performs all three tasks together within one inference pipeline. A network with two coupled streams, built on a 3D U-Net, first reconstructs an enhanced uLF volume and then combines the original and enhanced images for subcortical segmentation. To improve boundary stability, we add an auxiliary class covering brain tissue outside the target structures, derived from whole brain masks. A head conditioned on an artifact graph predicts the seven artifact ratings from reconstruction residuals and frozen segmentation features. We address the scarcity of dense annotations using diffeomorphic registration from atlas to target for label propagation and to regularize anatomical reconstruction. We report validation results across all three tasks.
Abstract (translated)
URL
https://arxiv.org/abs/2609.20562
PDF
-
Model-Agnostic and Language-Agnostic Voice Pipeline Improvement for the Agriculture Domain
2026-09-17 14:50:12Aakash Singh, Lakshmi Pedapudi, Chandrashekar M S, Sanyam Singh, Naga Ganesh, Vineet SinghAbstract
FarmerChat is Digital Green's AI-powered agricultural advisory assistant for smallholder farmers, who access it in their own language through text, voice, or photographs. Voice is a critical channel for this population, yet field-recorded speech is challenging for general-purpose automatic speech recognition (ASR) because recordings frequently contain machinery noise, background media, competing speakers, and domain-specific agricultural vocabulary. These conditions disproportionately affect crop, pest, chemical, and quantity terms that carry the meaning of a farmer's query. We present a modular, model-agnostic pipeline for improving ASR quality in FarmerChat without fine-tuning or replacing the underlying ASR model. The pipeline combines gated audio enhancement, speaker diarization and target-speaker selection, ASR, domain-aware correction using a weighted agricultural lexicon, and a quality gate for detecting unreliable transcripts. Only the diarization stage is fine-tuned; all other stages use off-the-shelf models behind common interfaces. We evaluate the pipeline on human-annotated FarmerChat recordings in Hindi, Telugu, and Odia using word error rate (WER) and a domain-weighted error rate that gives greater importance to agricultural terminology. The largest improvements occur on multi-speaker recordings, where target-speaker selection prevents competing speech from entering the transcript. Across the full corpus, the pipeline reduces WER by 16-23% relative on three cloud ASR models and by 5% on an on-device model. On multi-speaker recordings, the reductions are 32-42% for the cloud models and 16% for the on-device model. All reported reductions are statistically significant. These results show that targeted preprocessing, speaker selection, and domain-aware post-processing can substantially improve agricultural speech transcription while preserving the underlying ASR model.
Abstract (translated)
URL
https://arxiv.org/abs/2609.20504
PDF
-
LapaTrack-3D: 6 DoF pre-operative shape tracking for laparoscopic surgery
2026-09-17 09:27:29Jingwei Song, Javid Hussain Jakir, Ray Zhang, Wenwei Zhang, Hao Zhou, Xiaomeng Xian, Maani GhaffariAbstract
This work proposes a real-time 6 Degree-of-Freedom (6 DoF) tracking algorithm for monocular laparoscopic surgery. It provides alignment between intra-operative video and pre-operative data (e.g., CT). The 6 DoF tracking offers a solution for accurately locating the internal anatomy of the target organ despite the lack of tactile feedback and transparency. The ORB-SLAM2 framework is adopted and modified for prior-based 3D tracking with four major modifications. First, the primitive 3D shape is used for fast initialization of the ORB-SLAM2 monocular mode. Second, a pseudo-segmentation strategy is employed to separate the target organ from the background for tracking. Third, the 3D shape is incorporated as a geometric prior in its pose graph optimization. Fourth, the Multi-Scale Retinex with Chromaticity Preservation (MSRCP) algorithm is leveraged and modified for image enhancement in challenging illumination scenarios. In-vivo and ex-vivo experiments validate that LapaTrack-3D provides robust 3D tracking and effectively handles typical challenges such as poor illumination, fast motion, out-of-field-of-view scenarios, partial visibility, and ``organ-background'' relative motion. LapaTrack-3D achieves a processing rate of 13 Hz for 1280*720 pixel video.
Abstract (translated)
URL
https://arxiv.org/abs/2609.19954
PDF
-
PersonalAI 2.0: Enhancing knowledge graph traversal/retrieval with planning mechanism for Personalized LLM Agents
2026-09-17 05:06:02Mikhail Menschikov, Matvey Iskornev, Alexander Kharitonov, Alina Bogdanova, Mikhail Belkin, Ekaterina Lisitsyna, Artyom Sosedka, Victoria Dochkina, Ruslan Kostoev, Ilia Perepechkin, Evgeny BurnaevAbstract
We introduce PersonalAI 2.0 (PAI-2), a novel framework designed to enhance LLM-based systems through integration of external knowledge graphs (KGs). The proposed approach addresses key limitations of existing Graph Retrieval-Augmented Generation (GraphRAG) methods by incorporating a dynamic, multistage query-processing pipeline. The central point of the PAI-2 design is its ability to perform adaptive, iterative information search, guided by extracted entities, matched graph vertices, and generated clue-queries. An evaluation conducted on five benchmarks (Natural Questions, TriviaQA, HotpotQA, 2WikiMultihopQA, and MuSiQue) demonstrates an improvement in the factual correctness of generated answers compared to analogue methods (LightRAG, RAPTOR, HippoRAG 2, and PAI-1). PAI-2 achieves a 9% average gain by LLM-as-a-Judge on the 2WikiMultihopQA and MuSiQue benchmarks, and attains accuracy comparable to HippoRAG 2 on the TriviaQA and HotpotQA benchmarks, reflecting its effectiveness in reducing hallucination rates and increasing precision. We show that enabled search plan enhancement mechanism gain 18% boost compared to disabled one by LLM-as-a-Judge across five benchmarks. In addition, an ablation study reveals that PAI-2 achieves SOTA result on the MINE-1 benchmark, obtaining an 89% information-retention score with LLMs in the 7--15B tiers. Collectively, these findings underscore the potential of PAI-2 to serve as a reusable component for personalized AI applications, which require scalable, context-aware knowledge-representation and reasoning capabilities. The source code of PAI-2 is available at the following link: this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2605.13481
PDF
-
Red-Teaming Auto Mode: Improving Blocking Classifiers Against Malign Coding Agents
2026-09-17 02:14:53Alex Remedios, Simon Storf, Fabien Roger, John HughesAbstract
To keep coding agents from going off the rails, production systems now review each proposed action with a blocking monitor that can reject it before it runs (Auto Mode in Claude Code, Guardian in OpenAI's Codex). Prior evaluations of such monitors largely measure robustness to accidental harm or prompt injections from untrusted sources looking to hijack the agent. Less understood is how they hold up when the agent they monitor is persistently misaligned. To understand this risk, we task an adversarial agent with evading production blocking monitors and causing catastrophic harm, e.g. by exfiltrating its own weights. We find that when instructed with high-level attack strategies, adversarial agents can succeed through several distinct mechanisms, such as agent-generated prompt injection against the monitor, multi-agent attacks, and malicious compaction. In particular we find that in 79% of trials, the agent can use an injection attack against Auto Mode and Guardian to run arbitrary bash commands. We also find that it is possible to greatly improve Auto Mode through design changes like enhancements to tool coverage, transcript formatting and an agentic monitor stage. Even with these improvements, preventing multi-context attacks at an acceptable cost remains an open problem. By detailing our red-teaming methodology and highlighting new attack vectors, we aim to help defenders evaluate their mitigations against the possibility of persistent malign coding agents. Code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2609.19587
PDF
-
QMSR: Query-Conditioned Mask-wise Expert Routing for Robust Open-Vocabulary Underwater Object Retrieval
2026-09-16 14:43:26Fuming Zhang, Dongyue Huang, Junjie Wen, Lihua XieAbstract
Open-vocabulary object retrieval remains challenging in complex underwater environments. Although underwater image enhancement (UIE) can improve visual quality, fixed UIE strategies may even underperform the Raw representation in retrieval, indicating that enhancement should not be applied as a uniform preprocessing step. To address this problem, we propose \textbf{QMSR}, a query-conditioned mask-wise expert routing framework for underwater open-vocabulary retrieval. Specifically, QMSR selects one pretrained UIE expert for each query--candidate pair and predicts a continuous Raw--Expert fusion strength, enabling adaptive enhancement while preserving useful Raw semantics. During training, a privileged ranking oracle provides expert-selection and fusion-strength supervision, while an annealed soft-routing relaxation facilitates optimization of the hard Top-1 routing policy. Experiments show that QMSR improves NDCG@10 by 17.6\% over an image--query shared router, while consistently outperforming fixed UIE strategies and remaining effective on held-out query categories. These results demonstrate the effectiveness of query-conditioned and candidate-specific enhancement routing for underwater open-vocabulary retrieval.
Abstract (translated)
URL
https://arxiv.org/abs/2609.18752
PDF
-
Clueing up LLMs with Tool-Augmented Deductive Reasoning
2026-09-16 14:32:11Rebecca Ansell, Autumn Toney-WailsAbstract
Despite recent advances in large language models (LLMs), performing logically consistent deductive reasoning over extended interactions remains challenging. Tasks that require integrating evidence across multiple reasoning steps, maintaining consistency with prior inferences, and updating beliefs under new constraints can surface limitations in current models while providing a useful testbed for evaluating reasoning enhancements. In this paper, we implement a text-based, multi-agent version of the classic board game Clue as an environment to evaluate multi-step, agentic deductive reasoning. In this setting, agents must infer hidden information from a sequence of observations, maintain consistency across turns, and reason over an evolving set of logical constraints. We instantiate six LLM-based agents (GPT-4o-mini and Gemini-2.5-Flash) as players that engage in turn-based gameplay; using three agents per model family, we establish baseline performance across repeated games. We then introduce a tool-augmented approach in which a structured possibility matrix converts implicit game state from generated reasoning logs into an explicit representation of remaining possibilities. The possibility matrix encodes extended-turn memory and deductive constraints, offloading these tasks from the agent. We compare this approach against the baseline to evaluate how tool augmentation supports reasoning quality and task success for autonomous agents in a strategic reasoning environment.
Abstract (translated)
URL
https://arxiv.org/abs/2609.18736
PDF
