❌

Reading view

↗️

K-Bench: A Benchmark for LLM Unlearning in Agentic Deployments

βœ‡cs.AI, q-bio.NC updates on arXiv.org
By: Guangsheng Yu Β· Yanna Jiang Β· Qin Wang Β· Baihe Ma Β· Xu Wang
arXiv:2609.12808v1 Announce Type: new Abstract: Unlearning benchmarks such as TOFU and MUSE certify forgetting by reading the model's final answer, where a model that refuses to answer already counts as having forgotten. We show that this model-level certificate does not transfer once the model is deployed as an agent. We introduce K-Bench, a benchmark that scores LLM unlearning under agentic deployment. K-Bench inspects all six channels a ReAct agent exposes, including its chain-of-thought (CoT), tool calls and tool observations, and elicited summary. A query counts as leaked if the secret appears in any of them. Each experiment places the secret in exactly one of the agent's three sources (the weights, the prompt, or the retrieval store). The K-Score is computed separately for each source and credits forgetting only when the agent remains usable. Clearing the answer channel does not make the secret unrecoverable. On structured retrieval, the secret stays verbatim in the tool-observation channel and the aggregate leak rate is unchanged. When the secret lives in the prompt or the retrieval store, TOFU and MUSE report no leakage, while the deployed agent still leaks it on 22--86\% of queries. When the secret is in the weights, none of the twenty evaluated published methods demonstrably removes it, and only an input-corruption intervention reaches selective forgetting under the evaluated observer. The top-ranked method changes across base models. A refusal-tuning method resists the evaluated extraction without verified knowledge removal.
  •  
  • ↗️
↗️

Label-Guided Knowledge Distillation for 3D-CNNs in Action Recognition

βœ‡cs.AI, q-bio.NC updates on arXiv.org
By: Yanjiang Shi Β· Peng Zhao Β· Nan Qi Β· Guiqin Wang
arXiv:2609.13024v1 Announce Type: cross Abstract: As a key model compression technique, knowledge distillation aims to transfer knowledge from a high-capacity teacher model to a lightweight student model for enhancing the latter's performance. In this work, we reviewed the feature knowledge distillation for 3D-CNNs and observed that most feature distillation methods in video analysis are simple adaptations of those used in image analysis, often neglecting the differences of video features in the temporal dimension. To address this issue, we proposed Label-Guided Knowledge Distillation (LGKD) to guide the distillation of student model features using ground truth labels. Our method entails two components: sample-wise distillation and class-wise distillation, enabling the student model to learn feature representation of the teacher model at two levels. Sample-wise distillation utilizes label information and the teacher's probability distribution to guide the learning of features that significantly impact temporal accuracy while mitigating noise. Meanwhile, class-wise feature distillation employs a prototype network to further capture the relational knowledge among samples within the same category, enhancing the student's ability to learn higher-dimensional semantic information and improving model generalization. To demonstrate the effectiveness and superiority of our method, we conducted comprehensive experiments on two benchmark action recognition datasets, UCF101 and HMDB51, achieving competitive results.
  •  
  • ↗️
↗️

FoodMonitor: Benchmarking MLLMs for Explainable Compliance Analysis

βœ‡cs.AI, q-bio.NC updates on arXiv.org
By: Ruihao Xu Β· Xingming Shui Β· Jingxuan Niu Β· Yiqin Wang Β· Jilin Yu Β· Haoji Zhang Β· Yansong Tang
arXiv:2605.24503v1 Announce Type: cross Abstract: As AI-powered compliance monitoring becomes increasingly important in public governance and industrial safety, the ability to provide verifiable evidence and traceable accountability signals is essential. However, existing video anomaly detection datasets focus on event-level binary classification, lacking the rule-driven, explainable analysis required for real-world compliance scenarios. We introduce FoodMonitor, a benchmark for explainable compliance analysis in commercial kitchen surveillance. FoodMonitor comprises 477 video clips with 3,307 violation annotations across a dual-channel design covering both person-level and environment-level violations. Each annotation specifies which rule was violated, what non-compliant behavior occurred, and who committed it with frame-level bounding boxes. We establish a unified evaluation protocol with a two-stage matching mechanism that separately assesses spatial localization and semantic understanding, along with a composite metric ($C_{\text{score}}$) that balances environment and person detection performance. Systematic evaluation of several state-of-the-art multimodal large language models reveals that the best-performing model achieves only 0.360 $C_{\text{score}}$, with spatial localization and fine-grained rule understanding emerging as the primary bottlenecks. Our analysis identifies two distinct failure modes: localization-dominated errors and semantics-dominated errors, providing diagnostic insights for future model development.
  •  
  • ↗️
↗️

Circulating Tumor Cells in Pancreatic Ductal Adenocarcinoma: The Systemic Execution Hub of Metastasis

βœ‡MRD
By: Liang Hao Β· Shi-Yang Lu Β· Fu-Ying Zhao Β· Jia-Mei Wang Β· Bai-Qiang Li Β· Ting Zhang Β· Zhen-Xian Du Β· Hua-Qin Wang

Pharmacol Res. 2026 May 17:108253. doi: 10.1016/j.phrs.2026.108253. Online ahead of print.

ABSTRACT

Pancreatic ductal adenocarcinoma (PDAC) exemplifies early systemic dissemination, with circulating tumor cells (CTCs) at its core. We advance a unified conceptual framework that positions CTCs as the systemic execution hub of PDAC metastasis, dynamic entity that coordinates the metastatic cascade via four cardinal functions: Seeding, Adapting, Engineering, and Signaling. Integrating eco-evolutionary dynamics, this hub actively drives phenotypic selection, niche remodeling, and immune evasion, while providing real-time biologic intelligence through liquid biopsy. Robust clinical correlation has not yet translated into routine practice because of technical variability, biological complexity, and a lack of interventional evidence. We therefore propose an evidence-driven, phased roadmap: grounded in prospective clinical cohort data, progressing from immediate multi-center technical standardization and pragmatic trials, such as minimal residual disease (MRD)-triggered salvage therapy, to mid-term biomarker-driven adjuvant trials and long-term integration into multimodal liquid biopsy ecosystems, aimed at intercepting this execution hub. By reframing CTCs from correlative indicators to actionable therapeutic targets and dynamic sentinels, this framework charts a path toward transforming the management of this recalcitrant systemic disease.

PMID:42150733 | DOI:10.1016/j.phrs.2026.108253

  •  
  • ↗️
↗️

Refined immune-based molecular subtypes of gastric cancer: Integrating mismatch repair status and tumor microenvironment for enhanced immunotherapy prediction

βœ‡Omics in Gastric
By: Ke Shen Β· Yanqiang Zhang Β· Zhiqin Wang Β· Ruixing Xu Β· Jintao He Β· Mengya Zhou Β· Yiming Lu Β· Danni Zhu Β· Yizhou Wei Β· Weiwei Zhu Β· Qing Yang Β· Zhiyuan Xu Β· Can Hu Β· Siwei Pan Β· Xiangdong Cheng

Chin J Cancer Res. 2026 Apr 30;38(2):234-251. doi: 10.21147/j.issn.1000-9604.2026.02.09.

ABSTRACT

OBJECTIVE: Gastric cancer (GC) is heterogeneous, and current mismatch repair (MMR)-based classifications incompletely predict response to immune checkpoint inhibitors (ICIs).

METHODS: RNA sequencing (RNA-seq) and immune infiltration profiles from 189 resected GC were used to derive four refined immune-MMR subtypes (R1-R4) by integrating MMR status, survival, and tumor microenvironment (TME) features. Multi-omics profiling and pathway analysis defined subtype biology. External transcriptomic cohorts and an ICI-treated cohort were classified with Nearest Template Prediction (NTP). Immune response-associated genes were identified from responder vs. non-responder comparisons within the ICI-sensitive subtype and validated by multiplex immunohistochemistry (mIHC).

RESULTS: R1 showed the best prognosis and highest immunotherapy response with objective response rate (ORR) 54.5%, while R4 had the worst prognosis. R2 represented an immune-unresponsive deficient mismatch repair (dMMR) subset, and R3 captured an immune-active proficient mismatch repair (pMMR) subgroup with moderate therapy sensitivity. Multi-omics integration revealed subtype-specific pathways (e.g., ECM remodeling in R1, metabolic reprogramming in R2). Reclassification of pMMR tumors based on transcriptional similarity to R1 identified a New R3 subset with enhanced immune features and higher ICI response. Eight immune response-associated genes (e.g., CXCL10, CXCL11, ELN, GAD1, IL32, MT1E, OR2I1P, SLC3A1) were identified and validated by mIHC for predictive relevance.

CONCLUSIONS: This immune-based molecular framework refines risk stratification beyond conventional MMR categories, identifies ICI-sensitive subsets among both dMMR and pMMR tumors, and proposes candidate biomarkers for patient selection.

PMID:42147371 | PMC:PMC13171420 | DOI:10.21147/j.issn.1000-9604.2026.02.09

  •  
  • ↗️
↗️

Pyruvate is a natural suppressor of interferon signaling by inducing STAT1 protein pyruvylation

βœ‡Cell
By: Yibo Zuo Β· Qin Wang Β· Wanying Tian Β· Xinhe Wang Β· Zhijin Zheng Β· Wei He Β· Renxia Zhang Β· Qian Zhao Β· Ying Miao Β· Yukang Yuan Β· Tingting Zhang Β· Qun Cui Β· Yuerong Zhang Β· Chunyan Liu Β· Haiyan Zhou Β· Hui Zheng
Yibo et al. identify protein pyruvylation as a post-translational modification that can modulate immune signaling and host antiviral response.
  •  
  • ↗️
↗️

Differentially Private Retrieval-Augmented Generation

βœ‡cs.AI, q-bio.NC updates on arXiv.org
By: Tingting Tang Β· James Flemings Β· Yongqin Wang Β· Murali Annavaram
arXiv:2602.14374v1 Announce Type: cross Abstract: Retrieval-augmented generation (RAG) is a widely used framework for reducing hallucinations in large language models (LLMs) on domain-specific tasks by retrieving relevant documents from a database to support accurate responses. However, when the database contains sensitive corpora, such as medical records or legal documents, RAG poses serious privacy risks by potentially exposing private information through its outputs. Prior work has demonstrated that one can practically craft adversarial prompts that force an LLM to regurgitate the augmented contexts. A promising direction is to integrate differential privacy (DP), a privacy notion that offers strong formal guarantees, into RAG systems. However, naively applying DP mechanisms into existing systems often leads to significant utility degradation. Particularly for RAG systems, DP can reduce the usefulness of the augmented contexts leading to increase risk of hallucination from the LLMs. Motivated by these challenges, we present DP-KSA, a novel privacy-preserving RAG algorithm that integrates DP using the propose-test-release paradigm. DP-KSA follows from a key observation that most question-answering (QA) queries can be sufficiently answered with a few keywords. Hence, DP-KSA first obtains an ensemble of relevant contexts, each of which will be used to generate a response from an LLM. We utilize these responses to obtain the most frequent keywords in a differentially private manner. Lastly, the keywords are augmented into the prompt for the final output. This approach effectively compresses the semantic space while preserving both utility and privacy. We formally show that DP-KSA provides formal DP guarantees on the generated output with respect to the RAG database. We evaluate DP-KSA on two QA benchmarks using three instruction-tuned LLMs, and our empirical results demonstrate that DP-KSA achieves a strong privacy-utility tradeoff.
  •  
  • ↗️
There are no more articles
❌