The paper introduces CARE‑SD, a four‑module architecture designed to enhance the reasoning and explanation capabilities of small language models (SLMs) for analyzing self‑disclosure in cybersecurity interview transcripts. By integrating context‑adaptive processing, hierarchical disclosure reasoning, explanation‑consistency verification, and cross‑batch reconciliation, CARE‑SD improves the accuracy and evidence reliability of SLMs, raising Llama 3B’s Macro‑F1 from 0.80 to 0.88 and cutting unsupported claims by 60%. The study demonstrates that a 3B model can match the performance of a 20B reference model, highlighting the architecture’s effectiveness without needing larger models.
By Aisvarya Adeseye, Adeyemi Adeseye, Mohammad Tahir, Jouni Isoaho
The paper investigates whether visual generation can extend the knowledge of unified multimodal models (UMMs). Using a new Knowledge Transfer Probing protocol, the authors show that training UMMs to generate images can transfer knowledge to text-only question answering, though the transfer is uneven across categories and models. They further demonstrate that aligning textual and interleaved image–text outputs improves knowledge accessibility, and that knowledge accurately captured in images tends to be more retrievable via text, yet gaps persist.
By Wenkai Li, Wenyi Hong, Zhuoyi Yang, Jie Tang
The paper investigates privacy risks of language embeddings used in recommender systems. It demonstrates that a simple cosine‑similarity attack can recover item identities with high accuracy when an adversary has auxiliary text, even when the descriptions are independently sourced. The study also shows that such leakage can lead to user re‑identification from sparse ratings and evaluates obfuscated descriptions as a mitigation, revealing a trade‑off between privacy and utility.
By Linda Bekov\'a, Rodrigo Alves
BridgeAfford tackles the query‑target gap in language‑guided 3D affordance grounding by aligning interaction regions across objects and refining queries with visual evidence. The framework introduces Interaction Region Alignment to promote cross‑object consistency of geometrically similar regions, and Visual Evidence Aggregation to guide queries toward correct regions using a 2D affordance model. Experiments on LASO and PIAD show that BridgeAfford outperforms existing methods, especially on unseen objects, demonstrating robust generalization.
By Yijie Su, Yakun Zhu, Yi Bin, Pengpeng Zeng, Liang Chen, Guoqing Wang, Jingkuan Song, Heng Tao Shen
The paper examines how continual learning (CL) techniques used to update deepfake detectors over time affect fairness across different demographic groups. By training detectors on a chronological series of 11 real-world deepfake generation methods, the authors track four fairness dimensions and find that accuracy loss often worsens unfairness, while fairness gains are rare. The study reveals that inequitable forgetting is linked to representation learning rather than mere underrepresentation, urging CL methods to consider social outcomes when balancing stability and plasticity.
By Kajaree Das, Oliver B\"ose, Subarnaduti Paul, Gabriel Storch, Martin Mundt
GraphCompliance is a structure‑aware framework that models regulatory documents as a Policy Graph and contextual information as a Context Graph of entity‑relation triples. It aligns these graphs via a Compliance Gate to identify relevant regulatory requirements and evaluate them against contextual evidence, thereby preserving structural dependencies and reducing terminology mismatch. Experiments on EU2UK, GDPR‑13, and CONTRACT datasets show higher Matthews correlation coefficients than competing methods, and ablation studies confirm the importance of each component.
By Jiseong Chung, Ronny Ko, Won-Yong Shin, Wonchul Yoo, Tae-Wan Kim
The paper discusses how human feedback used to align language models aggregates conflicting judgments about desirable behavior, resulting in an objective that reflects a collective preference. It surveys recent research that applies social choice theory to this aggregation problem, showing how this perspective can uncover failure modes in the feedback layer and expand the design space for managing disagreement in a principled manner.
By Daniel Halpern, Evi Micha, Ariel D. Procaccia, Benjamin Schiffer, Itai Shapira, Shirley Zhang
The paper introduces FACTGUARD, a multi-task workflow that generates evidence‑grounded answerable and unanswerable questions to improve long‑context reasoning in large language models. It presents FACTGUARD‑BENCH, a benchmark of 25,220 examples with context lengths up to 128K, focusing on Lack of Evidence and Misleading Evidence scenarios. Experiments on eleven LLMs show a reliability gap, with the best model achieving only 71.62% accuracy; training on FACTGUARD‑BENCH raises accuracy to 81.17% and enhances reasoned rejection of unanswerable queries.
By Qian-Wen Zhang, Biao Liu, Fang Li, Jie Wang, Lingfeng Qiao, Yifei Yu, Di Yin, Xing Sun
UniFGVC introduces a training‑free framework for few‑shot fine‑grained visual classification that reframes the task as a multimodal retrieval problem. It uses a Category‑Discriminative Visual Captioner to generate attribute‑aware textual descriptions, then matches image‑description pairs with category templates via pre‑trained encoders. Experiments on 12 datasets show that UniFGVC consistently outperforms existing CLIP‑based and fully‑supervised MLLM methods, demonstrating the effectiveness of the retrieval paradigm.
By Haiyun Guo, Hongyu Guo, Xiangzhao Hao, Jiarui Guo, Jinqiao Wang, Tat-Seng Chua
The paper introduces AXIOM, a perturbation-based framework for creating a balanced benchmark of code evaluation problems. AXIOM refines program quality by applying controlled semantic perturbations and then calibrates scores through selective manual annotation, resulting in the multilingual AXIOMBENCH dataset with 1,962 problems. This approach addresses limitations of prior benchmarks that rely on coarse binary labels or subjective human scoring.
By Ruiqi Wang, Xinchen Wang, Cuiyun Gao, Chun Yong Chong, Xin Xia, Qing Liao
The paper introduces Span-Centric Learning (SCL), a method that decouples coding knowledge acquisition from its application in full clinical documents. SCL first trains on short text spans to learn coding knowledge, then fine‑tunes on a small set of annotated documents, enabling an 8B LLM to outperform state‑of‑the‑art methods with only 248 full documents. The approach also yields explicit, human‑aligned supporting evidence for each predicted ICD code and can learn code knowledge from spans even when the codes are absent from the full documents.
By Xu Zhang, Wenxin Ma, Chenxu Wu, Rongsheng Wang, Zhiyang He, Xiaodong Tao, Kun Zhang, S. Kevin Zhou
The paper introduces LLMON, an LLM-native markup language designed to encode the structure and semantic metadata of textual inputs for large language models. By distinguishing components such as instructions and data within prompts, LLMON aims to reduce model confusion and mitigate security risks like prompt injection. The authors outline design requirements, present the language, and provide preliminary evidence that incorporating LLMON can improve model accuracy, safety, and security during training and inference.
By Michael Hind, Basel Shbita, Bo Wu, Farhan Ahmed, Chad DeLuca, Nathan Fulton, David Cox, Dan Gutfreund
The paper examines privacy risks in large language models (LLMs) through a unified threat model and systematic ablation study of four attacks: Membership Inference, Attribute Inference, Data Extraction, and Backdoor Attacks. It finds that membership inference is highly sensitive to dataset and attack strategy, attribute inference depends on model and training characteristics, data extraction grows with model size and repeated training data, and backdoor attacks consistently succeed. The authors provide practical recommendations for privacy‑aware LLM deployment, such as prioritizing membership leakage checks in retrieval‑augmented systems, reducing repeated personal data in training, monitoring attribute inference, and protecting fine‑tuning pipelines from backdoors.
By Karima Makhlouf, Lamiaa Basyoni, Syed Khaderi, Gabriel Marquez, Peter Sotomango, Mahmoud Awawdah, Sami Zhioua
The study presents a clinically interpretable machine‑learning framework that estimates biological age (BA) using a CatBoost regression model trained on 44 diverse predictors from NHANES data. SHAP values are leveraged not only for visual interpretation but also through an automated quantitative procedure that identifies feature‑value ranges linked to the lowest and highest BA contributions, including zero‑crossing thresholds. The resulting BA estimates correlate strongly with chronological age and highlight key predictors such as blood urea nitrogen, serum glucose, and vigorous activity, offering transparent, reproducible insights for precision health research.
By Yanni Sun, Zheng Qu, Kevin Ren, Zhanglin Lin Cui
arXiv:2610. 21001v1 Announce Type: new Abstract: Column generation folklore warns that initial columns hurt: a warm start may bias the dual and slow the root of branch-price-and-cut.
By Haoran Liu, Yu Yang
The paper investigates how language models’ internal judgments can be separated from the way those judgments are expressed. By comparing hidden-state vectors under different instructions, the authors identify a low-dimensional expression subspace that can be removed to isolate the model’s true judgment. This isolation improves predictive performance across tasks such as correctness, entailment, and safety, even when the model’s final answers degrade.
By Chaemin Jang, Donggyu Lee, Jeesoo Kim, Dongman Lee, Jihee Kim
The paper introduces a federated minimax framework where the minimization variable is shared across clients while each client retains a personalized maximization variable. It proposes a projected ascent method for the personalized variables and shows that in nonconvex-concave settings it achieves a sublinear convergence rate, but in nonconvex-PL settings it suffers from a noise floor. To overcome this, the authors develop VR-SGDA, a variance‑reduced algorithm that eliminates the noise floor through a two‑timescale update scheme, and demonstrate its effectiveness on subpopulation‑robustness and NLP adversarial training tasks.
By M Yashwanth, Sharannya Ghosh, Saikiran Bulusu, Anirban Chakraborty
The paper introduces a graph deep learning framework that learns to forecast spatiotemporal data across multiple hierarchical structures, using variations among hierarchies as multi‑task regularization. By jointly training on different aggregation levels, the model reduces over‑specialization and improves forecasting accuracy. A novel top‑down operator ensures coherent forecasts with linear time and space complexity, making the approach computationally efficient.
By Valentina Moretti, Cesare Alippi, Andrea Cini
ShiftQAT is a quantization‑aware training workflow designed for the MAX7800x family of fixed‑function CNN accelerators. It enforces the accelerator’s numerical contract—power‑of‑two, per‑tensor symmetric scales, 8‑bit activations and biases, integer‑only arithmetic, and a single statically configured shift per layer—by training weights and biases on a joint power‑of‑two grid and resolving dependencies so that activation scales propagate statically. Compared to other methods, ShiftQAT uniquely satisfies all contract constraints while matching the accuracy of less constrained QAT approaches, achieving ImageNet performance within 2.9 percentage points of FP32 at W4A8 and enabling fast, energy‑efficient inference on MAX78002 silicon.
By Oguzhan Buyuksolak, Osman Erman Okman, Mehmet Gorkem Ulkar
The paper investigates why direct pairwise AUC surrogate training, though aligned with the ranking metric, can fail to provide corrective updates during stochastic training. It introduces the concept of Corrective Power, showing that in certain regions severe ranking errors persist without generating strong gradient signals, a phenomenon linked to degeneracies in the U‑statistics of pairwise gradients. The authors explain empirical observations—such as the effectiveness of squared‑hinge loss, the benefit of cross‑entropy warm‑up, and the advantage of parameter‑efficient fine‑tuning—through this theoretical lens, and validate their claims with experiments on eight computer‑vision and eight NLP tasks using various models.
By Jia Chen, Zhiyong Yang, Shilong Bao, Qianqian Xu, Qingming Huang