The paper "Fairness Definitions in Language Models Explained" surveys and clarifies various fairness notions applicable to language models (LMs). It introduces a taxonomy categorizing these concepts by transformer architecture—encoder-only, decoder-only, and encoder-decoder—and demonstrates each definition through experiments to illustrate practical implications. The authors also discuss current research challenges and open questions to encourage further innovation in the field.
By Zhipeng Yin, Zichong Wang, Avash Palikhe, Wenbin Zhang
The paper introduces SSR, a simple yet effective data selection strategy for million‑scale multi‑modal instruction‑tuning. SSR uses interaction styles as a proxy for structural diversity and decomposes data quality into capability‑specific scores, forming a rich profile for each sample. By casting selection as submodular coverage maximization, SSR consistently outperforms random sampling across two architecture stacks, multiple data pools, and 14 benchmarks, achieving 99.1% of full‑data performance with only 30% of a 2.6M pool.
By Mengyao Lyu, Yan Li, Huasong Zhong, Hui Chen, Jungong Han, Guiguang Ding, Zhenheng Yang
The paper examines whether modern drift‑adaptive malware classifiers provide reliable and stable confidence estimates when faced with distribution shifts. It introduces Aurora, a framework that evaluates confidence quality and operational resilience, and presents metrics that extend beyond point‑in‑time performance to assess stability over time. The study finds that state‑of‑the‑art models exhibit fragility across datasets with varying drift severity, indicating a need to revisit underlying assumptions.
By Alexander Herzog, Aliai Eusebi, Lorenzo Cavallaro
The paper introduces a doubly robust preference optimization algorithm for reinforcement learning from human feedback (RLHF) in large language models. It addresses sensitivity to misspecifications in the preference model, reference policy, or reward function by ensuring consistency when either the preference model or the reference policy is correctly specified. Experiments show that the method outperforms state‑of‑the‑art algorithms both theoretically and empirically.
By Erhan Xu, Kai Ye, Hongyi Zhou, Luhan Zhu, Francesco Quinzan, Chengchun Shi
The study explores how K‑12 teachers adapt generative AI (GenAI) to fit local classroom needs, highlighting a mismatch between GenAI’s default cultural outputs and teachers’ everyday practices. By interviewing 30 teachers from South Africa, Taiwan, and the United States, the authors introduce the concept of cultural distance, categorizing teacher effort into low, mid, and high levels based on the usability of AI outputs. The research suggests that while minor edits may suffice in some contexts, more substantial prompting or structural changes are required in others, underscoring the need for designers, policymakers, and educators to address these gaps.
By Qing Xiao, Ruiwei Xiao, Hanqi Jane Li, Phenyo Phemelo Moletsane, Xinying Hou, John Stamper, Hong Shen
The paper introduces SELF, an intrinsic weight‑based fingerprinting method for large language models that uses singular value and eigenvalue decomposition of attention weights to generate unique, transformation‑invariant fingerprints. SELF claims to eliminate the need for input data, resist false claim attacks, and reduce false positives compared to existing techniques. Experiments show that SELF achieves high IP infringement detection accuracy and remains robust against quantization, pruning, and fine‑tuning modifications.
By Hanxiu Zhang, Qiang Xu, Yue Zheng
The paper presents a novel dimensionality‑reduction framework for reach‑avoid games involving unmanned aerial vehicles. By decomposing the three‑dimensional problem into horizontal and vertical sub‑games, each solved with Hamilton‑Jacobi reachability analysis, the authors reconstruct the full solution using a tracking control algorithm that guarantees attacker capture and subsequent tracking. Numerical simulations and Gazebo simulations confirm that the decomposition preserves optimality and achieves successful quadrotor capture in three‑dimensional space.
By Minh Bui, Simon Monckton, Mo Chen
RePO‑VLA is a recovery‑driven policy optimization framework for Vision‑Language‑Action models that separates success, recovery, and failure trajectories. It introduces Recovery‑Aware Initialization, a Progress‑Aware Semantic Value Function, and Value‑Conditioned Refinement to learn from both successful and failed rollouts, enabling the policy to prefer high‑progress actions without online failure detectors. The authors also present the Failure‑Recovery Benchmark (FRBench) and demonstrate that RePO‑VLA raises adversarial success rates from 20% to 75% on average in simulated and real‑world bimanual tasks.
By Weijia Liufu, Xiaoyu Guo, Ruiyi Chen, Jingzhi Liu, Kaidong Zhang, Xiwen Liang, Jianqi Lin, Dawei Sun, Yuze Wang, Rongtao Xu, Bingqian Lin, Bowen Yang, Tongtong Cao, Bowen Peng, Dongyu Zhang, Guangrun Wang, Min Wang, Liang Lin, Xiaodan Liang
The paper investigates how pretraining and alignment stages of transformer training affect weight updates in different matrix groups. By examining the residual stream’s shared coordinate system, it finds that alignment updates to query and key matrices align with the principal input subspace, whereas updates to output matrices remain largely random relative to the prediction subspace. This pattern persists across objectives and training stages, indicating that the geometry of updates is determined by the training phase rather than the specific objective.
By Valeria Ruscio, Keiran Thompson
The paper introduces GAUGE, a training‑free post‑hoc module that rectifies view‑inconsistency in feed‑forward 4D tracking models. It identifies motion groups via direction consistency and spatial connectivity, then estimates per‑frame radial scale and group‑level translation using a small set of metric anchors. Applying GAUGE to eight trackers reduces endpoint error by 15.1%–62.6%, outperforming gradient fine‑tuning with the same anchors.
By Zhuoqian Feng, Weixing Chen, Ziliang Chen, Yang Liu, Liang Lin
Pinterest’s new system, RLBP, combines reinforcement learning with traditional PID controllers to tune bidding parameters for each campaign daily. The hybrid architecture uses a multi‑head Q‑network trained on a relative‑lift reward to adapt the cost‑delivery trade‑off without retraining. Online A/B tests and post‑rollout measurements show a statistically significant drop in effective cost‑per‑click with no revenue loss.
By Yichu Zhou, Qinyi Chen, Fei Feng, Jacob Loh, Mehdi Ben Ayed, Lin Yang, Andrei Curelea, Jiajing Xu
The paper introduces a node‑level relaxation of individual fairness called local predictive consistency, where each prediction is compared to a similarity‑weighted average of predictions in its local neighborhood. This approach yields an interpretable fairness score and a regularized learning method based on local residuals, and it explicitly constructs the neighborhood graph with threshold selection guided by graph‑coverage diagnostics. The authors apply the method to binary classification, regression, and time‑to‑event prediction, and compare it with pairwise constraint‑based fairness formulations to analyze fairness‑performance trade‑offs.
By Zhouting Zhao, Tin Lok James Ng
The paper introduces a unified attack framework that can simultaneously target multiple large language model (LLM) transformations such as quantization, pruning, and model merging. By treating each transformation as a weight perturbation, the authors jointly train a model to perform benignly before transformation and maliciously after any one of the transformations is applied. Experiments across various attack scenarios and model families show high success rates, illustrating that a single adversarial model can exploit diverse deployment-time transformations.
By Xiaohua Zhan, Kazuki Egashira, Robin Staab, Mark Vero, Martin Vechev
The paper introduces Augmentation-Guided Adversarial Training (AGAT), a selective adversarial training method that uses augmentation-induced local variation to identify unstable samples for training. AGAT employs Augmentation-based Local Stability (ALS) to rank samples based on the normalized change in model logits between an input and its label-preserving augmented version. Experiments on MNIST, CIFAR-10, CIFAR-100, and Tiny-ImageNet-200 show that AGAT achieves comparable or higher robust accuracy while applying adversarial supervision to only 20‑30% of training samples per minibatch, and it requires less training time than other adversarial training baselines.
By Daniel Agyei Asante, Mark Tetteh, Deborah Asamoah, Bernard Asante, Regina Esi Turkson
The paper introduces RBGA, an algorithm for offline-to-online safe reinforcement learning in dexterous manipulation. RBGA tracks the remaining safety budget during an episode and uses it to arbitrate between offline guidance and an evolving online policy, comparing cost-to-go estimates at each step. The authors provide theoretical analysis of safety implications and demonstrate through experiments that RBGA improves online policy performance while preserving safety.
By Zongyou Yu, Ao Zhou, Xiongbo Shi, Jiayi Guan, Shihang Du, Xudong Zhang, Shangding Gu, Junqiao Zhao, Hanli Wang, Guang Chen
The paper introduces EISep RNN, a recurrent neural network that enforces Dale's principle—each neuron is either exclusively excitatory or inhibitory—through a learnable excitatory‑inhibitory separation. Experiments across multi‑task learning, continual learning, and reinforcement learning show that EISep yields more stable training, competitive or improved performance, and sparser, more modular connectivity compared to vanilla RNNs and prior Dale‑constrained models. The architecture also produces stable and informative latent dynamics, demonstrating that Dale's principle can serve as a biologically grounded inductive bias for RNNs.
By Jiyi Wang, Chenxiao Yang, Yujia Zhao, Jingzhao Zhang, Lu Mi
The paper introduces Aligned‑PAC learning, a framework that reframes the alignment problem as a question of generalization: whether success on training requirements extends to unseen ones. It proposes representing intentions through multiple specific requirements, provides theoretical results on independent evaluation and the amount of evaluation needed, and develops an optimization method that combines learned models to satisfy these requirements under certain assumptions. The authors acknowledge that selecting appropriate requirements and ensuring generalization to new ones remain open challenges, positioning their work as a foundation for precise, learning‑theoretic study of alignment.
By Shai Shalev-Shwartz, Amnon Shashua
The paper presents a robust approximate unlearning method for Generalized Learning Vector Quantization (GLVQ) classifiers, enabling the removal of specific data samples from trained models. It evaluates how the size and characteristics of the forget set affect unlearning, proposes an adapted technique to mitigate instabilities, and demonstrates that the method achieves near-zero deviation from a fully retrained model while being up to 10^5 times more efficient on benchmark tabular datasets. The approach is applicable to any GLVQ model without altering its training scheme.
By Sreejita Ghosh, Mohammad Mohammadi
TRACE is a memory‑efficient, replay‑free framework for Time Series Class‑Incremental Learning that uses generative modeling to preserve class‑specific knowledge. It employs a compact, frozen temporal‑spectral representation and lightweight class‑specific autoencoders to capture intra‑class variations without retaining historical samples or large statistical matrices. Experiments on five benchmarks show that TRACE achieves competitive accuracy while reducing retained memory by about 98.26% compared with the strongest baseline and improving accuracy on GrabMyo by roughly 33.43%.
By Junhui Mao, Junru Zhang, Liming Ye, Cheng Peng, Haoran Shi, Xiaoli Tang, Baokun Wang, Bin Dou, Yabo Dong, Han Yu
IntNet is a multi-agent reinforcement learning framework designed to improve safety in autonomous driving by integrating vehicle intent transmission and adaptive communication scheduling into a single end-to-end learning system. The model uses an observation prediction module for agents to forecast future policy outputs and share them over a vehicle-to-vehicle network, while a Graph Attention Network processes incoming messages and a scheduler optimizes the communication graph for bandwidth efficiency. Experiments in complex urban environments show that IntNet achieves better coordination, lower collision rates, and reduces communication effort by up to 60% compared to state‑of‑the‑art communicative MARL methods.
By Leandro Parada, Kevin Yu, Panagiotis Angeloudis