Reward modeling lies at the core of reinforcement learning from human feedback (RLHF), yet most existing reward models rely on scalar or pairwise judgments that fail to capture the multifaceted nature of human preferences.Recent studies have explored rubrics-as-rewards (RaR) that uses structured criteria to capture multiple dimensions of response quality.However, producing rubrics that are both reliable and scalable remains a key challenge.In this work, we introduce OpenRubrics, a diverse, large-scale collection of (prompt, rubric) pairs for training rubric-generation and rubric-based reward models.To elicit discriminative and comprehensive evaluation signals, we introduce Contrastive Rubric Generation (CRG), which derives both hard rules (explicit constraints) and principles (implicit qualities) by contrasting preferred and rejected responses.We further remove noisy rubrics via preserving preference-label consistency.Across multiple reward-modeling benchmarks, our rubricbased reward model, RUBRIC-RM, surpasses strong size-matched baselines by 8.4%.These gains transfer to policy models on instructionfollowing and biomedical benchmarks.The model weights and datasets are publicly available at https://huggingface.co/OpenRubrics.* These authors contributed equally to this work, order was determined randomly (by rolling a die).

Fine-tuning large pre-trained foundation models often yields excellent downstream performance but is prohibitively expensive when updating all parameters. Parameter-efficient fine-tuning (PEFT) methods such as LoRA alleviate this by introducing lightweight update modules, yet they commonly rely on weight-agnostic linear approximations, limiting their expressiveness. In this work, we propose PEANuT, a novel PEFT framework that introduces weight-aware neural tweakers, compact neural modules that generate task-adaptive updates conditioned on frozen pre-trained weights. PEANuT provides a flexible yet efficient way to capture complex update patterns without full model tuning. We theoretically show that PEANuT achieves equivalent or greater expressivity than existing linear PEFT methods with comparable or fewer parameters. Extensive experiments across four benchmarks with over twenty datasets demonstrate that PEANuT consistently outperforms strong baselines in both NLP and vision tasks, while maintaining low computational overhead.

Large language models (LLMs) have achieved remarkable performance on vari- ous natural language tasks. However, they are trained on static corpora and their knowledge can become outdated quickly in the fast-changing world. This moti- vates the development of knowledge editing methods designed to update certain knowledge in LLMs without changing unrelated others. To make selective edits, previous efforts often sought to update a small amount of parameters in some spe- cific layer(s) of a LLM. Nonetheless, in challenging scenarios, they still fall short in making successful edits while preserving knowledge irrelevant to the updates simultaneously, resulting in a notable editing-locality trade-off. In this work, we question if the trade-offs are caused by the fact that parameter-based updates have a global effect, i.e., edited parameters affect all inputs indiscriminately. In light of this, we explore the feasibility of representation fine-tuning, which applied some linear update to a few representations in a learned subspace, for knowledge edit- ing. While being effective to enhance an LLM’s general ability as demonstrated in the previous work, we theoretically show that this linear update imposes a tension in editing-locality trade-off. Subsequently, BaFT is proposed to break the linear- ity. BaFT computes a weight for each basis that spans a dimension of the subspace based on the input representation. This input-dependent weighting mechanism al- lows BaFT to manage different types of knowledge in an adaptive way, thereby achieving a better editing-locality trade-off. Experiments on three LLMs with five editing benchmarks in diverse scenarios show the superiority of our method.

Large language models (LLMs) have achieved remarkable performance on various natural language tasks. However, they are trained on static corpora and their knowledge can become outdated quickly in the fast-changing world. This motivates the development of knowledge editing (KE) to update specific knowledge in LLMs without changing unrelated others or compromising their pre-trained capabilities. Previous efforts sought to update a small amount of parameters of a LLM and proved effective for making selective updates. Nonetheless, the edited LLM often exhibits degraded ability to reason about the new knowledge. In this work, we identify a key issue: heterogeneous token overfitting (HTO), where the LLM overfits different tokens in the provided knowledge at varying rates. To tackle this, we propose OVERTONE, a token-level smoothing method that mitigates HTO by adaptively refining the target distribution. Theoretically, OVERTONE offers better parameter updates with negligible computation overhead. It also induces an implicit DPO but does not require preference data pairs. Extensive experiments across four editing methods, two LLMs, and diverse scenarios demonstrate the effectiveness and versatility of our method.

Using 3D hand poses as the input of user interfaces can enable many novel human-computer interaction applications. However, conventional solutions for precisely reconstructing the hand poses are either vision-based, which are compute-intensive and may cause privacy issues, or wearable devices-based, which are intrusive to users. In this paper, we propose RAM-Hand, a Robust Acoustic 3D Multi-Hand pose reconstruction system built on a microphone array. Our RAM-Hand system can support multiple hands and is designed to be highly adaptable to new scenarios even when training data is limited. Specifically, it should robustly accommodate variations in environment, subject, and hand positions. To achieve this, on one hand, we propose a customized signal processing pipeline to segment multiple hands' reflections and extract the features corresponding to each hand, then feed those features into a transformer-based neural network for precise pose reconstruction. On the other hand, to tackle the challenge that the training data is limited, we propose a series of data augmentation methods to generate virtual training data, and utilize contrastive learning to ensure our model behaves well on new subjects. We conduct extensive experiments on a real-world microphone array testbed to evaluate the performance of the proposed system. The results show that our RAM-Hand system can localize each hand joint with an average error of 10.71 mm, handle multiple hands, and generalize well to the above mentioned new scenarios.

As AI advances, copyrighted content faces growing risk of unauthorized use, whether through model training or direct misuse. Building upon invisible adversarial perturbation, recent works developed copyright protections against specific AI techniques such as unauthorized personalization through DreamBooth that are misused. However, these methods offer only short-term security, as they require retraining whenever the underlying model architectures change. To establish long-term protection aiming at better robustness, we go beyond invisible perturbation, and propose a universal approach that embeds \textit{visible} watermarks that are \textit{hard-to-remove} into images. Grounded in a new probabilistic and inverse problem-based formulation, our framework maximizes the discrepancy between the \textit{optimal} reconstruction and the original content. We develop an effective and efficient approximation algorithm to circumvent a intractable bi-level optimization. Experimental results demonstrate superiority of our approach across diverse scenarios.

Pre-trained language models, trained on largescale corpora, demonstrate strong generalizability across various NLP tasks.Finetuning these models for specific tasks typically involves updating all parameters, which is resource-intensive.Parameter-efficient finetuning (PEFT) methods, such as the popular LoRA family, introduce low-rank matrices to learn only a few parameters efficiently.However, during inference, the product of these matrices updates all pre-trained parameters, complicating tasks like knowledge editing that require selective updates.We propose a novel PEFT method, which conducts row and column-wise sparse low-rank adaptation (RoseLoRA), to address this challenge.RoseLoRA identifies and updates only the most important parameters for a specific task, maintaining efficiency while preserving other model knowledge.By adding a sparsity constraint on the product of low-rank matrices and converting it to row and column-wise sparsity, we ensure efficient and precise model updates.Our theoretical analysis guarantees the lower bound of the sparsity with respective to the matrix product.Extensive experiments on five benchmarks across twenty datasets demonstrate that RoseLoRA outperforms baselines in both general fine-tuning and knowledge editing tasks.

Fair machine learning seeks to mitigate model prediction bias against certain demographic subgroups such as elder and female. Recently, fair representation learning (FRL) trained by deep neural networks has demonstrated superior performance, whereby representations containing no demographic information are inferred from the data and then used as the input to classification or other downstream tasks. Despite the development of FRL methods, their vulnerability under data poisoning attack, a popular protocol to benchmark model robustness under adversarial scenarios, is under-explored. Data poisoning attacks have been developed for classical fair machine learning methods which incorporate fairness constraints into shallow-model classifiers. Nonetheless, these attacks fall short in FRL due to notably different fairness goals and model architectures. This work proposes the first data poisoning framework attacking FRL. We induce the model to output unfair representations that contain as much demographic information as possible by injecting carefully crafted poisoning samples into the training data. This attack entails a prohibitive bilevel optimization, wherefore an effective approximated solution is proposed. A theoretical analysis on the needed number of poisoning samples is derived and sheds light on defending against the attack. Experiments on benchmark fairness datasets and state-of-the-art fair representation learning models demonstrate the superiority of our attack.

Large language models (LLMs) have achieved impressive performance on various natural language generation tasks. Nonetheless, they suffer from generating negative and harmful contents that are biased against certain demographic groups (e.g., female), raising severe fairness concerns. As remedies, prior works intervened the generation by removing attitude or demographic information, inevitably degrading the generation quality and resulting in notable fairness-fluency trade-offs. However, it is still under-explored to what extent the fluency has to be affected in order to achieve a desired level of fairness. In this work, we conduct the first formal study from an information-theoretic perspective. We show that previous approaches are excessive for debiasing and propose LIDAO, a general framework to debias a (L)LM at a better fluency provably. We further robustify LIDAO in adversarial scenarios, where a carefully-crafted prompt may stimulate LLMs exhibiting instruction-following abilities to generate texts with fairness issue appears only when the prompt is also taken into account. Experiments on three LMs ranging from 0.7B to 7B parameters demonstrate the superiority of our method.

In federated learning (FL), accommodating clients' varied computational capacities poses a challenge, often limiting the participation of those with constrained resources in global model training. To address this issue, the concept of model heterogeneity through submodel extraction has emerged, offering a tailored solution that aligns the model's complexity with each client's computational capacity. In this work, we propose Federated Importance-Aware Submodel Extraction (FIARSE), a novel approach that dynamically adjusts submodels based on the importance of model parameters, thereby overcoming the limitations of previous static and dynamic submodel extraction methods. Compared to existing works, the proposed method offers a theoretical foundation for the submodel extraction and eliminates the need for additional information beyond the model parameters themselves to determine parameter importance, significantly reducing the overhead on clients. Extensive experiments are conducted on various datasets to showcase the superior performance of the proposed FIARSE.

In high-stakes domains such as healthcare and hiring, the role of machine learning (ML) in decision-making raises significant fairness concerns. This work focuses on Counterfactual Fairness (CF), which posits that an ML model's outcome on any individual should remain unchanged if they had belonged to a different demographic group. Previous works have proposed methods that guarantee CF. Notwithstanding, their effects on the model's predictive performance remain largely unclear. To fill this gap, we provide a theoretical study on the inherent trade-off between CF and predictive performance in a model-agnostic manner. We first propose a simple but effective method to cast an optimal but potentially unfair predictor into a fair one with a minimal loss of performance. By analyzing the excess risk incurred by perfect CF, we quantify this inherent trade-off. Further analysis on our method's performance with access to only incomplete causal knowledge is also conducted. Built upon this, we propose a practical algorithm that can be applied in such scenarios. Experiments on both synthetic and semi-synthetic datasets demonstrate the validity of our analysis and methods.

Millimeter-wave (mmWave) based human activity recognition (HAR) systems have demonstrated promising performance in various applications, leveraging the power of deep neural networks. However, these systems are suffering from the scarcity of available mmWave data for model training. To address this challenge, we explore the possibility of transferring knowledge from large AI models built on massive text and visual data to enhance the generalizability of mmWave-based HAR models. Towards this end, we introduce mmCLIP, a novel system that aligns mmWave signal space and text space to facilitate zero-shot recognition for unseen activities. To enable this alignment, we employ cross-modality signal synthesis to augment mmWave signal data using large human mesh datasets and design an activity attribute decomposition and recomposition approach to characterize the semantic interconnections among activities. We conducted extensive experiments to demonstrate the effectiveness of our proposed framework.

Multi-modal sensing systems are becoming increasingly common in real-world applications like human activity recognition (HAR). To enable knowledge sharing among individuals, Federated Learning (FL) offers a solution as a distributed machine learning paradigm that retains user data locally, thereby safeguarding privacy. However, existing heterogeneous multi-modal Federated Learning (MMFL) solutions have yet to fully utilize all the potential knowledge-sharing opportunities, as they fail to capture fundamental common knowledge that is independent of both modality and client. In this paper, we propose Federated Hierarchical Knowledge Disentanglement (FedHKD), a new sensing system for heterogeneous multi-modal federated learning. FedHKD introduces a multi-stage training paradigm based on hierarchical knowledge disentanglement at both the modality and client levels. This design enhances collaboration among modality-heterogeneous clients while maintaining low storage overhead and high adaptation flexibility to new sensing modalities. Our evaluation of two public real-world multi-modal HAR datasets and a self-collected dataset demonstrates that FedHKD outperforms state-of-the-art baselines by up to 4.85% in accuracy while saving up to 2.29× in storage. Additionally, when adapting to new sensing modalities, it reduces communication overhead by up to 4.62×.

Recent years have witnessed increasing concerns towards unfair decisions made by machine learning algorithms. To improve fairness in model decisions, various fairness notions have been proposed and many fairness-aware methods are developed. However, most of existing definitions and methods focus only on single-label classification. Fairness for multi-label classification, where each instance is associated with more than one labels, is still yet to establish. To fill this gap, we study fairness-aware multi-label classification in this paper. We start by extending Demographic Parity (DP) and Equalized Opportunity (EOp), two popular fairness notions, to multi-label classification scenarios. Through a systematic study, we show that on multi-label data, because of unevenly distributed labels, EOp usually fails to construct a reliable estimate on labels with few instances. We then propose a new framework named Similarity s-induced Fairness (sγ -SimFair). This new framework utilizes data that have similar labels when estimating fairness on a particular label group for better stability, and can unify DP and EOp. Theoretical analysis and experimental results on real-world datasets together demonstrate the advantage of sγ -SimFair over existing methods on multi-label classification tasks.

Incorporating a deep generative model as the prior distribution in inverse problems has established substantial success in reconstructing images from corrupted observations. Notwithstanding, the existing optimization approaches use gradient descent largely without adapting to the non-convex nature of the problem and can be sensitive to initial values, impeding further performance improvement. In this paper, we propose an efficient amortized optimization scheme for inverse problems with a deep generative prior. Specifically, the optimization task with high degrees of difficulty is decomposed into optimizing a sequence of much easier ones. We provide a theoretical guarantee of the proposed algorithm and empirically validate it on different inverse problems. As a result, our approach outperforms baseline methods qualitatively and quantitatively by a large margin.