Aligning large language models (LLMs) with safety guidelines typically involves reinforcement learning from human feedback (RLHF), relying on human-generated preference annotations. However, assigning consistent overall quality ratings is challenging, prompting recent research to shift towards detailed evaluations based on multiple specific safety criteria. This paper uncovers a consistent observation: safety rules characterized by high rating entropy are generally less reliable in identifying responses preferred by humans. Leveraging this finding, we introduce ENCORE, a straightforward entropy-guided approach that composes multi-head rewards by downweighting rules exhibiting high rating entropy. Theoretically, we demonstrate that rules with elevated entropy naturally receive minimal weighting in the Bradley-Terry optimization framework, justifying our entropy-based penalization. Through extensive experiments on RewardBench safety tasks, our method significantly surpasses several competitive baselines, including random weighting, uniform weighting, single-head Bradley-Terry models, and LLM-based judging methods. Our proposed approach is training-free, broadly applicable to various datasets, and maintains interpretability, offering a practical and effective solution for multi-attribute reward modeling.

Reinforcement Learning from Human Feedback (RLHF) is commonly employed to tailor models to human preferences, especially to improve the safety of outputs from large language models (LLMs). Traditionally, this method depends on selecting preferred responses from pairs. However, due to the variability in human opinions and the challenges in directly comparing two responses, there is an increasing trend towards fine-grained annotation approaches that evaluate responses using multiple targeted metrics or rules. The challenge lies in efficiently choosing and applying these rules to handle the diverse range of preference data. In this paper, we propose a dynamic method that adaptively selects the most important rules for each response pair. We introduce a mathematical framework that utilizes the maximum discrepancy across paired responses and demonstrate theoretically that this approach maximizes the mutual information between the rule-based annotations and the underlying true preferences. We then train an 8B reward model using this adaptively labeled preference dataset and assess its efficacy using RewardBench. As of January 25, 2025, our model achieved the highest safety performance on the leaderboard, surpassing various larger models.

There are increasing studies using LLMs to rate and select data based on several human-crafted metrics (rules). However, these conventional rule-based approaches often depend too heavily on human heuristics, lack effective metrics for assessing rules, and exhibit limited adaptability to new tasks. In our study, we introduce an innovative rule-based framework that utilizes the orthogonality of score vectors associated with rules as a novel metric for rule evaluations. Our approach includes an automated pipeline that first uses LLMs to generate a diverse set of rules, encompassing various rating dimensions to evaluate data quality. Then it rates a batch of data based on these rules and uses the determinantal point process (DPP) from random matrix theory to select the most orthogonal score vectors, thereby identifying a set of independent rules. These rules are subsequently used to evaluate all data, selecting samples with the highest average scores for downstream tasks such as LLM training. We verify the effectiveness of our method through two experimental setups: 1) comparisons with ground truth ratings and 2) benchmarking LLMs trained with the chosen data. Our comprehensive experiments cover a range of scenarios, including general pre-training and domain-specific fine-tuning in areas such as IMDB, Medical, Math, and Code. The outcomes demonstrate that our DPP-based rule rating method consistently outperforms other approaches, including rule-free rating, uniform sampling, importance resampling, and QuRating, in terms of both rating precision and model performance.

Identifying key product features that influence consumer preferences is essential in the fashion industry. In this study, we introduce a robust methodology to ascertain the most impactful features in fashion product images, utilizing past market sales data. First, we propose the metric called "influence score" to quantitatively assess the importance of product features. Then we develop a forecasting model, the Fashion Demand Predictor (FDP), which integrates Transformer-based models and Random Forest to predict market popularity based on product images. We employ image-editing diffusion models to modify these images and perform an ablation study, which validates the impact of the highest and lowest-scoring features on the model's popularity predictions. Additionally, we further validate these results through surveys that gather human rankings of preferences, confirming the accuracy of the FDP model's predictions and the efficacy of our method in identifying influential features. Notably, products enhanced with "good" features show marked improvements in predicted popularity over their modified counterparts. Our approach develops a fully automated and systematic framework for fashion image analysis that provides valuable guidance for downstream tasks such as fashion product design and marketing strategy development.

Large language models (LLMs) have shown remarkable proficiency in generating text, benefiting from extensive training on vast textual corpora. Machine unlearning has been introduced as a viable solution to remove the influence of such problematic content without the need for costly and time-consuming retraining. This process aims to erase specific knowledge from LLMs while preserving as much model utility as possible. Despite the effectiveness of current unlearning methods, little attention has been given to whether existing unlearning methods for LLMs truly achieve forgetting or merely hide the knowledge, which current unlearning benchmarks fail to detect. This paper reveals that applying quantization to models that have undergone unlearning can restore the "forgotten" information. We conduct comprehensive experiments using various quantization techniques across multiple precision levels to thoroughly evaluate this phenomenon. We find that for unlearning methods with utility constraints, the unlearned model retains an average of 21% of the intended forgotten knowledge in full precision, which significantly increases to 83% after 4-bit quantization. Based on our empirical findings, we provide a theoretical explanation for the observed phenomenon and propose a quantization-robust unlearning strategy aimed at mitigating this intricate issue. Our results highlight a fundamental tension between preserving the utility of the unlearned model and preventing knowledge recovery through quantization, emphasizing the challenge of balancing these two objectives. Altogether, our study underscores a major failure in existing unlearning methods for LLMs, strongly advocating for more comprehensive and robust strategies to ensure authentic unlearning without compromising model utility. Large language models (LLMs) have exhibited remarkable abilities in generating human-like text, owing to their training on extensive datasets (Zhao et al., 2023). However, LLMs can also unintentionally learn and reproduce undesirable behaviors from sensitive training data (Liu et al., 2024a; Sun et al., 2024). Furthermore, laws such as the European Union General Data Protection Regulation (GDPR) (Voigt & Von dem Bussche, 2017) have introduced the "Right to be Forgotten", allowing users to request the removal of their personal data from trained models (Xu et al., 2024a). FP32 "There's more in the frying pan," Petunia, turning eyes on said Aunt her massive son.

High-resolution (HR) 3D magnetic resonance imaging (MRI) can provide detailed anatomical structural information, enabling precise segmentation of regions of interest for various medical image analysis tasks. Due to the high demands of acquisition device, collection of HR images with their annotations is always impractical in clinical scenarios. Consequently, segmentation results based on low-resolution (LR) images with large slice thickness are often unsatisfactory for subsequent tasks. In this paper, we propose a novel Resource-Efficient High-Resolution Segmentation framework (REHRSeg) to address the above-mentioned challenges in real-world applications, which can achieve HR segmentation while only employing the LR images as input. REHRSeg is designed to leverage self-supervised super-resolution (self-SR) to provide pseudo supervision, therefore the relatively easier-to-acquire LR annotated images generated by 2D scanning protocols can be directly used for model training. The main contribution to ensure the effectiveness in self-SR for enhancing segmentation is three-fold: (1) We mitigate the data scarcity problem in the medical field by using pseudo-data for training the segmentation model. (2) We design an uncertainty-aware super-resolution (UASR) head in self-SR to raise the awareness of segmentation uncertainty as commonly appeared on the ROI boundaries. (3) We align the spatial features for self-SR and segmentation through structural knowledge distillation to enable a better capture of region correlations. Experimental results demonstrate that REHRSeg achieves high-quality HR segmentation without intensive supervision, while also significantly improving the baseline performance for LR segmentation.

Machine learning tasks involving biomedical signals frequently grapple with issues such as limited data availability, imbalanced datasets, labeling complexities, and the interference of measurement noise. These challenges often hinder the optimal training of machine learning algorithms. Addressing these concerns, we introduce BioDiffusion, a diffusion-based probabilistic model optimized for the synthesis of multivariate biomedical signals. BioDiffusion demonstrates excellence in producing high-fidelity, non-stationary, multivariate signals for a range of tasks including unconditional, label-conditional, and signal-conditional generation. Leveraging these synthesized signals offers a notable solution to the aforementioned challenges. Our research encompasses both qualitative and quantitative assessments of the synthesized data quality, underscoring its capacity to bolster accuracy in machine learning tasks tied to biomedical signals. Furthermore, when juxtaposed with current leading time-series generative models, empirical evidence suggests that BioDiffusion outperforms them in biomedical signal generation quality.

Signal measurements appearing in the form of time series are one of the most common types of data used in medical machine learning applications. However, such datasets are often small, making the training of deep neural network architectures ineffective. For time-series, the suite of data augmentation tricks we can use to expand the size of the dataset is limited by the need to maintain the basic properties of the signal. Data generated by a Generative Adversarial Network (GAN) can be utilized as another data augmentation tool. RNN-based GANs suffer from the fact that they cannot effectively model long sequences of data points with irregular temporal relations. To tackle these problems, we introduce TTS-GAN, a transformer-based GAN which can successfully generate realistic synthetic time-series data sequences of arbitrary length, similar to the real ones. Both the generator and discriminator networks of the GAN model are built using a pure transformer encoder architecture. We use visualizations and dimensionality reduction techniques to demonstrate the similarity of real and generated time-series data. We also compare the quality of our generated data with the best existing alternative, which is an RNN-based time-series GAN.