Our findings show interesting trends, particularly pertaining to recent methods for data curation such as synthetic data generation and lookup based on CLIP embeddings. We show that although these strategies are highly competitive for certain tasks, the curation strategy used to assemble the original ImageNet-1K dataset remains the gold standard. We anticipate that our benchmark can illuminate the path for new methods to further reduce the gap.

Offline Reinforcement Learning (RL) presents a promising paradigm for training autonomous vehicle (AV) planning policies from large-scale, real-world driving logs. However, the extreme data imbalance in these logs, where mundane scenarios vastly outnumber rare "long-tail" events, leads to brittle and unsafe policies when using standard uniform data sampling. In this work, we address this challenge through a systematic, large-scale comparative study of data curation strategies designed to focus the learning process on information-rich samples. We investigate six distinct criticality weighting schemes which are categorized into three families: heuristic-based, uncertainty-based, and behavior-based. These are evaluated at two temporal scales, the individual timestep and the complete scenario. We train seven goal-conditioned Conservative Q-Learning (CQL) agents with a state-of-the-art, attention-based architecture and evaluate them in the high-fidelity Waymax simulator. Our results demonstrate that all data curation methods significantly outperform the baseline. Notably, data-driven curation using model uncertainty as a signal achieves the most significant safety improvements, reducing the collision rate by nearly three-fold (from 16.0% to 5.5%). Furthermore, we identify a clear trade-off where timestep-level weighting excels at reactive safety while scenario-level weighting improves long-horizon planning. Our work provides a comprehensive framework for data curation in Offline RL and underscores that intelligent, non-uniform sampling is a critical component for building safe and reliable autonomous agents.

Knowledge Distillation (KD) compresses large language models (LLMs) by transferring the teacher model's capabilities to a smaller student model, reducing inference cost and memory usage while maintaining performance. However, existing KD methods for LLMs often fail to prevent significant shifts in the student model's distribution during training, leading to issues such as catastrophic forgetting, mode collapse, and training-inference mismatch. To address these challenges, we propose a novel, plug-in curriculum learning framework inspired by the strength training principle of "progressive overload" (POCL), which can be seamlessly integrated into existing white-box KD approaches with minimal computational overhead. The framework comprises two core components: (1) a difficulty measurer that ranks and partitions training samples from easy to hard, and (2) a training scheduler that incrementally introduces these subsets into the distillation process at fixed intervals while applying loss functions with progressively rising temperatures. By starting with the easiest samples and progressively increasing the difficulty, the approach enhances both the stability and efficiency of learning. Extensive experiments in instruction-following settings demonstrate that POCL consistently improves the performance of distilled student models across various white-box KD methods and model families. Our findings highlight the effectiveness of sorted training samples in KD for LLMs. More generally, our work demonstrates how to structure training data within the KD process to enhance the stability and performance of distilled LLMs.

Data curation is the problem of how to collect and organize samples into a dataset that supports efficient learning. Despite the centrality of the task, little work has been devoted towards a large-scale, systematic comparison of various curation methods. In this work, we take steps towards a formal evaluation of data curation strategies and introduce SELECT, the first large-scale benchmark of curation strategies for image classification.In order to generate baseline methods for the SELECT benchmark, we create a new dataset, ImageNet, which constitutes the largest superset of ImageNet-1K to date. Our dataset extends ImageNet with 5 new training-data shifts, each approximately the size of ImageNet-1K, and each assembled using a distinct curation strategy. We evaluate our data curation baselines in two ways: (i) using each training-data shift to train identical image classification models from scratch (ii) using it to inspect a fixed pretrained self-supervised representation.Our findings show interesting trends, particularly pertaining to recent methods for data curation such as synthetic data generation and lookup based on CLIP embeddings.

Data curation is the problem of how to collect and organize samples into a dataset that supports efficient learning. Despite the centrality of the task, little work has been devoted towards a large-scale, systematic comparison of various curation methods. In this work, we take steps towards a formal evaluation of data curation strategies and introduce SELECT, the first large-scale benchmark of curation strategies for image classification. In order to generate baseline methods for the SELECT benchmark, we create a new dataset, ImageNet++, which constitutes the largest superset of ImageNet-1K to date. Our dataset extends ImageNet with 5 new training-data shifts, each approximately the size of ImageNet-1K itself, and each assembled using a distinct curation strategy. We evaluate our data curation baselines in two ways: (i) using each training-data shift to train identical image classification models from scratch (ii) using the data itself to fit a pretrained self-supervised representation. Our findings show interesting trends, particularly pertaining to recent methods for data curation such as synthetic data generation and lookup based on CLIP embeddings. We show that although these strategies are highly competitive for certain tasks, the curation strategy used to assemble the original ImageNet-1K dataset remains the gold standard. We anticipate that our benchmark can illuminate the path for new methods to further reduce the gap. We release our checkpoints, code, documentation, and a link to our dataset at https://github.com/jimmyxu123/SELECT.