Recent advances in Multimodal In-Context Learning (M-ICL) for Multimodal Large Language Models (MLLMs) have attracted considerable attention. These developments primarily focus on configuring an in-context sequence for a given test case based on instance-level semantic similarity. However, high similarity among demonstrations in the sequence introduces inductive biases, which may mislead MLLMs and ultimately degrade their overall performance. To address this, we propose a novel cluster-based in-context configuration method that adaptively groups candidate data and selects demonstrations from each cluster. This method enhances the diversity within the sequence while preserving semantic consistency, enabling MLLMs to focus on the main intent of the demonstrations. The experimental results on four Visual Question Answering (VQA) benchmarks, including OK-VQA, VQAv2, VizWiz, and TextVQA, demonstrate the effectiveness of our proposed method.

Data attribution methods quantify the influence of training data on model outputs and are becoming increasingly relevant for a wide range of LLM research and applications, including dataset curation, model interpretability, data valuation. However, there remain critical gaps in systematic LLM-centric evaluation of data attribution methods. To this end, we introduce DATE-LM (Data Attribution Evaluation in Language Models), a unified benchmark for evaluating data attribution methods through real-world LLM applications. DATE-LM measures attribution quality through three key tasks — training data selection, toxicity/bias filtering, and factual attribution. Our benchmark is designed for ease of use, enabling researchers to configure and run large-scale evaluations across diverse tasks and LLM architectures. Furthermore, we use DATE-LM to conduct a large-scale evaluation of existing data attribution methods. Our findings show that no single method dominates across all tasks, data attribution methods have trade-offs with simpler baselines, and method performance is sensitive to task-specific evaluation design. Finally, we release a public leaderboard for quick comparison of methods and to facilitate community engagement. We hope DATE-LM serves as a foundation for future data attribution research in LLMs.

Fine-grained action recognition typically faces challenges with lower inter-class variances and higher intra-class variances. Supervised contrastive learning is inherently suitable for this task, as it can decrease intra-class feature distances while increasing inter-class ones. However, directly applying it into fine-grained action recognition encounters two main problems. The first problem stems from the heavy training cost associated with supervised contrastive learning, which requires numerous training epochs, each involving double augmentation views per instance. To address this issue, we propose the late-stage supervised contrastive learning (late-SC) strategy, which effectively reduces the number of training epochs needed for the contrastive learning process. The second problem is that supervised contrastive loss does not explicitly consider the semantic distances between fine-grained actions when adjusting representation distances. This results in less reasonable and efficient adjustments to the representation space. To overcome this limitation, we introduce the semantic-aware temperature adaptation (STA) mechanism, enhancing the suitability of the supervised contrastive loss for fine-grained action recognition. We conduct experiments on several benchmark datasets for fine-grained action recognition, including Epic-Kitchens-55/100, SomethingSomething-V1, and Diving48-V2. The results demonstrate that our proposed method (referred to as LSC-STA) consistently enhances performance across various base feature extractors, without introducing additional inference overhead and incurring only a marginal increase in training expenses.

Optical flow is a widely used data source for learning motion information, but the complete loss of appearance information limits its ability for action recognition. Therefore we think of enriching optical flow frames with supplementary appearance information to form a new motion data source denoted as Appearance-Supplemented Optical Flow (ASOF). Specifically, we propose a data embedding layer and a stagewise training method to mitigate the scale-wise and density-wise data distribution divergence between the RGB and optical flow frames respectively. We conduct experiments on three benchmark datasets: UCF101 [1], HMDB51 [2] and SomethingSomething-V1 [3]. The results show that our methods can prominently improve optical flow stream recognition accuracy, and further improve the performances of two-stream (score fusion with the RGB stream) with only a little storage increase.