arXiv:2602.18584v1 Announce Type: cross Abstract: Targeted data selection has emerged as a crucial paradigm for efficient instruction tuning, aiming to identify a small yet influential subset of training examples for a specific target task. In practice, influence is often measured through the effect of an example on parameter updates. To make selection scalable, many approaches leverage optimizer statistics (e.g., Adam states) as an axis-aligned surrogate for update geometry (i.e., diagonal precondition), implicitly treating parameters as coordinate-wise independent. We show that this assumption breaks down in parameter-efficient fine-tuning (PEFT) methods such as LoRA. In this setting, the induced optimization geometry exhibits strong cross-parameter coupling with non-trivial off-diagonal interactions, while the task-relevant update directions are confined to a low-dimensional subspace. Motivated by this mismatch, we propose GIST (Gradient Isometric Subspace Transformation), a simple yet principled alternative that replaces axis-aligned scaling with robust subspace alignment. GIST recovers a task-specific subspace from validation gradients via spectral filtering (SVD), projects training gradients into this coupled subspace, and scores examples by their alignment with target directions.Extensive experiments have demonstrated that GIST matches or outperforms the state-of-the-art baseline with only 0.29% of the storage and 25% of the computational time under the same selection budget.

arXiv:2602.18745v1 Announce Type: cross Abstract: Multimodal geometry reasoning requires models to jointly understand visual diagrams and perform structured symbolic inference, yet current vision--language models struggle with complex geometric constructions due to limited training data and weak visual--symbolic alignment. We propose a pipeline for synthesizing complex multimodal geometry problems from scratch and construct a dataset named \textbf{GeoCode}, which decouples problem generation into symbolic seed construction, grounded instantiation with verification, and code-based diagram rendering, ensuring consistency across structure, text, reasoning, and images. Leveraging the plotting code provided in GeoCode, we further introduce code prediction as an explicit alignment objective, transforming visual understanding into a supervised structured prediction task. GeoCode exhibits substantially higher structural complexity and reasoning difficulty than existing benchmarks, while maintaining mathematical correctness through multi-stage validation. Extensive experiments show that models trained on GeoCode achieve consistent improvements on multiple geometry benchmarks, demonstrating both the effectiveness of the dataset and the proposed alignment strategy. The code will be available at https://github.com/would1920/GeoCode.

arXiv:2602.18759v1 Announce Type: cross Abstract: Learning from implicit feedback is a fundamental problem in modern recommender systems, where only positive interactions are observed and explicit negative signals are unavailable. In such settings, negative sampling plays a critical role in model training by constructing negative items that enable effective preference learning and ranking optimization. However, designing reliable negative sampling strategies remains challenging, as they must simultaneously ensure realness, hardness, and interpretability. To this end, we propose \textbf{ICPNS (In-Community Popularity Negative Sampling)}, a novel framework that leverages user community structure to identify reliable and informative negative samples. Our approach is grounded in the insight that item exposure is driven by latent user communities. By identifying these communities and utilizing in-community popularity, ICPNS effectively approximates the probability of item exposure. Consequently, items that are popular within a user's community but remain unclicked are identified as more reliable true negatives. Extensive experiments on four benchmark datasets demonstrate that ICPNS yields consistent improvements on graph-based recommenders and competitive performance on MF-based models, outperforming representative negative sampling strategies under a unified evaluation protocol.