Computational prediction of crystal properties plays a pivotal role in materials science. With the accelerated progress in machine learning, crystal property prediction has seen remarkable advancements. Nevertheless, the utilization of machine learning in this context faces several challenges. First, existing methods that utilize the smallest repeatable unit cell of a crystal often have a limited receptive field. Second, as experiments measuring crystal properties are time-consuming, labeled data is often scarce. To address these challenges, we propose a
S
elect
I
ve mu
L
ti-
V
iew representation
A
ugmentation framework (SILVA) for crystal property prediction. To go beyond limited receptive fields, we introduce the notion of a crystal supercell, which enables more comprehensive explorations of crystal structure. To fully combine insights from multi-view structures, i.e., from unit cells and supercells, we propose a multi-view representation learning (MRL) module that features a representation space that enhances the learning of representative features specific to different views. To alleviate the limited availability of labeled data, we propose a selective representation augmentation (SRA) module. Given representations of labeled training data, we carefully select nearby representations in the representation space established by the MRL module so that labels can be reused. An experimental study offers evidence that SILVA is capable of outperforming state-of-the-art methods.
Haomin Yu, Jilin Hu, K. Tolborg et al.· ACM Transactions on AI for S...· 0 citations
Time series generation is essential for data augmentation and privacy-preserving analysis across many real-world domains. Recent progress in discrete token modeling~(DTM) has demonstrated strong potential by transforming continuous sequences into discrete representations and performing generation in the latent space. However, existing DTM-based methods, particularly VQ-VAE based methods, suffer from imprecise manifold modeling, severe codebook collapse, and high generative complexity caused by entangled latent attributes and heuristic sequence decomposition. In this work, we propose PolarFormer, a novel framework for time series generation that addresses these limitations through a polar decomposition-based discrete representation. Specifically, we decompose continuous latent embeddings into radial and angular components, and model them using two independent codebooks, enabling a multi-layer spherical-shell latent topology that more faithfully captures the underlying data manifold. To stabilize representation learning, we reinterpret VQ-VAE quantization as a Mixture-of-Experts routing process and introduce a load-balancing loss to effectively mitigate codebook collapse. In the generative stage, we further propose a structurally decoupled generation strategy that models radial and angular token sequences jointly while leveraging their orthogonality to substantially reduce modeling complexity. Extensive experiments on multiple datasets demonstrate that PolarFormer consistently achieves state-of-the-art performance in time series generation, validating its effectiveness in representation fidelity and generative quality. Our source code has been made publicly available at https://github.com/decisionintelligence/PolarFormer
Jiahong Lyu, Hongfan Gao, Wangmeng Shen et al.· Proceedings of the 32nd ACM...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
