CGTime：通过计算与语言解耦实现多变量时间序列对齐
原标题：Decoupling Perception from Description: Computation-Grounded Representation Alignment between Multivariate Time Series and Language
AI 摘要
arXiv 上发布了一篇关于多变量时间序列与语言对齐的研究论文。研究指出，传统方法依赖 LLM 描述时间序列，导致标签质量受限于模型感知能力，且多变量模式难以处理。为此，研究者提出解耦感知与描述的方法，用确定性代码计算统计量，再由 LLM 进行表达，构建了 4B 参数的 CGTime 模型。该模型在多变量理解任务上优于 GPT-4o-mini 和 GPT-5.4-nano 等更大模型，并在生成描述中更准确地陈述可验证的数值事实。
正文节选
Computer Science > Machine Learning Title:Decoupling Perception from Description: Computation-Grounded Representation Alignment between Multivariate Time Series and Language View PDF Abstract:Training multimodal models to align time series with language runs into a self-supervision trap. The usual recipe asks an LLM to read a series and write a description, so label quality is capped by the perceptual skill the model is supposed to learn. The data can never teach more than the labele