TinyCast: Probabilistic Zero-Shot Forecasting with Computed Periodicity
- 类型：arxiv
- 标识：2608.15767
- 链接：https://arxiv.org/abs/2608.15767
- 主分类：evaluation
- 形态：method
- TLDR：We introduce TinyCast, an attention-free zero-shot forecaster that emits a predictive distribution from 146,505 parameters, on the premise that at this size the periodic structure of a context is worth computing rather than learning. A zero-parameter spectral detector supplies the dominant periods, the context is folded on their phase, and a dilated convolutional encoder and a block-autoregressive quantile decoder model the rest. It is smaller than every zero-shot entry on the GIFT-Eval board whose parameter count can be established. On probabilistic accuracy it defines the size-accuracy front
- 待LLM分类：否
- 标题中文：TinyCast：基于计算周期性的概率性零样本预测
- TLDR中文：我们提出 TinyCast，一个注意力无关的零样本预测器，仅用 146,505 个参数输出预测分布，其前提是在该规模下，上下文中的周期结构值得通过计算而非学习方式得到。一个零参数谱检测器给出主导周期，上下文按其相位进行折叠，再由一个膨胀卷积编码器和一个分块自回归分位数解码器建模其余部分。它在 GIFT-Eval 榜单上所有可确认参数量的零样本条目中体积最小；在概率准确性方面，它划定了 size-accuracy 前沿。
- 来源文件：
- /inbox/tom/_candidates/2026-08-22-agent-rag-longcontext-candidates.json
- /inbox/tom/_candidates/2026-08-23-agent-rag-longcontext-candidates.json