RAID: Semantic Graph Diffusion for True Cold-Start and Cross-Lingual Forecasting

A new framework called RAID enables time-series forecasting models to make predictions for items with no historical data, a scenario known as true cold-start, according to research posted on the arXiv preprint server [1][2]. The framework, detailed in a paper submitted on June 15, 2026, replaces traditional history-based correlation learning with a method that uses metadata-driven semantic retrieval and graph-conditioned diffusion [1][2]. RAID maps textual metadata into a shared semantic space using a frozen multilingual embedding model and constructs an inductive retrieval graph that extends to previously unseen items [2]. It first forms a base forecast by aggregating information from semantically related neighbors, then refines this forecast with a gated diffusion module to model residual uncertainty [2]. Under a strict true cold-start protocol, RAID outperforms strong foundation models and competitive baselines on both forecasting accuracy and prediction interval coverage, while reducing inference latency by an order of magnitude through non-autoregressive decoding [2]. The shared semantic space also enables zero-shot cross-lingual transfer, allowing a model trained on English descriptions to generalize to items described in other languages without direct supervision [2]. The paper appears on arXiv, an open-access repository of electronic preprints that began on August 14, 1991, and now receives about 24,000 submissions per month as of November 2024 [6]. The research was shared through arXivLabs, a framework launched by arXiv to enable community collaborations that add features to the platform [4]. arXiv Executive Director Eleonora Presani said at the time of the framework's launch that "members of our community want to contribute tools that enhance the arXiv experience, and we value that kind of community engagement" [4]. The arXivLabs tabs highlight experimental tools developed by collaborators and are updated as new tools become available [4]. The repository passed the half-million-article milestone in October 2008, surpassed one million by the end of 2014, and reached two million by the end of 2021 [6].