[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-flowstate-enables-forecasting-at-any-sampling-rate-without-retraining":10},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":22,"tags":34,"sources":39,"feedback":43,"feedback_at":22,"cost_usd":43,"total_tokens":43},1321,"flowstate-enables-forecasting-at-any-sampling-rate-without-retraining","FlowState enables forecasting at any sampling rate without retraining","A new time‑series model combines a state‑space encoder with a functional decoder to handle variable temporal resolutions while staying small.","FlowState lets users forecast data sampled at any interval without re‑training the model.\n\nThe paper introduces a time‑series foundation model that pairs a state‑space model encoder with a functional‑basis decoder. This combination yields continuous‑time representations, so the model can adjust to different input sampling rates and forecast horizons on the fly. The authors also describe a pre‑training scheme that speeds up convergence and improves robustness. In benchmark tests on GIFT‑Eval, FlowState—one of the smallest models in the category—outperformed larger transformer‑based TSFMs and handled unseen sampling rates better.\n\nWhy it matters: Most existing TSFMs require a fixed data cadence and often need retraining when the sampling frequency changes, which adds operational overhead. FlowState’s equivariance cuts that cost and broadens applicability to domains like IoT sensor streams or irregular financial data where resolution varies. Its efficiency also challenges the notion that bigger models are always needed for state‑of‑the‑art performance.\n\nThe takeaway is simple: a compact model that works across time scales could make continuous‑time forecasting more practical, and its open‑source release will let the community test whether the approach scales beyond the benchmark set.","[\"time-series\",\"state-space\",\"forecasting\",\"arxiv\"]","2026-06-16T04:00:00.000Z","2026-06-17T03:49:57.893Z","2026-06-17T03:50:00.720Z","published",null,[24,30],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"Add a clear concluding paragraph summarizing the key takeaway and why readers should track this development.","resolved",{"id":31,"reviewer":26,"round":32,"reason":33,"status":29},"editor-r2",2,"Add a distinct concluding paragraph that cleanly summarises the main finding, its significance and why readers should keep an eye on FlowState's progress.",[35,36,37,38],"time-series","state-space","forecasting","arxiv",[40],{"name":41,"url":42},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.05287",0]