{"paper":{"title":"MSTN: A Lightweight and Fast Model for General TimeSeries Analysis","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"The Multi-scale Temporal Network uses early aggregation of convolutional features, sequence modeling, and self-gated fusion to set new performance marks on time series tasks while staying under one million parameters.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chandresh K Maurya, Sumit S Shevtekar","submitted_at":"2025-11-25T18:09:42Z","abstract_excerpt":"Real-world time series often exhibit strong non-stationarity, complex nonlinear dynamics, and behavior expressed across multiple temporal scales, from rapid local fluctuations to slow-evolving long-range trends. However, many contemporary architectures impose rigid, fixed-scale structural priors-such as patch-based tokenization, predefined receptive fields, or frozen backbone encoders-which can over-regularize temporal dynamics and limit adaptability to abrupt high-magnitude events. To handle this, we introduce the Multi-scale Temporal Network (MSTN), a hybrid neural architecture grounded in a"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across extensive benchmarks covering imputation, long term forecasting, short term forecasting, classification, and cross-dataset generalization, MSTN achieves state-of-the-art performance, establishing new best results on 33 of 40 datasets, while remaining lightweight (~278,520 params for MSTN-BiLSTM and ~950,776 ≈ 1M for MSTN-Transformer) and suitable for low-latency inference (<1 sec, often in milliseconds).","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the specific combination of multi-scale convolutional encoder, sequence module, and self-gated fusion generalizes to new time series distributions without requiring extensive per-dataset hyperparameter retuning or suffering from benchmark overfitting.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MSTN is a lightweight hybrid model that reports new state-of-the-art results on 33 of 40 time series benchmarks for imputation, forecasting, and classification while using under one million parameters and sub-second inference.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"The Multi-scale Temporal Network uses early aggregation of convolutional features, sequence modeling, and self-gated fusion to set new performance marks on time series tasks while staying under one million parameters.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"bbdbdb5aa31f5e9821f096360fbfaf19ebb65b51c926640df7353a47d5430835"},"source":{"id":"2511.20577","kind":"arxiv","version":4},"verdict":{"id":"7cf8b128-3d74-43a7-84fe-ff7e46df4868","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T04:30:50.160183Z","strongest_claim":"Across extensive benchmarks covering imputation, long term forecasting, short term forecasting, classification, and cross-dataset generalization, MSTN achieves state-of-the-art performance, establishing new best results on 33 of 40 datasets, while remaining lightweight (~278,520 params for MSTN-BiLSTM and ~950,776 ≈ 1M for MSTN-Transformer) and suitable for low-latency inference (<1 sec, often in milliseconds).","one_line_summary":"MSTN is a lightweight hybrid model that reports new state-of-the-art results on 33 of 40 time series benchmarks for imputation, forecasting, and classification while using under one million parameters and sub-second inference.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the specific combination of multi-scale convolutional encoder, sequence module, and self-gated fusion generalizes to new time series distributions without requiring extensive per-dataset hyperparameter retuning or suffering from benchmark overfitting.","pith_extraction_headline":"The Multi-scale Temporal Network uses early aggregation of convolutional features, sequence modeling, and self-gated fusion to set new performance marks on time series tasks while staying under one million parameters."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2511.20577/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"4b3d77795e27cae11112039b67de189ef60faeb42ef4ef5af9cebc0051928d2d"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}