Pith. sign in

REVIEW 4 cited by

TOTEM: TOkenized Time Series EMbeddings for General Time Series Analysis

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2402.16412 v2 pith:47KCW5CU submitted 2024-02-26 cs.LG

TOTEM: TOkenized Time Series EMbeddings for General Time Series Analysis

classification cs.LG
keywords timeseriesmodelsmanytotemdatadatasetsgeneralist
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

This work studies the problem of time series analysis with generalist (or foundation) models, which are models trained across many data domains. Drawing inspiration from the widespread success of large language models, we consider the simple strategy of discretely tokenizing time series data drawn from a myriad of datasets via self-supervision, then using the fixed tokenization to solve a variety of tasks across many data domains. Canonically, time series models are either trained on a single dataset or built in a task-specific manner (e.g., a forecasting-only model), where many use patches of time as inputs to the model. As such, performant generalist, discrete representation time series models explored across many tasks are of value. Our method, TOkenized Time Series EMbeddings (TOTEM), produces such generalist time series models with minimal or no fine-tuning while exhibiting strong zero-shot performance. We evaluate TOTEM extensively over nearly 500 experiments on three commonly-studied time series tasks with real-world data: imputation (17 baselines, 12 datasets), anomaly detection (19 baselines, 25 datasets), and forecasting (14 baselines, 12 datasets). We conclude that TOTEM matches or outperforms existing state-of-the-art models in both the canonical specialist setting (i.e., training one model on one domain) as well as the generalist setting (i.e., training a single model on many domains), which demonstrates the efficacy of tokenization for general time series analysis. The open-source implementation is available here: https://github.com/SaberaTalukder/TOTEM; a video summary is available here: https://www.youtube.com/watch?v=OqrCpdb6MJk.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Data-Driven Forecasting of three-Component Seismograms Using Transformer Architectures

    astro-ph.IM 2026-06 unverdicted novelty 6.0

    SeismoGPT is a transformer autoregressive model achieving median normalized cross-correlation above 0.93 when forecasting synthetic three-component seismograms up to 240 s ahead from P- and S-wave context.

  2. Data-Driven Forecasting of three-Component Seismograms Using Transformer Architectures

    astro-ph.IM 2026-06 conditional novelty 6.0

    A causal transformer trained on synthetic seismograms autoregressively continues three-component waveforms past the S-wave with median NCC of 0.93 or higher across controlled geometries.

  3. Seeing Without Eyes: 4D Human-Scene Understanding from Wearable IMUs

    cs.CV 2026-04 unverdicted novelty 6.0

    IMU-to-4D uses wearable IMU data and repurposed LLMs to predict coherent 4D human motion plus coarse scene structure, outperforming cascaded state-of-the-art pipelines in temporal stability.

  4. On Identifying Why and When Foundation Models Perform Well on Time-Series Forecasting Using Automated Explanations and Rating

    cs.LG 2025-08 conditional novelty 4.0

    On four public datasets, Gradient Boosting with hand-built features beat Chronos, Llama, and ARIMA on most accuracy metrics, while Chronos only led on financial sMAPE.