Research map / Time series / Foundation models for time series
CoGenCast: A Coupled Autoregressive–Flow Generative Framework for Time Series Forecasting
Mingyue Cheng, Yaguo Liu, Daoyu Wang, Xiaoyu Tao, Qi Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the LLM encoder (bidirectional) to the decoder-only variant (ablation) | CoGenCast · Energy, Environ., Exchange, Wind | MSE: Energy 0.953 -> 0.293; Wind 2.010 -> 0.532 | Improved |
| add decoder with flow matching on top of encoder-only variant (ablation) | CoGenCast · Energy | MSE: 0.331 -> 0.293 | Improved |
| add autoregressive mechanism (ablation) | CoGenCast · Wind | MSE: 0.673 -> 0.532 | Improved |
| add flow-matching mechanism (ablation) | CoGenCast · Wind | MSE: 0.611 -> 0.532 | Improved |
| couple both AR and flow versus neither (ablation) | CoGenCast · ETTh1 | MSE: 0.372 -> 0.328 | Improved |
| use average-velocity one-step generator instead of vanilla flow matching (ablation) | CoGenCast · Exchange | MSE: 0.051 -> 0.031 | Improved |
| use average-velocity generator instead of vanilla diffusion (ablation) | CoGenCast · Energy | MSE: 0.325 -> 0.293 | Improved |
| increase function evaluations from 1 to 3 (ablation) | CoGenCast · Energy, Environ., ETTh1, Solar | MSE | No clear effect |
| use linear noise scheduler instead of cosine (ablation) | CoGenCast · Energy, Environ., ETTh1 | MSE | Improved |
| use pretrained Qwen3-0.6B instead of vanilla Transformer backbone (ablation) | CoGenCast · Wind | MSE: 0.587 -> 0.532 | Improved |
| scale backbone from Qwen3-0.6B to Qwen3-4B (ablation) | CoGenCast · Energy, Environ., Exchange, Wind | MSE: Wind 0.532 -> 0.525; Energy 0.293 -> 0.290 | No clear effect |
| swap Qwen3-0.6B for Llama3.2-1B backbone (ablation) | CoGenCast · ETTh2, ETTm1, ETTm2, Exchange | MSE: ETTm1 0.224 -> 0.221 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- TimeDART — TimeDART: A Diffusion Autoregressive Transformer for Self-Supervised Time Series Representation (ICML 2025) · better on 1 of 2 shared results
- iTransformer — iTransformer: Inverted Transformers Are Effective for Time Series Forecasting (ICLR 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- AutoTimes: Autoregressive Time Series Forecasters via Large Language Models (NeurIPS 2024 · similarity 0.89)
- TS-RAG: Retrieval-Augmented Generation based Time Series Foundation Models are Stronger Zero-Shot Forecaster (NeurIPS 2025 · similarity 0.89)
- TEMPO: Prompt-based Generative Pre-trained Transformer for Time Series Forecasting (ICLR 2024 · similarity 0.89)
- TARFVAE: Efficient One-Step Generative Time Series Forecasting via TARFLOW based VAE (NeurIPS 2025 · similarity 0.89)
- Time Series Generation Under Data Scarcity: A Unified Generative Modeling Approach (NeurIPS 2025 · similarity 0.89)
- Time-LLM: Time Series Forecasting by Reprogramming Large Language Models (ICLR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
