[KongchangAI]
ModelFlexCast 框架

FlexCast

A time-series foundation model framework that reformulates forecasting as variable-length masked prediction over time-series segments, disentangling the three entangled factors of autoregression (context conditioning, forecast-unit decomposition, and autoregressive rollout) and supporting multiple inference modes such as one-step forecasting, iterative refinement, and next-token generation.

Timeline (last 90 days)

Nov 16

The FlexCast framework decouples three factors in autoregressive training—variable context conditions, prediction unit decomposition, and autoregressive rollout—and proposes that sequential decoding is not performance-critical, while flexible training distributions are the true source of performance.

Unverified50%
Nov 16

FlexCast reframes the prediction task as variable-length masked prediction over time-series segments, dynamically sampling visible context and masked target segments during pretraining without relying on sequential decoding.

Unverified50%
Nov 16

In the inference stage, FlexCast supports three modes through a unified mask-predict interface: one-step prediction, confidence-based iterative refinement, and next-token-style generation.

Unverified50%
Nov 16

In FlexCast's variable-length masked prediction, each pretraining step randomly selects several contiguous segments as prediction targets and the remaining segments as visible context, with both mask length and position varying dynamically.

Unverified50%
Nov 16

FlexCast's ablation experiments show that contiguous target masking and context diversity together constitute the main source of performance improvement.

Unverified50%
Nov 16

Evaluation on BLAST and multiple zero-shot benchmarks shows that FlexCast's accuracy is comparable to or even better than strong TSFM baselines, while inference latency is significantly reduced.

Unverified50%
Nov 16

Once the model has learned flexible context-target relationships, left-to-right decoding does not bring consistent gains.

Unverified50%
Nov 16

The research code for FlexCast has been open-sourced on GitHub, with the repository address BenchCouncil/FlexCast.

Unverified50%

All Facts (8)

Unverified

The FlexCast framework decouples three factors in autoregressive training—variable context conditions, prediction unit decomposition, and autoregressive rollout—and proposes that sequential decoding is not performance-critical, while flexible training distributions are the true source of performance.

50%
Unverified

FlexCast reframes the prediction task as variable-length masked prediction over time-series segments, dynamically sampling visible context and masked target segments during pretraining without relying on sequential decoding.

50%
Unverified

In the inference stage, FlexCast supports three modes through a unified mask-predict interface: one-step prediction, confidence-based iterative refinement, and next-token-style generation.

50%
Unverified

In FlexCast's variable-length masked prediction, each pretraining step randomly selects several contiguous segments as prediction targets and the remaining segments as visible context, with both mask length and position varying dynamically.

50%
Unverified

FlexCast's ablation experiments show that contiguous target masking and context diversity together constitute the main source of performance improvement.

50%
Unverified

Evaluation on BLAST and multiple zero-shot benchmarks shows that FlexCast's accuracy is comparable to or even better than strong TSFM baselines, while inference latency is significantly reduced.

50%
Unverified

Once the model has learned flexible context-target relationships, left-to-right decoding does not bring consistent gains.

50%
Unverified

The research code for FlexCast has been open-sourced on GitHub, with the repository address BenchCouncil/FlexCast.

50%

Source Articles