DEFT Improves Expert-Guided Time-Series Forecast Editing
Summary
DEFT is a new framework that enhances expert-guided forecast editing for time-series foundation models by balancing exploitation of the model's predictive distribution with structured exploration in a decomposed trend-seasonal space.
Why it matters
Professionals relying on time-series forecasts, especially in critical domains, can use DEFT to incorporate valuable expert insights more efficiently, leading to more accurate and trustworthy predictions from foundation models.
How to implement this in your domain
- 1Evaluate DEFT's approach for integrating expert feedback into your time-series forecasting workflows.
- 2Decompose your time-series forecasts into trend and seasonal components for more structured expert interaction.
- 3Design expert feedback mechanisms that can score complete trajectories while providing component-level insights.
- 4Compare DEFT's performance against simpler best-of-N or direct optimization methods under limited expert query budgets.
Who benefits
Key takeaways
- DEFT enables efficient expert-guided editing of time-series foundation model forecasts.
- It balances exploiting model predictions with structured exploration in a decomposed space.
- Expert queries are maximized by reusing scores for trend and seasonal components.
- DEFT consistently outperforms direct search methods across various benchmarks.
Original post by Hung Le, Minh Hoang Nguyen, Manh Nguyen, Huu Hiep Nguyen, Dai Do
"arXiv:2607.19659v1 Announce Type: new Abstract: Time-series foundation models can forecast across heterogeneous domains without task-specific training, but their forecasts are fixed once produced and cannot directly incorporate task-specific expert feedback. We study expert-guide…"
View on XOriginally posted by Hung Le, Minh Hoang Nguyen, Manh Nguyen, Huu Hiep Nguyen, Dai Do on X · view source
Want to go deeper?
Turn these trends into skills with Learnijoy's hands-on AI & tech courses.
Explore coursesMore in AI Engineering & DevTools
New Adaptive Filter Improves Time-Series Prediction with Input Noise
Researchers developed the RFFBCGA algorithm, a new nonlinear adaptive filter that effectively mitigates both input and output noise in time-series prediction. This method maintains a fixed network structure while enhancing robustness across various noise scenarios.
New Algorithm Learns Local Causal Structures with Latent Variables
Researchers propose LoCaLS, a new algorithm for learning local causal structures around a target variable from observational data, even when latent variables and selection bias are present. LoCaLS achieves high accuracy with significantly less computational effort than global causal discovery methods.
New Framework Evaluates AI Robustness with Minimum-Norm Attacks
Researchers introduce a unified framework for evaluating adversarial robustness using a comprehensive pool of minimum-norm attacks and robustness-perturbation curves across multiple norms. This approach addresses limitations of fixed-epsilon evaluations, providing a more stable and controllable assessment of AI model defenses.