Back to feed
arXiv cs.LG
arXiv cs.LG
7/23/2026
Challenges of Explainability in Continual Learning for Time Series Forecasting

Challenges of Explainability in Continual Learning for Time Series Forecasting

Short summary

This paper investigates explainability as a tool for understanding continual learning in adaptive time series forecasting using Experience Replay. The authors study PatchMixer, PatchTST, and DLinear architectures with attention-based sampling, applying Grad-CAM and attention rollout to analyze predictive behavior on real-world piezometric data. Results show that attribution patterns evolve over time and can inform data selection and adaptation strategies in non-stationary forecasting scenarios.

  • •Explainability methods (attention rollout, Grad-CAM) applied to continual learning in time series forecasting
  • •Neural forecasters (PatchMixer, PatchTST, DLinear) tested on real-world piezometric data with regime shifts
  • •Attribution pattern evolution reveals insights for data selection and adaptation in non-stationary scenarios

Generated with AI, which can make mistakes.

#research-breakthrough
Read full article at arXiv cs.LG

Is this a good recommendation for you?

Comments

Failed to load comments. Please try again.

Explore more

Towards the Explainability of Temporal Graph Networks via Memory Backtracking and Topological Attribution
arXiv cs.LG
From Graphs to Gradients: Physics-Inspired Structural Attribution for Cyber-Physical IoT Systems and Beyond
arXiv CS.AI
Position: Explainability Research Must Prioritize Foundations over Ad-hoc Methods
arXiv cs.LG
Federated Explainable Artificial Intelligence: Roles, Architectures, Evaluation, and Open Challenges
arXiv cs.LG