ON THIS PAGE

• Introduction
• Motivation
• Challenges
• Proposed Approach
• Main Contributions
• Problem Definition
• Model Architecture
• 1. Patching
• 2. Decoder-only Model
• 3. Longer Output Patches
• 4. Patch Masking
• TimesFM Model Architecture Detail
• 1. Input Layers
• 2. Stacked Transformer
• 3. Output Layers
• 4. Loss Function
• Training & Inference
• Pretraining Details
• Result
• 평가지표
• MAE (Mean Absolute Error)
• Scaled MAE
• Zero-shot Evaluation
• (a) Monash Archive
• (b) Darts
• (c) ETT (Electricity Transformer Temperature)
• Ablation Study
• (a) Model Scaling
• (b) Output Patch Length
• (c) Input Patch Length
• (d) Synthetic Data Ablation
• Conclusion