Feature-aligned N-BEATS with Sinkhorn divergence

Authors: Joonhun Lee, Myeongho Jeon, Myungjoo Kang, Kyunghyun Park

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experimental evaluations with ablation studies are provided and the corresponding results demonstrate the proposed model s forecasting and generalization capabilities.
Researcher Affiliation Academia Joonhun Lee , Myeongho Jeon & Myungjoo Kang Seoul National University {niceguy718,andyjeon,mkang}@snu.ac.kr Kyunghyun Park Nanyang Technological University kyunghyun.park@ntu.edu.sg
Pseudocode Yes Algorithm 1: Training Feature-aligned N-BEATS.
Open Source Code Yes The comprehensive software setup can be found on Git Hub3. (footnote 3: https://github.com/leejoonhun/fan-beats)
Open Datasets Yes For (i), we use financial data from the Federal Reserve Economic Data (FRED)1 and weather data from the National Centers for Environmental Information (NCEI)2. (footnote 1: https://fred.stlouisfed.org, footnote 2: https://ncei.noaa.gov) ... 4. We randomly split each domain into three sets: 70% for training, 10% for validation, and 20% for testing.
Dataset Splits Yes 4. We randomly split each domain into three sets: 70% for training, 10% for validation, and 20% for testing.
Hardware Specification Yes CPU: Intel(R) Xeon(R) Platinum 8163 CPU @ 2.50GHz. GPU: NVIDIA TITAN RTX.
Software Dependencies No The paper mentions using
Experiment Setup Yes Hyperparameter Considered Values Lookback horizon. α = 50 Forecast horizon. β = 10 Number of stacks. M = 3 Number of blocks in each stack. L = 4 Activation function. RELU Feature dimension. γ = 512 Loss function. L = s MAPE Regularizing temperature. λ {0.1, 0.3, 1, 3} Learning rate scheduling. Cyclic LR(base lr=2e-7, max lr=2e-5, mode=