Shot Forecasting

Shot forecasting, particularly few-shot forecasting, aims to accurately predict future time series data using limited training examples. Current research emphasizes developing efficient and effective model architectures, including lightweight pre-trained models and those leveraging techniques like deep sets and temporal embeddings, to handle diverse data characteristics and heterogeneous channels within multivariate time series. This focus on efficient learning is driven by the need to reduce computational costs and improve generalization across various datasets, ultimately enhancing the applicability of time series forecasting in resource-constrained environments and real-world applications. Improved methods for quantifying model uncertainty are also being explored to increase the reliability of predictions.

Papers

May 23, 2024

When predict can also explain: few-shot prediction to select better neural latents
Kabir Dabholkar, Omri Barak
Latent Variable Latent Variable Model Latent Dynamic Shot Forecasting

January 8, 2024

Tiny Time Mixers (TTMs): Fast Pre-trained Models for Enhanced Zero/Few-Shot Forecasting of Multivariate Time Series
Vijay Ekambaram, Arindam Jati, Pankaj Dayama, Sumanta Mukherjee, Nam H. Nguyen, Wesley M. Gifford, Chandra Reddy, Jayant Kalagnanam
Time Series Pre Training Multivariate Time Series State of the Art Forecasting Large Pre Trained Model Better Zero Shot Forecasting

April 7, 2022

Few-Shot Forecasting of Time-Series with Heterogeneous Channels
Lukas Brinkmeyer, Rafael Rego Drumond, Johannes Burchert, Lars Schmidt-Thieme
Time Series Time Series Forecasting Multivariate Time Series State of the Art Forecasting Shot Classification Heterogeneous Wireless Shot Forecasting

March 23, 2022

Multidimensional Belief Quantification for Label-Efficient Meta-Learning
Deep Pandey, Qi Yu
Shot Learning Shot Image Classification Shot Task Belief Distribution Task Selection Multi Label Meta Shot Forecasting

March 4, 2022

WPNAS: Neural Architecture Search by jointly using Weight Sharing and Predictor
Ke Lin, Yong A, Zhuoxin Gan, Yingying Jiang
Neural Architecture Search Supervised ImageNet Weight Sharing Deodata Predictor Shot Forecasting

Shot Forecasting

Papers

When predict can also explain: few-shot prediction to select better neural latents

Tiny Time Mixers (TTMs): Fast Pre-trained Models for Enhanced Zero/Few-Shot Forecasting of Multivariate Time Series

Few-Shot Forecasting of Time-Series with Heterogeneous Channels

Multidimensional Belief Quantification for Label-Efficient Meta-Learning

WPNAS: Neural Architecture Search by jointly using Weight Sharing and Predictor