SOTA (State of the Art) in Time Series

SOTA (State of the Art) in Time Series

Time Series Classification

Irregularly Sampled

• GRU-D
• IP-NETS, Interpolation-Prediction Networks, for Irregularly Sampled Time Series
• Transformer
• m-TAND, Multi-Time Attention Networks, for Irregularly Sampled Time Series

Regularly Sampled

• MALSTM-FCN, Multivariate LSTM-Fully Convolutional network
• FCN-SNLST (FCN-LS2T)
• GP-based, Gaussian Process (GP-LSTM, GP-GRU, GP-Conv1D)

References of Models

• MALSTM-FCN, Multivariate LSTM-FCNs for Time Series Classification, houshd/MLSTM-FCN: Multivariate LSTM Fully Convolutional Networks for Time Series Classification
• FCN-SNLST, Seq2Tens: An Efficient Representation of Sequences by Low-Rank Tensor Projections tgcsaba/seq2tens: Seq2Tens: An efficient representation of sequences by low-rank tensor projections
• GP-based, Bayesian Learning from Sequential Data using Gaussian Processes with Signature Covariances, tgcsaba/GPSig: Bayesian Learning from Sequential Data using Gaussian Processes with Signature Covariances

Time Series Forecasting

Univariate

• NLinear
• FiLM. Frequency improved Legendre Memory Model
• SCINet

Multivariate

• DLinear
• Query Selector
• Yformer
• Informer
• NLinear
• FiLM
• SCINet

References of Model

• FiLM FiLM: Frequency improved Legendre Memory Model for Long-term Time Series Forecasting tianzhou2011/FiLM
• SCINet SCINet: Time Series Modeling and Forecasting with Sample Convolution and Interaction | OpenReview cure-lab/SCINet: The GitHub repository for the paper: “Time Series is a Special Sequence: Forecasting with Sample Convolution and Interaction“. (NeurIPS 2022)
• LTSF-Linear (DLinear, NLinear) Are Transformers Effective for Time Series Forecasting? cure-lab/LTSF-Linear: This is the official implementation for AAAI-23 paper "Are Transformers Effective for Time Series Forecasting?"
• FEDformer
• Autoformer
• Informer
• Yformer

Library of Timeseries Classification and Forecasting

• timeseriesAI/tsai: Time series Timeseries Deep Learning Machine Learning Pytorch fastai | State-of-the-art Deep Learning library for Time Series and Sequences in Pytorch / fastai tsai mainly for classification
  • RNN (LSTM, GRU)
  • FCN
  • LSTM-FCN, GRU-FCN, MLSTM-FCN
• microsoft/EdgeML: This repository provides code for machine learning algorithms for edge devices developed at Microsoft Research India. mainly for classification for edge devices
• darts for forecasting and anomaly detection
  • Classic (ARIMA, VARIMA, BATS, Kalman, Random Forest)
  • RNN (GRU, LSTM)
  • TCN
  • Dlinear, Nlinear
• eural Forecast mainly for forecasting
  • RNN (LSTM)
  • TCN
  • Autoformer, Informer

SOTA in other Area

• OpenScience’s Bloom (opensource, 175B params),
• Google’s T5 (opensource, 11B)
• Riffusion,
• VALL-E,
• Point-E
• ChatGPT is not all you need. A State of the Art Review of large Generative AI models