DOCSEARCHbyAlgolia
Submit Docs

TSLANet (Time Series Lightweight Adaptive Network)

repository·main·Indexed 42 days ago

https://github.com/emadeldeen24/tslanet

A lightweight, adaptive convolutional model for time series representation learning. TSLANet utilizes an Adaptive Spectral Block and Interactive Convolution Block to perform classification, forecasting, and anomaly detection while addressing noise sensitivity and computational inefficiency.

Tokens
426
Snippets
0
Records
2
Agent score
4%

What's inside TSLANet

  1. Overview of TSLANet

    main

    TSLANet (Time Series Lightweight Adaptive Network) is a universal convolutional model designed for time series representation learning. It is intended to address the limitations of Transformer-based models, such as noise sensitivity, computational inefficiency, and overfitting on small datasets.

    Key components include:

    • Adaptive Spectral Block: Uses Fourier analysis to capture long-term and short-term interactions while using adaptive thresholding to mitigate noise.
    • Interactive Convolution Block: Enhances feature representation.
    • Self-Supervised Learning: Used to refine the model's ability to decode complex temporal patterns and improve robustness across different datasets.

    TSLANet is applicable to various tasks including Classification, Forecasting, and Anomaly Detection.

  2. Prepare datasets for TSLANet

    main

    Depending on your task, you will need to source or preprocess your data as follows:

    Forecasting and Anomaly Detection (AD)

    Download datasets from the TimesNet repository: https://github.com/thuml/Time-Series-Library.

    Classification

    Custom Datasets

    To use a custom dataset, you must convert it to the .pt (PyTorch) format. Follow the preprocessing steps provided in the TS-TCC repository: https://github.com/emadeldeen24/TS-TCC/tree/main/data_preprocessing.