README.md

Audio_GW

A repository which contains work on the data analysis for audio signals and gravitational waves signals.

Speech Classification with NNs

• plot_classif_torch.ipynb is a jupyter notebook that contains the application of the 1-D scattering transform to classify audio files with a simple logistic regression. It is used for the classification of spoken digits and various environmental sounds.

• SpeechClassification.ipynb Speech classification of the Speech Commands dataset using pytorch. We tested three models:

  1. Two dimensional CNN trained on transformed waveforms using 1-D scattering transform.
  2. Softmax regression trained on the "log scattering transform", see: https://www.kymat.io/index.html
  3. One dimensional CNN applied to raw waveforms, see: Tutorial Paper

  We also reported some possible future developments.

Datasets Description

O3a_scatt_light_strain_whitened.zip contains hdf5 files of scattered-light glitches from Virgo.

Gravity_Spy_Glitches_whitened_#.zip contain various glitches from LIGO. They were identified starting from a slightly modified Gravity Spy dataset which can be found here. The dataset was used to obtain the classification of the glitches and their time coordinate. Once this was done, it was possible to retrieve the time-signal from the GWOSC website using the gwpy package.

Creation of the Dataset

• Gravity_Spy_Analysis.ipynb the notebook used to perform the whiteneing of the time-signals and to create the Gravity_Spy_Glitches_whitened_# files.
• Read_Whitened_GW.ipynb a notebook showing how to read the Gravity_Spy_Glitches_whitened_# files.

Gravity Spy Machine Learning

• DEEP MULTI-VIEW MODELS FOR GLITCH CLASSIFICATION, Bahaadini et al., 2017
• Machine learning for Gravity Spy: Glitch classification and dataset, Bahaadini et al., 2018
• Classifying the unknown: Discovering novel gravitational-wavedetector glitches using similarity learning, Coughlin et al., 2019.
• Discriminative Dimensionality Reduction usingDeep Neural Networks for Clustering of LIGO data, Bahaadin et al., 2022
• Data quality up to the third observing run ofAdvanced LIGO: Gravity Spy glitch classifications, Glanzer et al., 2023
• Classification of raw data with pytorch:
• Kaggle dataset: Gravity spy dataset (Q-transform data), Pytorch CNN: , Pytorch Vision Transformer

Glitch Generation:

• Gengli: GAN for glitch generation, paper1 paper2. To do: build dataset

• Wavenet: a pytorch implementation: Code. To do: build proper dataset. Wavenet_Pytorch_WIP.ipynb is a notebook in which we try to perform the generation

• Denoising Diffusion Probabilistic Model: best state-of-the-art generative models, pytorch implementation: Code. To do: guess? ahah

• Scratchy CycleGAN adapted from Kaggle notebook

Time series Forecasting:

some paper:

• Temporal Fusion Transformers for Interpretable Multi-horizon Time Series Forecasting, Lim et al., (2019)
• Are Transformers Effective for Time Series Forecasting?, Zeng et al., (2022)
• Long-term Forecasting with TiDE: Time-series Dense Encoder, Kong et al., (2023)

some repos:

• PyTorch Forecasting forecasting
• Darts forecasting and anomaly detection. Some examples:

some codes:

• D,N and Vanilla Linear Forecasting Scratchy Pytorch: