ICML 21 - Voice2Series: Reprogramming Acoustic Models for Time Series Classification

Last update: Jan 03, 2023

Overview

Voice2Series-Reprogramming

Voice2Series: Reprogramming Acoustic Models for Time Series Classification

International Conference on Machine Learning (ICML), 2021 | Paper | Colab Demo

Environment

Tensorflow 2.2 (CUDA=10.0) and Kapre 0.2.0.

Noted: Echo to many interests from the community, we will also provide Pytorch V2S layers and frameworks around this September, incoperating the new torch audio layers. Feel free to email the authors for further collaboration.
option 1 (from yml)

conda env create -f V2S.yml

option 2 (from clean python 3.6)

pip install tensorflow-gpu==2.1.0
pip install kapre==0.2.0
pip install h5py==2.10.0

Training

This is tengible Version. Please also check the paper for actual validation details. Many Thanks!

python v2s_main.py --dataset 0 --eps 100 --mapping 3

Result

seg idx: 0 --> start: 0, end: 500
seg idx: 1 --> start: 5000, end: 5500
seg idx: 2 --> start: 10000, end: 10500
Tensor("AddV2_2:0", shape=(None, 16000, 1), dtype=float32)
--- Preparing Masking Matrix
Model: "model_1"
__________________________________________________________________________________________________
Layer (type)                    Output Shape         Param #     Connected to                     
==================================================================================================
input_1 (InputLayer)            [(None, 500, 1)]     0                                            
__________________________________________________________________________________________________
zero_padding1d (ZeroPadding1D)  (None, 16000, 1)     0           input_1[0][0]                    
__________________________________________________________________________________________________
tf_op_layer_AddV2 (TensorFlowOp [(None, 16000, 1)]   0           zero_padding1d[0][0]             
__________________________________________________________________________________________________
zero_padding1d_1 (ZeroPadding1D (None, 16000, 1)     0           input_1[0][0]                    
__________________________________________________________________________________________________
tf_op_layer_AddV2_1 (TensorFlow [(None, 16000, 1)]   0           tf_op_layer_AddV2[0][0]          
                                                                 zero_padding1d_1[0][0]           
__________________________________________________________________________________________________
zero_padding1d_2 (ZeroPadding1D (None, 16000, 1)     0           input_1[0][0]                    
__________________________________________________________________________________________________
tf_op_layer_AddV2_2 (TensorFlow [(None, 16000, 1)]   0           tf_op_layer_AddV2_1[0][0]        
                                                                 zero_padding1d_2[0][0]           
__________________________________________________________________________________________________
art_layer (ARTLayer)            (None, 16000, 1)     16000       tf_op_layer_AddV2_2[0][0]        
__________________________________________________________________________________________________
reshape_1 (Reshape)             (None, 16000)        0           art_layer[0][0]                  
__________________________________________________________________________________________________
model (Model)                   (None, 36)           1292911     reshape_1[0][0]                  
__________________________________________________________________________________________________
tf_op_layer_MatMul (TensorFlowO [(None, 6)]          0           model[1][0]                      
__________________________________________________________________________________________________
tf_op_layer_Shape (TensorFlowOp [(2,)]               0           tf_op_layer_MatMul[0][0]         
__________________________________________________________________________________________________
tf_op_layer_strided_slice (Tens [()]                 0           tf_op_layer_Shape[0][0]          
__________________________________________________________________________________________________
tf_op_layer_Reshape_2/shape (Te [(3,)]               0           tf_op_layer_strided_slice[0][0]  
__________________________________________________________________________________________________
tf_op_layer_Reshape_2 (TensorFl [(None, 2, 3)]       0           tf_op_layer_MatMul[0][0]         
                                                                 tf_op_layer_Reshape_2/shape[0][0]
__________________________________________________________________________________________________
tf_op_layer_Mean (TensorFlowOpL [(None, 2)]          0           tf_op_layer_Reshape_2[0][0]      
==================================================================================================
Total params: 1,308,911
Trainable params: 217,225
Non-trainable params: 1,091,686
__________________________________________________________________________________________________
Epoch 1/100
2021-07-19 01:43:32.690913: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2021-07-19 01:43:32.919343: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
113/113 [==============================] - 6s 50ms/step - loss: 0.0811 - accuracy: 1.0000 - val_loss: 1.5589e-04 - val_accuracy: 1.0000
Epoch 2/100
113/113 [==============================] - 5s 41ms/step - loss: 5.0098e-05 - accuracy: 1.0000 - val_loss: 1.0906e-05 - val_accuracy: 1.0000

Class Activation Mapping

python cam_v2s.py --dataset 5 --weight wNo5_map6-88-0.7662.h5 --mapping 6 --layer conv2d_1

Reference

Voice2Series: Reprogramming Acoustic Models for Time Series Classification

@InProceedings{pmlr-v139-yang21j,
  title = 	 {Voice2Series: Reprogramming Acoustic Models for Time Series Classification},
  author =       {Yang, Chao-Han Huck and Tsai, Yun-Yun and Chen, Pin-Yu},
  booktitle = 	 {Proceedings of the 38th International Conference on Machine Learning},
  pages = 	 {11808--11819},
  year = 	 {2021},
  volume = 	 {139},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {18--24 Jul},
  publisher =    {PMLR},
}

ICML 21 - Voice2Series: Reprogramming Acoustic Models for Time Series Classification

Related tags

Overview

Voice2Series-Reprogramming

Environment

Training

Class Activation Mapping

Reference

Owner

"Moshpit SGD: Communication-Efficient Decentralized Training on Heterogeneous Unreliable Devices", official implementation

Unsupervised clustering of high content screen samples

Code for the paper "Unsupervised Contrastive Learning of Sound Event Representations", ICASSP 2021.

[CVPR'21] FedDG: Federated Domain Generalization on Medical Image Segmentation via Episodic Learning in Continuous Frequency Space

Deep learning based hand gesture recognition using LSTM and MediaPipie.

Research into Forex price prediction from price history using Deep Sequence Modeling with Stacked LSTMs.

A lightweight library designed to accelerate the process of training PyTorch models by providing a minimal

BlockUnexpectedPackets - Preventing BungeeCord CPU overload due to Layer 7 DDoS attacks by scanning BungeeCord's logs

Neural-Pull: Learning Signed Distance Functions from Point Clouds by Learning to Pull Space onto Surfaces(ICML 2021)

InterfaceGAN++: Exploring the limits of InterfaceGAN

Script for getting information in discord

[CVPR2021] DoDNet: Learning to segment multi-organ and tumors from multiple partially labeled datasets

TCube generates rich and fluent narratives that describes the characteristics, trends, and anomalies of any time-series data (domain-agnostic) using the transfer learning capabilities of PLMs.

Pre-Training 3D Point Cloud Transformers with Masked Point Modeling

meProp: Sparsified Back Propagation for Accelerated Deep Learning (ICML 2017)

This project aims to segment 4 common retinal lesions from Fundus Images.

🕹️ Official Implementation of Conditional Motion In-betweening (CMIB) 🏃

A new play-and-plug method of controlling an existing generative model with conditioning attributes and their compositions.

Code for the SIGGRAPH 2021 paper "Consistent Depth of Moving Objects in Video".

Efficient Training of Audio Transformers with Patchout