A Hybrid Spiking Recurrent Neural Network on Hardware for Efficient Emotion Recognition

This code can be used as the supplemental material for the paper: "A Hybrid Spiking Recurrent Neural Network on Hardware for Efficient Emotion Recognition". (Published on IEEE AICAS, June, 2022).

Citation:

C. Zou, X. Cui, Y. Kuang, Y. Wang, and X. Wang, "A Hybrid Spiking Recurrent Neural Network on Hardware for Efficient Emotion Recognition," 2022 IEEE 4th International Conference on Artificial Intelligence Circuits and Systems (AICAS), 2022, pp. 1-4, doi: xxx.

Features:

This supplemental material gives a reproduction function of training and testing experiments of vanillaRNN, LSTM, Text-CNN and proposed spiking RNN (SRNN) in our paper. Two kinds of emotion recognition datasets with different sentence lengths (also in English and Chinese language) are considered.

File overview:

README.md - this readme file.
Cars - the workspace folder for networks on the Car dataset with time step = 30.
Movies - the workspace folder for networks on the Movie dataset with time step = 500.

Requirements

Dependencies and Libraries:

python 3.5 (https://www.python.org/ or https://www.anaconda.com/)
pytorch 0.4.1 (https://pytorch.org/)
torchtext 0.3.1, glove.6B
CPU: Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz
GPU: Tesla V100

Datasets:

Movies: Large Movie Review Dataset, preprocessing, reference
Cars: dataset, preprocessing, reference

Run the code:

for example (networks training and testing, Movies dataset):

$ cd Movies
$ python movies_main.py
$ python text_cnn.py

for example (networks training and testing, Cars dataset):

$ cd Cars
$ python cars_main.py
$ python text_cnn.py

Others

You can run the plot_sop.py and plot_sparsity.py to get illustration information.

Results

Please refer to our paper for more information.

edwardzcl / SRNN