yjjuan/eSNN_cplusplus

Evolving spiking neural network in C++

Exploring the basic properties of eSNN in C++

Anomaly detection in time series w/ evolving spiking neural network

My works:

• Apply OeSNN to data stream of two consecutive bearing
• Explore how different hyperparameters of SNN affect performance of SNN for anomaly detection. (in c++ jupyter notebook)
  • how many hyperparameters are there in eSNN?
• Embed the SNN into running app: detection the drift of my velocity. Android sensor

Milestones:

• 20200203: I have to turn window-based ReadDirectory into Linux-based one
• 20200214: Precision 0.00438116 Recall 1 fMeasure 0.0087241. Why precision is so low?. Have to check the predictions w/ cout
• 20200228: how to feed java arrayList to cpp function?
• 20200314: start to explore eSNN for drift detection.
• 20200328: eSNN is not effective to locate the drift. Found a good substitute: ADWIN.

Notes:

• syn w/ jupyter7788
• To compile w/ Cmake,
  mkdir build
  cd build
  cmake ..
  make
• compile the main.cpp: g++ test_main2.cpp eSNN.cpp -o test_main2 -std=c++14

Sources

• SNN in Pytorch Link; Gitlab
• Datasets for concept drift; documentation
• Adapted from Piotr Maciag's github

Reference:

• Easy Outlier Detection in Data Streams
• Medium post: Detecting real-time and unsupervised anomalies in streaming data: a starting point
• Liu, B., Qi, Y. & Chen, K.-J. Sequential online prediction in the presence of outliers and change points: an instant temporal structure learning approach.(2020)
• Introduction to online learning @ creme
• Albert Bifet's work: Github, book;free book
• Deployment of online model
• Stream Learning in Energy IoT Systems