ºÝºÝߣ

ºÝºÝߣShare a Scribd company logo

Intro to Machine Learning - Hierarchical Temporal Memory

?Download as PPTX, PDF?
?
M
Martin Chang

This document discusses hierarchical temporal memory (HTM) and its applications for ECG heartbeat categorization and anomaly detection. HTM is an unsupervised learning approach that uses a temporal memory to predict the next value in a time series based on context to detect anomalies early without seeing the future. It provides links to libraries for HTM and a Kaggle dataset for training and testing an HTM model on ECG heartbeats.

1 of 17
Hierarchical temporal memory ECG Heartbeat Categorization
Hierarchical temporal memory ?No backprop ?Binary input, unsupervised ?fast learning
Binary input
Temporal Memory
(Realtime) Anomaly Detection ?Predict next value base on context ?No seeing the future ?Detect anomalies as early as possible
Extract Anomaly ?Predict value of next timestep ?Rredicted value vs Real value
Intro to Machine Learning - Hierarchical Temporal Memory
(basiclly) numpy for C++ ?https://github.com/QuantStack/xtensor/
HTM library ?https://github.com/numenta/nupic.core
? https://www.kaggle.com/shayanfazeli/heartbeat
load dataset
Dirrive traning set
Train the model
Test model
Intro to Machine Learning - Hierarchical Temporal Memory
Intro to Machine Learning - Hierarchical Temporal Memory
Result

More Related Content

Intro to Machine Learning - Hierarchical Temporal Memory