Positive Unlabeled Learning for Tiny ML

Explore positive unlabeled learning for tiny machine learning applications in this tinyML Talks webcast featuring Kristen Jaskie from Arizona State University's SenSIP Center. Discover how PU learning algorithms can create effective models with low power and memory requirements, ideal for embedded systems. Learn about potential applications in wearable health monitoring, early disease detection, machine monitoring, and wildlife tracking. Understand key concepts such as the SCAR assumption, 2-step approach, and weighted approach. Dive into logistic regression and its modifications for PU learning, including the Modified Logistic Regression (MLR) algorithm. See practical examples using the MNIST dataset to differentiate handwritten digits.

Intro
Summit Sponsors
The Classification Problem
The Positive Unlabeled Problem
Wearable Sensors for Health Monitor
Early Novel Disease Detection
Machine Monitoring
Wildlife Monitoring
Assumption 2: The SCAR Assumption
The 2-Step Approach
The Weighted Approach
Logistic Regression - Refresher
The MLR is a Non-Traditional Classifieds
A Modified Logistic Regression Non-Traditional PU Classifier
A Non-Traditional Classifier
Intuition Digression
MLR Algorithm
A Modified Logistic Regression - It Works
MNIST: 60,000 Handwritten Digits
MNIST Dataset 3s vs 5s