lectures.alex.balgavy.eu

commit 958587373ae267156f5aaec7e5f0d534d18f162b
parent e90b27ffbbd6175653ba0f6de364f01a5b1d806f
Author: Alex Balgavy <alex@balgavy.eu>
Date:   Mon, 31 May 2021 09:19:51 +0200

Add notes for ML4QS

Diffstat:
M
content/_index.md
 | 
1
+
A
content/ml4qs/_index.md
 | 
6
++++++
A
content/ml4qs/introduction-basics-of-sensory-data.md
 | 
25
+++++++++++++++++++++++++

3 files changed, 32 insertions(+), 0 deletions(-)
diff --git a/content/_index.md b/content/_index.md
@@ -13,6 +13,7 @@ title = "Alex's university course notes"
 * [Coding and Cryptography](coding-and-cryptography)
 * [Binary and Malware Analysis](binary-malware-analysis-notes)
 * [Distributed Algorithms](distributed-algorithms-notes)
+* [Machine Learning for the Quantified Self](ml4qs)
 
 # BSc Computer Science (VU Amsterdam)
 ---
diff --git a/content/ml4qs/_index.md b/content/ml4qs/_index.md
@@ -0,0 +1,6 @@
++++
+title = 'Machine Learning for the Quantified Self'
++++
+
+# Machine Learning for the Quantified Self
+1. [Introduction & Basics of Sensory Data](introduction-basics-of-sensory-data)
diff --git a/content/ml4qs/introduction-basics-of-sensory-data.md b/content/ml4qs/introduction-basics-of-sensory-data.md
@@ -0,0 +1,25 @@
++++
+title = 'Introduction & Basics of Sensory Data'
++++
+# Introduction & Basics of Sensory Data
+In this course, use machine learning with self/sensory data.
+
+"Quantified self": self-tracking of biological, physical, behavioral, environmental info. Driven by a goal of individual, they want to do something with the collected info.
+
+Why? Health, better work performance...self-healing, self-discipline, self-design, self-association, self-entertainment.
+
+Quantified self is different because sensory noise, missing measurements. It's temporal data and there's interaction with user. Use multiple datasets to learn.
+
+Terminology:
+- measurement: one value for one attribute at one time point
+- time series: measurements in temporal order
+- supervised learning: inferring function from set of labelled training data
+- unsupervised learning: no target label, goal is to describe associations and patterns among attribute
+- reinforcement learning: find optimal actions in given situation to maximize numerical reward later in time
+
+## Sensory data
+Transforming raw data: combine tables by selecting step size Δt considered in data, start at earliest time point. Combine values of measurements within each interval [t, t+Δt)
+
+Machine learning tasks:
+- classification: predicting label (e.g. activity) based on sensors
+- regression: predicting e.g. heart rate based on other sensory values and activity