Register   Login   About   Study   Enterprise   Share
Internet / AI Technology University (ITU/AITU)
Fast Login - available after registration







|

Top Links: >> 80. Technology >> Internet Technology Summit Program >> 7. Enterprise, Knowledge Architecture, IoT, AI and ML >> 7.7. Machine Learning, Data Fabric, IoT
Current Topic: 7.7.3. My experience of learning about ML
Sub-Topics: 7.7.3.1. A brief excursion into Machine Learning
-- Scroll to check for more content below...
You have a privilege to create a quiz (QnA) related to this subject and obtain creativity score...
Dear diary (all readers welcome) by Vincent Randal

These notes describe my effort to learn something about Machine Learning.. Very smart people with background in linear transform techniques (Fourier transform, wavelets, etc) are curious what?s going on inside neural networks. Ingrid Daubechies comments on that at 2:50 (t=170s) into this video:

https://youtu.be/vlGYETlHivQ&t=170s

Because Daubechies and Mallat comment on neural networks in their 2017 Abel prize lectures, I recommend watching them (including Condes? lecture) here:

https://abelprize.no/abel-prize-laureates/2017

Also, Ole Christensen?s lectures on Real Analysis are interesting for their coverage of Fourier and wavelet transforms.

https://youtube.com/playlist?list=PLMn2aW3wpAtOqo0g0OnHndXB1LnYBeMaX

In these notes I will write about an example using PWM (pulse width modulation) that illustrates the use of the Fourier transform in Machine Learning. As Ole Christensen says, electrical engineers are quite happy to talk about time varying ?signals? which are merely functions. We can sample an analog signal by multiplying it by a sampling function. We will use the fact that time domain multiplication is equivalent to frequency domain convolution *AND* the faster we sample the more we spread the features apart in the frequency domain.
Was it clear so far?


By the way that last point is the reason for 8x, 16x, or more oversampling found in audio CD players that were common 30 years ago. Substantial oversampling allows the signal to be recovered via low cost first-order low-pass filter implemented with just a capacitor and resistor.

[optional section]
Gratitude: It was a very brief time some 30+ years ago when I had the privilege to learn something from mathematicians, computer scientists, and engineers at the University of Texas at San Antonio. It was a brief and marvelous time for which I am grateful to many people including Michael Humphries, Larry Williams, Mary Lou Zeeman, Dmitry Gokhman, Kay Robbins, Steve Robbins, John Schmalzel, Parimal Patel, Marvin Chatkoff, Benham Kamali, and probably more whose names will be added as I remember them.

Vincent Randal
Sample
Please provide your comments.



Assignments
Suggest your assignments and quiz.

| Check Your Progress | Propose QnA | Have a question or comments for open discussion?

Have a suggestion? - shoot an email
Looking for something special? - Talk to me
Read: IT of the future: AI and Semantic Cloud Architecture | Fixing Education
Do you want to move from theory to practice and become a magician? Learn and work with us at Internet Technology University (ITU) - JavaSchool.com.

Technology that we offer and How this works: English | Spanish | Russian | French

Internet Technology University | JavaSchool.com | Copyrights © Since 1997 | All Rights Reserved
Patents: US10956676, US7032006, US7774751, US7966093, US8051026, US8863234
Including conversational semantic decision support systems (CSDS) and bringing us closer to The message from 2040
Privacy Policy