Interesting Free Course With Certificate Taught By Prof. Andrew Ng, Stanford University

[RCA]

Machine Learning

Offered by Stanford University and DeepLearning.AI. #BreakIntoAI with Machine Learning Specialization. Master fundamental AI concepts and ... Enroll for free.

www.coursera.org

I signed up for the Convolutional Neural Networks course.

Convolutional Neural Networks

Offered by DeepLearning.AI. In the fourth course of the Deep Learning Specialization, you will understand how computer vision has evolved ... Enroll for free.

www.coursera.org

 

[cpursley]

Machine Learning

Offered by Stanford University and DeepLearning.AI. #BreakIntoAI with Machine Learning Specialization. Master fundamental AI concepts and ... Enroll for free.

www.coursera.org

I signed up for the Convolutional Neural Networks course.

Convolutional Neural Networks

Offered by DeepLearning.AI. In the fourth course of the Deep Learning Specialization, you will understand how computer vision has evolved ... Enroll for free.

www.coursera.org

He's a great teacher. Very approachable.

 

[RCA]

He's a great teacher. Very approachable.

I'm already into the CNN (Convolutional Neural Networks) course. CNN, of course, is used to break photos down into components through a computationally expensive process. For example, it can be trained to identify various objects in an image, such as kitchen sinks. It does this by using, among other things, "ReLUs" - Rectified Linear Units. And guess what? These use piecewise linear functions based on the hinge function max(0, x), which is similar to one of MARS' hinge functions max(0,x-c) or max(0,x-0). In a sense, analyzing a large photo, extracting its components, recognizing what they are, and reassembling them into some description is a special case of what MARS does in analyzing, for example, a neighborhood of houses. CNN like MARS models non-linear data. CNN is better at handling the higher dimentions of image transformations but simpler in the math. MARS has more complex math - but can't (and doesn't need to) deal with the extraordinarily large dimentisons needed for image processing. They both have their similarities and dissimilarities.

Also studying the Toeplitz matrix transformations which are used to both blur and sharpen images. ...

 

[Digger88]

I'm already into the CNN (Convolutional Neural Networks) course. CNN, of course, is used to break photos down into components through a computationally expensive process. For example, it can be trained to identify various objects in an image, such as kitchen sinks. It does this by using, among other things, "ReLUs" - Rectified Linear Units. And guess what? These use piecewise linear functions based on the hinge function max(0, x), which is similar to one of MARS' hinge functions max(0,x-c) or max(0,x-0). In a sense, analyzing a large photo, extracting its components, recognizing what they are, and reassembling them into some description is a special case of what MARS does in analyzing, for example, a neighborhood of houses. CNN like MARS models non-linear data. CNN is better at handling the higher dimentions of image transformations but simpler in the math. MARS has more complex math - but can't (and doesn't need to) deal with the extraordinarily large dimentisons needed for image processing. They both have their similarities and dissimilarities.

Also studying the Toeplitz matrix transformations which are used to both blur and sharpen images. ...

same bro

lol jk I was talking to my car finance guy yesterday who is into AI (chatgpt and gemini) and building apps etc but has never used Grok (anti Elon I guess). I told him about Grok Heavy he called me back and said it has blown him away. What are you doing with all of this knowledge are you going to build something for appraisers to use or just like learn new things

 

[RCA]

same bro

lol jk I was talking to my car finance guy yesterday who is into AI (chatgpt and gemini) and building apps etc but has never used Grok (anti Elon I guess). I told him about Grok Heavy he called me back and said it has blown him away. What are you doing with all of this knowledge are you going to build something for appraisers to use or just like learn new things

Well there certainly are a lot of things to learn.

If I write an excellent R program that is nearly perfect, I could make it open source; other appraisers, if they are able, could copy it and use it for themselves, with the ability to modify it to suit their own tastes. Or I might use it to make neighborhood pricing models and try to sell those online, or both. I could just keep doing research and publication, and hope to sustain myself on my current income for at least the next 10 years before truly retiring. I'm not sure at this point.

Current residential and commercial appraisals are almost completely flawed if you are looking for accuracy in the +/- 2% range. Many will argue that it is not possible. It depends on how you define and understand accuracy in the context of the different valuation approaches. That's a discussion for another day.

In the meantime, I already know that certain aspects of the RCA approach can be improved. There are actually many things I need to do.

To utilize my RCA method or even just to use MARS regression, appraisers need to get to work and study how to use MARS, gaining experience in its application. I

In addition to the Stanford/Coursera courses, there are also many free courses on:

https://community.deeplearning.AI/top

If you have the intelligence, you should be able to start from knowing nothing in the beginning courses, and then go to Grok, ChatGPT, or Claude to ask questions about things you don't understand - until you do understand them. There is no reason you shouldn't be able to progress as far as you are able from a HS education. In fact a number of people have done it.

 

[RCA]

Working with MARS for many years and taking these Stanford/Coursera "Deep Learning" and "Machine Learning" courses, is extremely interesting, MARS is another variation of AI methods such as RNN *Recurrent Neural Networks", CNN "Convolutional Nerual Networks" and so on. In fact working with MARS, is much like training neural networks. The work flow of training, cross-validation and testing is pretty much the same. MARS gives you an explanation of the model that most people can understand, but the neural networks are just a jumble of many weight matrices, hidden matrices or functions and other functions - with so many variations. ... So, I will spend, I suppose the next six months working on these courses --- and building up my infrastructure.

I want to create for example, a program that can work on MLS photos - which I may get started with soon.

Also, I came to the realization that I just don't have time to do appraisal and software development with all of its overhead. Appraisal had to go, not enough money in it anyway.

[ Where is the economy going to be in 12 months? We might be surprised. ]

 

[RCA]

BTW: Here's the set of Stanford/Coursera courses for the "Deep Learning Specialization" certificate (using TheBrain app) :



It's doable in 5-6 months, although you can go at your own pace. Programming is in Python, and assignments are automatically graded - if you are paying the $49 monthly fee. If you don't want to receive a grade or certificate, you can audit the course for free.