• Think of a machine learning application and draw a diagram of a neural network, speculating which features the two hidden layers might detect. (see 5:31f in the 3Blue1Brown video) Bonus points if your application does not use camera/vision at all.

<aside> <img src="/icons/chat_lightgray.svg" alt="/icons/chat_lightgray.svg" width="40px" />

I am thinking about an advertisement post detector. When I search for item recommendations online, I get very worried that bloggers are charging for advertisements for what they recommend, rather than sharing real recommendations. I hope this advertisement post detector could help me find the real deal.

Hidden layer 1:

In this layer, the network examines how the post is arranged visually. It pays attention to things like the placement of images, text, and tags it adds.

It will look for the patterns and different templates of the post, and then link them to different features that will be listed in layer 2.

Hidden layer 2:

This layer is responsible for more subtle features and content in the post. Different layouts may have different features. Some advertisements like to have different emojis. Some of them may have “buy it now” or “I won't allow no one to know it” or something like that.

</aside>

• What specifically caused the end of the most recent “AI winter”? What, in your opinion, could be a plausible scenario that might lead to the next “AI winter”?

<aside> <img src="/icons/chat_lightgray.svg" alt="/icons/chat_lightgray.svg" width="40px" /> The end of “AI winter”

• The development of technology and the open source

The improvement of the Algorithms makes it more effective.

Also, different types “learning” for machine to make it able to be applied to various fields.

Open sources make all the improvement accessible to all of us, as people are pushing together.

• More data available and people’s consciousness

Leading to the next “AI winter”

• Ethical Issues Raised by Incomplete AI Laws

When inappropriate uses of AI proliferate, people may spontaneously resist its use and development.

• Resource Constraints

• Lack of Breakthroughs and Overhyped Expectations </aside>

• Alone or with a partner, come up with some small application that uses a neural network to do some classification or regression from numeric inputs. See the tutorial for some examples, or reach out to @gohai on Slack if you have questions!

<aside> <img src="/icons/chat_lightgray.svg" alt="/icons/chat_lightgray.svg" width="40px" /> link to the project

To drag objects with your hand!!!

(developed from my previous project)

Still lots of problems🤕 • Since the pose detection is from ML5, I’m wondering if I can use the value from it in the class? (like the value stored in array) • After I changed the size of the canvas, the key points no longer appeared in the right place. This made me confused about how the skeleton and key points are detected and how they could be draw on the canvas? Based on what? - try Scale

• With mousePressed , mouseDragged, and mouseReleased I can solve the overlapping issue when dragging since those are the functions installed in p5 library I think. But only with the pose the camera detected, I don’t know how to avoid moving multiple objects at the same time. …

And actually I don’t think neutral network is effectively used here. I just use it to replace the code to detect the hand posture. I’m still thinking of the way it could be effectively applied.

This picture of code is from my original project, which uses mouse to drag the object. And this part highlights how to solve the dragging issue.

</aside>