臺南市美術館

To draw a fractal within 10 secs, do this:

[[abs(((x^y)+(2_x)^(2_y))-256) for x in range(1,128)] for y in range(1,128)]

[[((x^y)-(64-x&y)+x+y) for x in range(1,128)] for y in range(1,128)]

[[(x+y)%((x&y)+8) for x in range(1,256)] for y in range(1,256)]

from math import sin [[int(sin((x-y)/42)*50)^int(sin((x+y)/42)*50) for x in range(1,512)] for y in range(1,512)]

from math import sin [[(int(sin((x)/84)*50)^int(sin((y)/84)*50)) for x in range(1,256)] for y in range(1,256)]

複數作為平面空間運算/向量（線性代數）

Node behavior/ trait

人類的能力有極限

也許世界上有 10 億人喜歡音樂，其中有 1 億人至少有一次腦中閃過這個念頭：我如果獲得和音樂家一樣的能力就好了。但為什麼他們沒有因為想要而真的變成 1 億個音樂家？因為這世界又不是心想事成；學會使用樂器有非常高的學習成本，而且就算去學也可能因為沒有天賦而效果不好。一大部分的人想到這裡，就以＂不切實際＂為由拒絕了這個剛萌芽2秒鐘的想法，甚至覺得這件事情是可笑的。

是阿，人是有極限的，不能想做到什麼就做到什麼。既可悲又合理。

機器的能力沒有極限

讓我們更一般地把樂器考慮成機器。

既然人類不擅長學如何操作機器，那就讓機器學如何讓人類操作。後者容易多了，因為我們可以隨意發明機器，而不能發明人種；機器的介面改良目前是看不到極限的，而人類操作機器的極限就擺在那。

「讓機器學如何讓人類操作」這件事有時需要人類花心力設計，但有時可以真的讓機器自己學。這是機器學習的一大價值。

怪罪機器難用吧

為什麼我不是音樂家？我不會說「因為天賦和練習不夠」。我會說「因為樂器不夠好用」。怪完機器難用之後，別忘了做一個好用的。做完就可以當音樂家了。

模擬一個族群演化出語言

The Value of Being Creative

Human brains are exceptionally able to process abstract ideas, and that is the premise to say that we human are conscious, intellectual, or emotional, distinguishing us from other kinds of objects in the world.

Therefore, to appreciate and make most use of the rare ability we’re granted with, we should put most of our efforts in the realm of concepts, like, be creative, and give those “low-level and trivial chores” to machines.

One of my resarches.

piano music generation with diffusion model.pdf

The Problem

I was wondering how to quantitatively represent the capacity of an VAE’s latent variable (or a reconstruction step of a diffusion model) in bits, making it comparable to other distributions, such as the data distribution?

Book on a Rod

Then, a paradox I saw like 8 years ago came to my mind:

“Theoretically, you can store all information of a book on a rod. To do that, encode the book’s content into a decimal string, put it behind 0. , and mark at the position of that number fraction of the rod’s length.”

• Pass on the gene
• Generalization
• Speed
• Match multiple rules simultaneously