Pneuma.to
從日常生活探知宇宙規律。
某日深宵，我在煮麵間萌生念頭：把液態的水加熱，抵達沸點蒸發，過程可否模擬恆星的誕生與死亡？星系主要由氫和氧構成，分子雲塌縮後，位能轉成熱能而催生恆星，水分子（H²O） 的相變過程，已可牽涉恆星能量轉換及形態變化的想像。從分子變化到星球生成，從個體像素到整體影像，或有一套相近的法則。
To know the laws of the universe from everyday life.
Late one night, I had an idea while cooking a pot of noodles: Can the process of heating and evaporating water simulate the birth and death of stars? Galaxies are mainly composed of hydrogen and oxygen, and stars form after the collapse of a molecular cloud, converting their potential energy into thermal energy. The phase change of water molecules can thus be linked to the imagination of stellar-scale energy conversion and morphological changes. From molecular change to planet generation, from an individual pixel to an overall image, there may be a similar set of laws.
人工智慧影像處理近來備受大眾關注。不過，牽動我內心困惑的，反而是其神經網絡演算法的思考過程、研究目標及影像數據來源。近年，研究人員受細胞通訊啟發，以經典的細胞自動機為基礎，發展出神經細胞自動機，從零和一的網格生成，拓展至由零至一的浮點數生成。程式能夠學習圖片，即使破壞影像，隨後也能自我重組。這項以影像處理出發的技術，期望應用於合成生物學的領域。
Al image processing has recently attracted a lot of public attention. However, what puzzles me most is the neural network algorithm's thinking process, research goal and sources of image data. Inspired by cell communication, researchers have recently developed neural cellular automata on the foundation of classical cellular automata, expanding from a zero-and-one pixel generation to a zero-to-one floating point number generation. The program can be trained to regenerate an image after its destruction. Taking image processing as its starting point, this technology looks towards applications in synthetic biology.
《Stochastic Camera (version 0.3）》 意圖應用這些專注於組織紋理的電腦科學研究，製作地景影像，令星球看似擁有自我生長及修復的能力。作品以紅外線熱影像儀攝錄燒水的過程，以神經細胞自動機處理平面影像，生成地勢海拔圖及恆星光譜，由日常細節探究亙古不滅的恆星。
"Stochastic Camera (version 0.3) - the boiling terra "intends to apply these computer science studies on tissue texture to produce landscape images where a planet appears capable of self-growth and self-repair. Recording the process of boiling water with an infrared thermal imager and running neural cellular automata on flat images, the work generates hypsometric maps and stellar spectra to explore immortal stars from everyday details.
