Lossy text compression, for some reason?!
summary
Applies JPEG-style lossy compression to text by converting letters to numbers and by compressing semantics with word embeddings, using a discrete cosine transform, KD-tree nearest-word reconstruction and simple Python code to invert vectors back to words. Experiment uses Genesis 1 (letters as letter-to-number 0–26 and words as word2vec 300‑dim vectors), shows examples at 2%, 50%, 91% and even 99% compression; outputs range from garbled letters to plausibly summarized, sometimes poetic phrases, suggesting possible low‑memory embedding or summarization uses.
AI-written summary · about
chapters (6)
related talks
11mMutating text toward better nonsense?!
39mExperimental Creative Writing with the Vectorized Word
41mKeynote: computational stichography! (or: connecting lines)
10mContinuous and Gradual Style Changes of Graphic Designs with Generative Model
19mSketchforme: Composing Sketched Scenes from Text Descriptions for Interactive Applications
01h 17mDeep generative models for image...
03h 22mNonsense verse with Python and machine learning
40mUnlock the Fun!