2d:null 2e:["$","title",null,{"children":"SM3809_A9 | John C | Observable"}] 33:T153e,html`

SM3809_A9

Neural Networks

Read

Generation with Recurrent Neural Networks by Andrej Karpathy: from fixed input and output to dynamic input and output -> outputs influenced by input and also history data, a hiden state 'memorizing' the previous steps.

// while processing each layer, something like this is called function step(input){ hiddenState = updateHiddenState(input,result,hiddenState); var result = doSth(input,hiddenState); return result; }
Long Short-Term Memory (LSTM) network: same idea, but the updateHiddenState() function is more complicated.
This hiddenState can be used to adjust the scores toward the wanted direction internally, and work with the training database. In terms of text processing, this hiddenState seems to be useful for the model to figure out the context, i.e. its current location in the text.
One thing is, those models need a crazy long time to train, seems that to obtain a good result a model needs to be trained for over 1000 iterations. And a giant database is also required.
Neural Cellular Automata: It is cool, and complex. I wonder what are the aesthetic possibilities inside it (What's its aesthetic advantage compared to normal cellular automata?).

This is a webpage that using machine learning language models to tell fables in an interactive way.

Technical: the dataset used in this project is Aesop's Fables (make sense as this is a project designed for bed time stories). The author then used a tensorflow.js model of Universal Sentence Encoder to embed the sentences of the fables into a high dimensional vector space. The 2D visualization of the vector space looks like this

The author uses Node.JS (with tensorflow.js and ml5.js) for the server side for vector analyzing, sentiment analysis and serving the content. So the whole project is entirely based on JavaScript, HTML and CSS.

For more details about how it was made, visit this post by the author.

Aesthetic: This piece has a very clean, simple, direct interface. The users input the beginning of the story via sound to text interface (babies don't know how to type) and generated story will be displayed a with simple, childish drawing illustrating the lines. The users then can add their own drawing to the canvas, and decide what happens next.
To be honest, this is not visually interesting if you saw it in a gallery. But that's because the target of the pieces is the kids. I think if I were a kid I will love it, because it creates a bridge for me to have conversations with the fables, both visually and with words.

Conceptual: I am just going to quote some words from the author (who quoted Friedrich Schiller) here:
"The arrival of my child was a great inspiration on my thought process. I understood quite fast that I wanted to build something with technology that would allow me to connect with him. As my partner and I started to read texts about child development and parenthood, I came across this quote by the poet Friedrich Schiller:
'Deeper meaning resides in the fairy tales told to me in my childhood than in the truth that is taught by life.'"

Something else

Seleted Stories by Cristóbal Valenzuela

A web text editor that runs a LSTM model while you write.

Machine Hallucination by Refik Anadol Studio

The project focuses on latent cinematic experiences derived from representations of urban memories as they are re-imagined by machine intelligence. For Artechouse’s New York location, Anadol presents a data universe of New York City in 1025 latent dimensions that he creates by deploying machine learning algorithms on over 100 million photographic memories of New York City found publicly in social networks.

Create

Leung Man Tao: "Naai Gong" or "Gang Du":
Observable page
page on my website

In Liz's:
Observable page
page on my website

`34:[] 31:{"id":"844fd5c5f9b1e236","slug":"sm3809-a9","title":"SM3809_A9","thumbnail":"c0b759b82dd2668e4eef375133ce7d7c8c7532392c6d9e50e11c94c82afa4d1e","access_level":"public","create_time":"$D2021-03-22T07:43:48.372Z","update_time":"$D2021-03-24T18:44:08.944Z","view_time":null,"first_public_time":"$D2021-03-24T18:14:44.458Z","has_unpublished_edits":null,"fork":false,"stars":0,"starred":false,"owner":{"type":"team","id":"b2e9840be7b1b607","login":"real-john-cheung","name":"John C","avatar_url":"https://avatars-next.observableusercontent.com/1d9ef6121a0bbd1a75fab479f9891fe1d78cc65f76c394585ae1a7dad0d3abf6","tier":"starter_2024"},"version":79,"publish_version":79,"role":null,"authored":false,"authors":[{"type":"individual","id":"e8e99acdbe3aebfb","login":"real-john-cheung","name":"John C","avatar_url":"https://avatars-next.observableusercontent.com/1d9ef6121a0bbd1a75fab479f9891fe1d78cc65f76c394585ae1a7dad0d3abf6","tier":null}],"members":[{"type":"individual","id":"e8e99acdbe3aebfb","login":"real-john-cheung","name":"John C","avatar_url":"https://avatars-next.observableusercontent.com/1d9ef6121a0bbd1a75fab479f9891fe1d78cc65f76c394585ae1a7dad0d3abf6","tier":null,"role":"editor"}],"comments":[],"custom_thumbnail":false,"trashed":false,"lapsed":null,"subscribed":null,"forks":0,"collection":null,"type":"notebook","body":{"title":"SM3809_A9","cells":[{"id":0,"value":"$33","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":6,"value":"html`

`","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":9,"value":"html`\n

SM3809_A9

Neural Networks

Read

Share

Create