Algoliterary Encounters: Difference between revisions
From Algolit
| Line 52: | Line 52: | ||
=== How a Machine Might Speak === | === How a Machine Might Speak === | ||
| − | If a neural | + | If a neural network could speak, what would it say? |
* [[We Are A Sentiment Thermometer]] | * [[We Are A Sentiment Thermometer]] | ||
Revision as of 08:37, 2 November 2017
Algoliterary Encounters
Algoliterary works
A selection of works by members of Algolit presented in other contexts before.
- Oulipo recipes
- i-could-have-written-that
- The Weekly Address, A model for a politician
- In the company of CluebotNG
Algoliterary explorations
This chapter presents part of the research of Algolit over the past year.
What the Machine Writes: a closer look at the output
Two neural networks are presented more closely, what content do they produce?
How the Machine Reads: Dissecting Neural Networks
Datasets
Working with Neural Networks includes collecting big amounts of textual data. A comparison with the collection of words of the Library of St-Gilles:
Common public datasets
Most commonly used public datasets are gathered at Amazon. We looked closely at the following two:
Algoliterary datasets
Working with literary texts allows for poetic beauty in the reading/writing of the algorithms. This is a small collection used for experiments.
From words to numbers
As machine learning is based on statistics and math, in order to process text, words need to be transformed to numbers. In the following section we present three technologies to do so.
- A Bag of Words
- A One Hot Vector
- Exploring Multidimensional Landscapes: Word Embeddings
- Word Embeddings Casestudy: Crowd embeddings
Different vizualisations of word embeddings
Inspecting the technique behind word embeddings
How a Machine Might Speak
If a neural network could speak, what would it say?
Sources
The scripts we used and a selection of texts that kept us company.