Difference between revisions of "Workshop:Algolit"

Revision as of 08:26, 8 October 2018

programme (provisional)

Monday

Machine Learning & datasets

• 10:00-10:20: Introduction algolit http://www.algolit.net
  • References: https://pad.constantvzw.org/p/algolit-workshop-mons
• 10:20-10:30: Introduction Etherbox https://networksofonesown.constantvzw.org
• 10:30-11:00 Introduction AI / Machine learning
• 11:00-11:15 Break
• 11:15-11:45 Rulebased model: Markov Chain
• 11:45-12:15 Supervised machine learning & labeled datasets: Naive Bayes & sentiment analysis
• 12:15-13:00 Unsupervised machine learning & common crawl: word2vec
• 13-14u: Lunch
• 14-15u Guided tour Mundaneum in relation to dataset selection
• 15-17u
  • For non-coders: Oulipo exercise on datasets
  • For coders: Run the scripts per group, depending of level of programming
  • Get the code running, play / variate on input and output (visualisations)
  • Mundaneum dataset: what to do with it?
  • Labeled - nltk corpora
  • Unlabeled using gensim/tensorflow
  • CommonCrawl
  • ImageNet:
  • https://distill.pub/2017/feature-visualization/
  • https://distill.pub/2018/building-blocks/
• Rule-based
• 17-18u: presentation & sharing

Tuesday

Text processing

• 10:00-10:30: What is in a good/crappy dataset, crawl wikipedia, nltk corpora/import twitter, gutenberg, mundaneum scans
• 10:30-11:50 Choose the type of element you want the machine to score and measure/count
• 10:30-10:50 Tokenization
• 11:10-11:30 Stemming
• 11:30-11:50 Part-of-speech
• 11:50-12:00 Break
• 12:00-13:00 algoliterary reading using the scripts of texts that match the action
• 13-14u: Lunch
• 14:00-17:00 Markov Chain: history, explanation, game, script
  • Play the game
  • Try another dataset, vary n-gram size
  • Read and understand the code, introduce start and end tokens.
  • Modify the algorithm, chain on stems or POS tags?
• 17:00-18:00 Presentation & sharing

Wednesday

From words to number / Pattern recognition - statistical techniques

• 10:00- 12:00 Counting / creating vectors
  • Transforming human readable text into numbers: information classifiers can process/calculate with.
• 10:00-10:30 frequency count
• 10:30-11:00 tf-idf
  • Normalize your numbers to be able to compare documents / paragraphs / sentences of different lengths
  • Demistifying vectors: show how they form vectors (of the same length)
• 11:00:11:15: Break
• 11:15-11:45 one-hot-vector / softmax
• 11:45-12:15 most common words (word2vec)
• 12:15-13:00: algoliterary reading using the scripts of texts that match the action
• 13-14u: Lunch
• 14:00-15:00 Naive Bayes: history, explanation, game, script
• 15:00-15:15: Break
• 15:15-16:15 Linear Regression OR Perceptron: history, explanation with videos, script
  • https://cs.stanford.edu/people/karpathy/convnetjs/demo/classify2d.html
  • http://www.arxiv-sanity.com/
  • http://colah.github.io/posts/2014-03-NN-Manifolds-Topology/
  • http://colah.github.io/
• 16:15-17:00: play the Naive Bayes game
• 17:00-18:00: Brainstorm 1 on projects, in collaboration

Donderdag

Evaluation methods

• 10:00-10:15: baseline
• 10:15-11:00: confusion matrix
• 11:00-11:15: Break
• 11:15-12:00: F-score, precision, recall
• 12:00-13:00 Recurrent Neural Networks - average loss
• 13-14u: Lunch
• 14:00-17:00: Questions & Brainstorm
• 17:00-18:00: presentation & sharing

Vrijdag

Prototypes ideas

• 10:00-13:00 presentation, brainstorm, in collaboration
• 13-14u: Lunch
• 14:00-17:00 finetuning & needs
• 17:00-18:00 final round

resources

• pad du workshop
• Algolit, site du collectif
• Annonce du workshop
• cours et notes de python

visuels

NOISE IS FULL OF WORDS

See http://t2i.cvalenzuelab.com - article about the process: http://aiweirdness.com/post/177091486527/this-ai-is-bad-at-drawing-but-will-try-anyways

Visual Dictionary - Click on top of the map to visualize the images in that region of the visual dictionary.

libraries

• Tensorflow: https://www.tensorflow.org
• GloVe: https://nlp.stanford.edu/projects/glove