Akten, 2016 : Machine Learning failure in algorithms, data, models and context.

JG's intentions and context

Akten has been a good source of information for an AI artistic perspective. I found in that article really basic descriptions of machine learning and deep learning that I might refer to later on. Description of the components of a machine learning system and generic understanding of what we do with machine learning.

The learning algorithms themselves are simply abstract mathematical (primarily statistical) formulations. (Akten, 2016)

[...] understand the wider implications of such data-driven systems. Without the full context, trying to understand the impact of an algorithm is like trying to understand the impact of the command ‘turn right’ — which is meaningless without knowing your current position,[...] what you are trying to achieve or avoid. (Akten, 2016)

Notes

BELLOW MIGHT BE OUTDATED

SEE: ma-dist/wiki/RN-Phases-2012011822

Algorithmes et données

The algorithm themselves are simply abstract mathematical (primarily statistical) formulations. (Akten, 2016)

L'apprentissage sera fait avec ces algorithmes. C'est-à-dire s'ils seront l'élément structurant la phase d'apprentissage a l'aide des données. C'est de ces deux éléments (algorithmes et données) que ce conçoit le modèle, c'est à dire un réseau capable de prédire selon les buts déterminés autant dans l'algorithme que dans les données fournies.

They need (training) data to give shape to how the algorithm will make predictions. (Akten, 2016)

It is only when trained that this abstraction is turned into concrete decisions and predictions.

That end result is called a model.

why is that called a model?

because we're trying to build a model of a particular system, to understand and predict its behavior.

Components of machine learning

  • training data
  • Preprocessed training
  • Architecture
  • Learning algorithm
  • Deployment

training data

factor

  • type of data

preprocessing training / feature extraction

With deep learning we won't have that manual feature training operation.

[...] motivations behind deep learning is to skip this ‘hand-crafted feature engineering’ step, and learn the features end to end).

Architecture

  • suits our data
  • Suits our end goal

learning algorithms

Everything related to learning :

  • functions
  • Optimization
  • Regularisation
  • Etc

end result of this step: a model

deployment - prediction phase

  • feed the model new data then:
  • Get prediction or decisions

Le déploiement est la phase de la prédiction. C'est le moment où l'on donne de nouvelle donnée au modèle pour voir ce qu'il en pense!

ref

Akten, 2016. medium.com/@memoakten/machine-learning-failure-in-algorithms-data-models-and-context-341d07d504db

autre article d'Akten et buts

2. (Collaborative creativity with Monte­ Carlo Tree Search and Convolutional Neural Networks (and other image classifiers), Akten, n.d.,

https://medium.com/@memoakten/collaborative-creativity-with-monte-carlo-tree-search-and-convolutional-neural-networks-and-other-69d7107385a0)

But: voir sa conception de la convolution du réseau de neurone, de l'IA et de son intérêt pour le type d'interaction artistique (d'entrée de données dans le réseau de l'IA)

Akten a un sujet intéressant : L'expression créative

feedback loop between the user and the system.

  • Temps réel et IA
  • Problématique du temps réel et ses réseau d'apprentissage profond discuté dans l'article (02:00-03:30)

L'apprentissage machine constructif (constructive machine learning) (Akten, n.d.)

Dissemination

Constructive learning workshop in Barcelona

Report abuse