Artificial Intelligence applied to art collections

Artificial Intelligence applied to art collections can provide unsuspected insights to both art amateurs and experts.

Let’s explore how this can be accomplished both technically and commercially.

Technical approach

1.A neural network that trains to learn the complete works of a painter

A CNN could be trained to classify the paintings of Eugène Delacroix, for example, in a few steps:
  • -     Run a CNN such as described in the “Getting Your Neurons to Work” chapter in Artificial Intelligence by Example: kernels, ReLU activations, pooling, flattening, dense layers, a cross-entropy loss function an Adam optimizer and more.
  • -         Once the architecture has been designed the training dataset will contain Eugène Delacroix’s paintings, some data augmentation with noise to avoid overfitting and some bad images to avoid overfitting
  • -         The trained model can then classify Eugène Delacroix’s paintings among any other image
  • -         An extended version could identify any painter’s work in a set of images

2.Comparative training

A sequence of paintings of various painters could be introduced using a historical, chronological dataset using the following steps:

  • -          Run RNN such as an LSTM as described in “Improve the Emotional Deficiencies of Chatbots,” Artificial Intelligence by Example
  • -          The model will then be able to locate a painting in a historical sequence

3.Conceptual Representation Learning meta-models(CRL-MM)

As explained in Artificial Intelligence by Example, mathematics alone cannot solve all AI problems.
In this case, we can take text descriptions and create concepts using:
  • -          Eugène Delacroix’s journal for example
  • -          A book on Eugène Delacroix or several depending on the resources available
  • -          A full-text dataset of texts describing a painter’s work, life and views
  • -          a link between the text and the paintings can be established

Once this dataset has been created:
  • -          Create a mental PCA model as described in Artificial Intelligence by Example. This representation will contain concepts with texts and images.
  • -          Use RNN’s to automatically generate text based on the texts trained by the CRL-MM

4. Deep Learning a life

Another CNN and LSTM could be trained to detect:
  • -          The difference in the style of a painter’s life. For example, for Eugène Delacroix, the way colors were used before he visited Algeria and after.
  • -          The difference in colors used in various paintings
  • -          The way forms were drawn in various circumstances
  • -          Finding influences automatically in a full collection of all painters

5.Image generation

Once the steps mentioned above are complete, it is possible to :
  • -          Take a picture with a smartphone
  • -          Send it to a cloud server
  • -          Convert that picture into a Delacroix (or other) style painting

This can be done with selfies, landscapes or any type of image.

6.Text generation

A Cognitive chatbot can be designed as explained in Artificial Intelligence by Example, to speak as a painter would. Once trained through the steps mentioned above, a “Delacroix” (or any other artist) chatbot could be designed that would speak as Delacroix on his works and art in general.

7.Cloud service and smartphone app

The functions mentioned above can be developed with present-day technology and hosted on a cloud server to be used from any client machine including a smartphone app.


This two-year(estimate) project follows a standard start-up roadmap for a team who wishes to pursue this avenue.
1.An art SME(Subject Matter Expert) and a backup SME
2.An AI Ph.D. student and developer and a backup developer
3.An art Ph.D. student and a backup
4.A University to host these Ph.D. students
5.A Start-up project manager: budget, market
6.A product project manager and planner
7. Public and private financing of the project


Posts les plus consultés de ce blog

How to Build Machine Learning Models in C#

Artificial Intelligence By Example