Skip to content
— CH. 1 · ORIGINS AND DEVELOPMENT —

DeepDream

~2 min read · Ch. 1 of 5
5 sections

  • Google engineer Alexander Mordvintsev created the DeepDream software program in 2014. The project began as a deep convolutional network codenamed Inception after the film of the same name. This specific code was developed for the ImageNet Large-Scale Visual Recognition Challenge that year. Google released the public version of the program in July 2015. Before this release, similar methods had been used to synthesize visual textures by other research groups. The dreaming idea and name became popular on the internet in 2015 thanks to Google's DeepDream program.

  • The software uses a convolutional neural network to find and enhance patterns in images via algorithmic pareidolia. Once trained, the network can be run in reverse to adjust the original image slightly. A given output neuron yields a higher confidence score when the input is adjusted correctly. For example, an existing image can be altered so that it is more cat-like. The resulting enhanced image can be again input to the procedure. Applying gradient descent independently to each pixel produces images where adjacent pixels have little relation. The generated images improve greatly by including a prior or regularizer that prefers inputs with natural image statistics.

  • After Google published their techniques and made their code open-source, tools appeared on the market. Web services, mobile applications, and desktop software enabled users to transform their own photos. These tools allowed people to apply the dream-like appearance to deliberately overprocessed images. The transition from proprietary code to public access changed how the technology was used globally. Users could now create psychedelic and surreal images algorithmically without needing deep technical knowledge of the underlying system.

  • In 2021, a study published in the journal Entropy demonstrated the similarity between DeepDream and actual psychedelic experience. Researchers recorded Electroencephalography EEG of human participants during passive vision of a movie clip and its DeepDream-generated counterpart. They found that DeepDream video triggered a higher entropy in the EEG signal. This result showed a higher level of functional connectivity between brain areas. Both metrics are well-known biomarkers of actual psychedelic experience. Hallucinogens such as DMT alter the function of the serotonergic system which is present within the layers of the visual cortex.

  • DeepDream was used for Foster the People's music video for the song Doing It for the Money. In 2017, a research group out of the University of Sussex created a Hallucination Machine. This tool applied the DeepDream algorithm to a pre-recorded panoramic video. Users could explore virtual reality environments to mimic the experience of psychoactive substances. The subjective experiences induced by the machine differed significantly from control non-hallucinogenic videos. In 2022, a research group coordinated by the University of Trento measured participants' cognitive flexibility after exposure to hallucinatory-like counterparts generated by the DeepDream algorithm.

Common questions

Who created the DeepDream software program and when was it developed?

Google engineer Alexander Mordvintsev created the DeepDream software program in 2014. The project began as a deep convolutional network codenamed Inception after the film of the same name.

When did Google release the public version of the DeepDream program to users?

Google released the public version of the program in July 2015. Before this release, similar methods had been used to synthesize visual textures by other research groups.

How does the DeepDream algorithm process images to create enhanced patterns?

The software uses a convolutional neural network to find and enhance patterns in images via algorithmic pareidolia. Applying gradient descent independently to each pixel produces images where adjacent pixels have little relation until regularizers are applied.

What scientific study published in 2021 demonstrated the similarity between DeepDream and actual psychedelic experience?

A study published in the journal Entropy in 2021 demonstrated the similarity between DeepDream and actual psychedelic experience. Researchers recorded Electroencephalography EEG of human participants during passive vision of a movie clip and its DeepDream-generated counterpart.

Which music video utilized the DeepDream technology for Foster the People's song Doing It for the Money?

DeepDream was used for Foster the People's music video for the song Doing It for the Money. The software enabled users to apply the dream-like appearance to deliberately overprocessed images.

See all questions about DeepDream →

All sources

23 references cited across the entry

  1. 1webDeepDream - a code example for visualizing Neural NetworksAlexander Mordvintsev et al. — Google Research — 2015
  2. 2webInceptionism: Going Deeper into Neural NetworksAlexander Mordvintsev et al. — Google Research — 2015
  3. 3conferenceIEEE Conference on Computer Vision and Pattern Recognition, CVPR 2015, Boston, MA, USA, June 7–12, 2015Christian Szegedy et al. — IEEE Computer Society — 2015
  4. 4conferenceIEEE International Conference on Neural NetworksJ.P. Lewis — 1988
  5. 5journalA parametric texture model based on joint statistics of complex wavelet coefficientsJ Portilla et al. — 2000
  6. 6conferenceVisualizing Higher-Layer Features of a Deep NetworkDumitru. Erhan — 2009
  7. 7conferenceDeep Inside Convolutional Networks: Visualising Image Classification Models and Saliency MapsKaren Simonyan et al. — 2014
  8. 8magazineThese Google "Deep Dream" Images Are Weirdly MesmerisingDaniel Culpan — 2015-07-03
  9. 9webFear and Loathing in Las Vegas is terrifying through the eyes of a computerRich McCormick — 7 July 2015
  10. 10journalComputer Vision and Computer HallucinationsBrian Hayes — 2015
  11. 11conference2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)Aravindh Mahendran et al. — 2015
  12. 12conferenceUnderstanding Neural Networks Through Deep VisualizationJason Yosinski et al. — 2015
  13. 13journalFeature VisualizationChris Olah et al. — 2017-11-07
  14. 14webWhen Robots HallucinateAdrienne LaFrance — The Atlantic — 2015-09-03
  15. 15journalNeural Network Models for DMT-induced Visual HallucinationsChristopher Timmermann — NIH — 2020-12-12
  16. 16conferenceSynthesizing the preferred inputs for neurons in neural networks via deep generator networksAnh Nguyen et al. — 2016
  17. 17conferenceWhy are deep nets reversible: A simple theory, with implications for trainingSanjeev Arora et al. — 2016
  18. 18journalDream Formulations and Deep Neural Networks: Humanistic Themes in the Iconology of the Machine-Learned ImageEmily L. Spratt — Humboldt-Universität zu Berlin — 2017
  19. 19citationFoster The People - Doing It for the MoneyfosterthepeopleVEVO — 2017-08-11
  20. 20journalA Deep-Dream Virtual Reality Platform for Studying Altered Perceptual PhenomenologyKeisuke Suzuki — 22 November 2017
  21. 21journalIncreased Entropic Brain Dynamics during DeepDream-Induced Altered Perceptual PhenomenologyAntonino Greco et al. — July 2021
  22. 22journalThe entropic brain: a theory of conscious states informed by neuroimaging research with psychedelic drugsRobin Carhart-Harris et al. — 2014
  23. 23journalSimulated visual hallucinations in virtual reality enhance cognitive flexibilityClara Rastelli et al. — 7 March 2022