Immersive immersion inspired by machine learning – Ars Electronica / Linz
Abysmal means bottomless; resembling an abyss deep down; unfathomable.
Perception is a process of acquisition, interpretation, selection and organization of sensory information. Perception supposes detection. In people, perception is helped by sensory organs. In the field of AI, the perception mechanism gathers in a significant way the data acquired by the sensors. Machine perception is the ability of a computer system to interpret data in a manner similar to the way humans use their senses to relate to the world around them. Inspired by the brain, deep neural networks (DNNs) are thought to learn abstract representations through their hierarchical architecture.
In-depth learning is part of a larger family of machine learning methods based on learning data representations as opposed to task-specific algorithms.
In-depth learning has emerged over the past decade and has changed dramatically and continues to change our current and future world. It is a "deep data extraction" with deep neural networks: neural networks with as many layers. What a network does is cascading simple linear transformations to represent highly non-linear functions capable of efficiently extracting the basic structures and models from the data and mapping them to a "meaningful" output. entry ". Yes, Neural Nets is a cascade of layers: These are hidden: who knows exactly what is happening! As more and more "hidden" layers are added, the network deepens and makes it able to represent any function: they are universal approximators. But you have to pay more to go further: more layers means more settings to adjust. Learning millions of parameters requires large data, otherwise the neural networks will fail. The learning / tuning process is a back-and-forth game in the number space with a well-known technique of back propagation. This game is played in the formation of networks – just like the training of a human who draws from his experience – finally, mainly from his mistakes.
One type of neuronal layer is the convolutional neuronal layer that converts a network into a convolutional neural network – a neural network with a particular ability to extract rich contextual information from image-type data, mimicking the way a human observer understands the world "seen". expressing it in terms of non-visible, non-frequent, non-human-expressible basic structures. Why and how it works better than any other machine learning technique and continues to beat even human performance, this is a hot topic and several technical proofs from optimizations, probabilities, statistics, mathematics, control theory etc. are available, but it's still a pipeline of linear transformations, nothing more …
The work mainly shows the "hidden" transformations occurring in a network: summing and multiplying, adding nonlinearities, creating common basic structures, models inside the data. It creates highly non-linear functions that associate "non-knowing" with "knowing". While our time creates quintillions of bytes of information a day, how can we make sense of this huge amount of data? Let the networks do it four ourselves. We give all the human knowledge and experience to the network to make sense of everything. As our lives become senseless, we may expect NN to learn and give us meaning.
We enter the tunnel, deeper and deeper, to understand things more deeply. . as we become less deep.
By interpreting the learning mechanism of an NN with an abstract approach, we wanted to challenge the dominant perception system of an artificial intelligence as it is practiced today, which is purely objective and reductionist.
Direction & Animation: Vide bevoid.co
Director: Yusuf Emre Kucur, Bahadır Dağdelen
Producer: Evren Erbaşol
Project Manager: Selay Karasu
Concept development and researcher:
Edaerife Seda Kucur Ergünay, PhD student, University of Bern
Selman Ergünay, PhD, EPFL
A / V Artists:
Yusuf Emre Kucur, Gödhan Doğan, Kerem Akgün, Bahadır Dağdelen
Cüneyt Korkut Keleşoglu, Taylan Türkkan
Erhan Kabakci, Barış Yalaz
Special thanks to forrender.com