Group items tagged

Following up on our coffee-time discussion, here's an Evolutionary Algorithm where you are the fitness function, and evolution is guided by your subjective artistic sense. Start from scratch, or pick an existing image in the database, and start evolving. At every generation, you are presented with the individuals/images in the population. Pick the ones you like. Those will be the parents from which the next generation will be bred. Repeat, repeat... where do you get to? If you want to learn more about the science behind this, check the tutorial below by Kenneth Stanley, who is also this site's supervisor: http://dx.doi.org/10.1145/1830761.1830920

A paper about evolutionary algorithms. Could be something for Dario or Christos. Also has some space things in it.

This has been around for over a year. The current trend in deep learning is "deeper is better". But a consequence of this is that for a given network depth, we can only feasibly evaluate a tiny fraction of the "search space" of NN architectures. The current approach to choosing a network architecture is to iteratively add more layers/units and keeping the architecture which gives an increase in the accuracy on some held-out data set i.e. we have the following information: {NN, accuracy}. Clearly, this process can be automated by using the accuracy as a 'signal' to a learning algorithm. The novelty in this work is they use reinforcement learning with a recurrent neural network controller which is trained by a policy gradient - a gradient-based method. Previously, evolutionary algorithms would typically be used. In summary, yes, the results are impressive - BUT this was only possible because they had access to Google's resources. An evolutionary approach would probably end up with the same architecture - it would just take longer. This is part of a broader research area in deep learning called 'meta-learning' which seeks to automate all aspects of neural network training.

Btw that techxplore article was cringing to read - if interested read this article instead: https://research.googleblog.com/2017/05/using-machine-learning-to-explore.html

We made it to Slashdot's frontpage !!! :)

Congratulations, gentlemen!

Evolutionary algorithms were the fitness function is assigned according to the users like (aesthetic touch)

"When a control laser is fired at the crystal, a complex quantum-level reaction turns it the opaque crystal transparent. A second light source is beamed into the crystal before the control laser is shut off, returning the crystal to its opaque state. This leaves the light trapped inside the crystal, and the opacity of the crystal keeps the light trapped inside from bouncing around, effectively bringing light to a full stop." is the simple explanation, but I am not sure how this is actually possible with the current laws of physics

There are two ways to make slow light: material slow light and structural slow light, where you either change the material or the structural properties of your system. Here they used EIT to make material slow light, by inducing transparency inside an otherwise opaque material. As you change the absorption properties of a material you also change its dispersion properties, the so-called Kramers-Kronig relations. A rapid positive change in the dispersion properties of a material will give rise to slow light. To effectively stop light they switched off the control beam, bringing back the opaque state. Another control beam is then used to retrieve the probe pulse that was 'frozen' inside the medium. Light will be halted according to the population lifetime on the energy level (~ 100s). They used an evolutionary algorithm to find an optimal pulse preparation sequence to reach close to the maximum possible storage duration of 100s. Interesting paper!

So it is not real storage then in a sense, as you are stimulating an excitation population which retains the phase information of your original pulse? Still it is amazing that they could store this up to 100s and retrieve it with a probe pulse, but light has never been halted.

Robot self-adapts its gait when limbs are damaged.

"A list of the many different animals, plants, microbes, natural phenomena and supernatural activities that can be spotted in the wild lands of the metaphor-based computation literature"