A human artificial neural network, BeatSaber, and proper Time Series Prediction – Friday Faves

Excuse the break, but I had my thesis to finish, but that is the topic for another post. In this Friday Faves, we have a solid mix of a human neural network, playing BeatSaber without VR headsets using machine learning, and a new scikit for time series learning.

The Stilwell Brain

The series Mind Field by Michael Stevens, the figure behind the very successful Youtube channel Vsauce, created a compelling series into consciousness and exploring the human mind. This series was a Youtube Premium series for some time, but has been made available to everyone recently. Episode 3 in Season 3 (yes there are seasons) explores neural networks for digit recognition. The twist? Every Neuron is a human on a football pitch. While it’s a symbolic approach instead of a machine learning approach, I feel it’s extremely interesting to see.

If you’re interested in the book mentioned by Chris Eliasmith from the University of Waterloo here called “How to build a brain”.

BeatSaber without VR

Tensorflow has been used for Pose Estimation all over the place, with incredible results. BeatSaber is an extremely popular game for VR, where you slice flying cubes with lightsabers to the beat of the music. VR can be really expensive, so why not use Tensorflow and a projector to bootstrap BeatSaber with pose estimation?

https://github.com/charliegerard/beat-pose

Scikit-Time

The SciKits are a family of Python libraries that build on the SciPy universe to build specific implementations for scientific fields. Scikit-Learn probably being the most famous. The amazing Alan Turing Institute now made Scikit Time available for Time Series prediction and modelling.

This should meet all your tagging, classification, and regression needs. Also, remember to cite the library if you use it to better recognize the need for Research Software Engineering to be recognized as a valuable endeavour.

@misc{sktime,
      author = {Markus Löning and Anthony Bagnall and Sajaysurya Ganesh and Viktor Kazakov
      and Jason Lines and Franz J. Király},
      title = {sktime: A Unified Interface for Machine Learning with Time Series},
      year = {2019},
      eprint = {arXiv:1909.07872},
}