thanks - I was already wondering several times what had happened to this technique that he used at the talk we looked at several times when it was first uploaded ... good that they have made it open source! are they easy to use?
the easiest way to use them is:
Google Docs > open/create a spreadsheet > Insert > Gadget > Charts > Motion Chart !! :)
You have here a tutorial describing all the steps to get it running.
Related to our discussion on crowdfunding from Friday's science coffee.
(Another sad example of how Tony Abbott's policy negatively affects the Australian science community... )
A team of astronomers have resorted to raising funds through crowdsourcing to try and save an Australian telescope involved in mapping the Milky Way. The 22-metre diameter Mopra Radio Telescope, based near Coonabarabran in western New South Wales, is slated to be shut down by the end of the year after $110-million was slashed from CSIRO in last year's federal budget.
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.
About "Papers and patents are becoming less disruptive over time" https://www.nature.com/articles/s41586-022-05543-x.
"Overall, our results deepen understanding of the evolution of knowledge and may guide career planning and science policy. To promote disruptive science and technology, scholars may be encouraged to read widely and given time to keep up with the rapidly expanding knowledge frontier. Universities may forgo the focus on quantity, and more strongly reward research quality56, and perhaps more fully subsidize year-long sabbaticals. Federal agencies may invest in the riskier and longer-term individual awards that support careers and not simply specific projects57, giving scholars the gift of time needed to step outside the fray, inoculate themselves from the publish or perish culture, and produce truly consequential work. Understanding the decline in disruptive science and technology more fully permits a much-needed rethinking of strategies for organizing the production of science and technology in the future."