"Similar to Neuro-Science, where reverse engineering methods like fMRI reveal structure in brain activity, we demonstrated how to describe the agent’s policy with simple logic rules by processing the network’s neural activity. This is important since often humans can understand the optimal policy and therefore understand what are the agent’s weaknesses. The ability to understand the hierarchical structure of the policy can help in distilling it into a simpler architecture. Moreover, we can direct learning resources to clusters with inferior performance by prioritized sampling"
On Thu, Mar 31, 2016 at 8:10 AM, Peter Kollarik <peter.kolla...@gmail.com> wrote: > this is also interesting, to visualize "how the NN thinks" > > > http://blog.acolyer.org/2016/03/02/graying-the-black-box-understanding-dqns/ > > On Wed, Mar 30, 2016 at 10:38 PM, Ben <ben_computer...@hemio.de> wrote: > >> It would be very interesting to see what these go playing neural networks >> dream about [1]. Admittedly it does not explain any specific moves the AI >> does - but it might show some interesting patterns that are encoded in the >> NN and might even give some insight into "how the NN thinks". >> >> Put differently: select a single neuron and find a board pattern such >> that the excitation of this neuron is maximal. With some luck you might be >> able to give meaning to this individual neuron or to single layers of the >> network (like how the first layers in pattern recognition basically detect >> edges). >> >> >> ~ Ben >> >> [1] >> http://googleresearch.blogspot.de/2015/06/inceptionism-going-deeper-into-neural.html >> >> >> Am 30.03.2016 22:23, schrieb Jim O'Flaherty: >> >>> I agree, "cannot" is too strong. But, values close enough to >>> "extremely difficult as to be unlikely" is why I used it. >>> >>> On Mar 30, 2016 11:12 AM, "Robert Jasiek" <jas...@snafu.de> wrote: >>> >>> On 30.03.2016 16:58, Jim O'Flaherty wrote: >>>> >>>> My own study says that we cannot top down include "English >>>>> explanations" of >>>>> how the ANNs (Artificial Neural Networks, of which DCNN is just >>>>> one type) >>>>> arrive a conclusions. >>>>> >>>> >>>> "cannot" is a strong word. I would use it only if it were proven >>>> mathematically. >>>> >>>> -- >>>> robert jasiek >>>> _______________________________________________ >>>> Computer-go mailing list >>>> Computer-go@computer-go.org >>>> http://computer-go.org/mailman/listinfo/computer-go >>>> >>> >>> _______________________________________________ >>> Computer-go mailing list >>> Computer-go@computer-go.org >>> http://computer-go.org/mailman/listinfo/computer-go >>> >> _______________________________________________ >> Computer-go mailing list >> Computer-go@computer-go.org >> http://computer-go.org/mailman/listinfo/computer-go >> > >
_______________________________________________ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
