algomanic

5 ms·

1. 1.▲
   by algomanic 11y ago

   Interesting, thank you.
2. 2.▲
   by algomanic 11y ago

   My understanding is that in SAT world, random instances are typically easily, even for large numbers of variables. Hard instances only exist in a very narrow band of instance space pertaining to the #constraints/#variables ratio. And y
3. 3.▲
   by algomanic 11y ago

   Related: "On the Expressive Power of Deep Learning: A Tensor Analysis". http://arxiv.org/abs/1509.05009
4. 4.▲
   by algomanic 11y ago

   Could you explain weight matrix "spurious interactions"?
5. 5.▲
   by algomanic 11y ago

   The fact that the class of NN functions is universal is almost vacuous: the basic idea is that if you allow a "neuron" for every point in your input space then you can mimic any function you like (i.e. each neuron handles a single