3 ms·
Has anyone followed Smolensky’s Harmonium approach? I think it is so cool that “goodness of fit” was based on “harmony”…
by dr_dshiv 5y ago
Has anyone followed Smolensky’s Harmonium approach? I think it is so cool that “goodness of fit” was based on “harmony”…
- etiam 5y agoHave a search under the alternative name Restricted Boltzmann Machine if you haven't. Several interesting developments since the 80s.
- dr_dshiv 5y agoOh yeah well aware that it was the first restricted Boltzmann Machine. It minimizes energy, but that’s just harmony with a sign change. When Smolensky Hinton and Rummelhart collaborated, they decided to call it “goodness of fit.” So neural networks still optimize for harmony by another name. I think that’s so cool. The original Smolensky paper is worth reading. Good ideas.