On Sat, Feb 23, 2019 at 11:48 AM Linas Vepstas <[email protected]> wrote:
> > > On Fri, Feb 22, 2019 at 4:34 PM Rob Freeman <[email protected]> > wrote: > >> >> Can you summarize it in a line? >> > > There's a graph. Here's where it is and what it looks like. Here's how > neural nets factor it. Here are other ways of factoring it. > Here? Where? Somewhere in your "Neural-Net vs. Symbolic Machine Learning" pdf? This is explaining why distributed representation works better than symbolism? Why neural networks have dominated in the last 6-8 years and improved speech recognition by 20%+ etc? This might be a candidate: p.g. 34 "The key driver behind the deep-learning models is the replacement of intractable probabilistic models by those that are computationally efficient." You might be saying distributed representation has dominated these last few years because it is more computationally efficient. Is it Fig. 5, p.g. 51? I don't understand this sentence on that figure: "Due to the fact that words are combinations of word-senses, hard-clustering in this fashion is undesirable; by contrast, it seems that wordsenses could be validly hard-clustered." Doesn't that contradict? Oh, I like this, Note 10, p.g. 52.: "In physics, such an ambiguity of factorization is known as a global gauge symmetry; fixing a gauge removes the ambiguity. In natural language, the ambiguity is spontaneously broken: words have only a few senses, and are often sysnonymous, making the left and right factors sparse. For this reason, the analogy to physics is entertaining but mostly pointless." (BTW typo there for you Linas: "sysnonymous" :-) Gauge symmetry. What I mentioned earlier I see in category theory. But you are dismissing it. This might be a good point of contrast. I'm making it significant, and even central to my analysis of our problems. I don't see anything hard and fast, as an argument why distributed representation works better. Unless it is because it can represent such gauge symmetry... -R ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T581199cf280badd7-M7528245deeb1f0e4dc9cfed2 Delivery options: https://agi.topicbox.com/groups/agi/subscription
