You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
One way to think of a neural network is as a hashtable where the hashing function is locality-sensitive. It memorizes training inputs & targets, and is capable of successfully querying targets for test inputs that are very close to what it has already seen.
In this light, the "intelligence" of the network comes purely from its training data. The network is sample-inefficient and only performs local generalization. The next frontier is abstraction & reasoning, which will enable extreme generalization and decent sample efficiency.
Thinking of it as interpolation also comes to my mind in that specific frame. Thinking of it as a hashtable would imply that the targets might be fixed value whereas NNets can provide continuous targets, wouldn't it ?
Clearly... BTW, considering abstraction, I was considering the fact that Embeddings like word2vec behave like linear vector space by construction but couldn't find theoretical proof of it... Do you know if there is such theoretical ground beyond empirical evidence?
And, as always, the question is if our brains are any better than locally-generalized hashtables
Even though I agree, I wonder if the next frontier hasn't been abstraction and reasoning for a long time. I did some work on that, but most related work were from the 90s, early 2000s, but it drowned because they couldn't do the "hash-table" right at the time, just didn't know it
Data is the king, but "Machine Induction" is prince...
Adversarial examples don’t seem to fit into the locality sensitive hashtag idea here
"In this light, the "intelligence" of the network comes purely from its training data." Brilliant!
Here's an example of a binary pattern game with very little training data that NNs just don't handle (and an example of human bias in pattern search) https://github.com/westurner/notebooks/blob/gh-pages/maths/binary-patterns_001.ipynb …
Don’t generative models go in this direction? Can’t the generator be seen as a “world” model and be used as some kind of simulator to sample as many samples as you like?
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
