Visual rotation invariance is not absolute. It is harder to recognize faces
upside down.

Compression doesn't work well for evaluating vision. First, there isn't a
visual equivalent of the Turing test. Second, the information content of
video is dominated by noise. Long term memory capacity for visual
information is 5-10 bits per second, the same as language.

On Thu, Feb 20, 2020, 12:06 AM <[email protected]> wrote:

> Share your questions here. Here's mine for now.
>
> OpenAI shows their 'net' can work for vision, text, and motor, like the
> brain does. It is therefore agnostic to the data type, like the brain is.
> But how does vision and text translate? Yes, they both have frequencies,
> similar synonyms, segmentation, reward, for nodes. But images, ignoring
> video, need robustness for rotation, scale, location. Does text have those?
> Text has only location offset ex. "the cat was not at" = "that cat I had
> was only at". I think I get it, location and stretching scale are just
> offsets. But how does vision be robust to 180 degree rotation? Explain very
> clearly, using a neural hierarchy. I know the features are relatively
> unmoved to each other, but doesn't the brain have to try rotating it
> different ways to get it in right way?
>
> @Matt, if we were to attempt those prize compressors against image
> prediction/completion, how would we change the algorithm? How would my trie
> tree be? I know less about vision, so I wish others would fill me in using
> clear examples. I seen OpenAI used a Transformer to predict the rest of an
> image, see their blog post from months back if need.
> *Artificial General Intelligence List <https://agi.topicbox.com/latest>*
> / AGI / see discussions <https://agi.topicbox.com/groups/agi> +
> participants <https://agi.topicbox.com/groups/agi/members> + delivery
> options <https://agi.topicbox.com/groups/agi/subscription> Permalink
> <https://agi.topicbox.com/groups/agi/Tc67faac3048278cf-M578c73220feb73dd3878e1c5>
>

------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/Tc67faac3048278cf-M4916b582633abac16027728e
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to