On Thursday, September 17, 2020, at 7:56 PM, Matt Mahoney wrote:
> That's how we evaluate lossy compression. It's wrong, but close enough. There
> is no getting around subjective evaluation.
But a semantic evaluator would improve lossless text compression. If the
predictor predicts cats 70% dogs 30%, and dogs is the correct answer, then it'd
get a combined arithmetic space of say ex. 80%, not 70%. We do that for each
prediction. Do you have any tests on such a evaluator/compressor?
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/T5031dc71532ea161-M7b1e5d83edcf3d65ba5dfd62
Delivery options: https://agi.topicbox.com/groups/agi/subscription
