Attentive Explanations: Justifying Decisions and Pointing to the Evidence — mentioned by deepnotderp

Wed, 08 Feb 2017 18:30:00 GMT

I'm not sure I fully buy the "explainability" argument against deep learning since our goal is human level intelligence and human intelligence isn't easily explained either. On the other hand, we're nowhere near human level intelligence in most tasks (Go, Poker and image classificaiton non withholding) so I can understand the argument for explainability from a practical perspective. I think we're making some good progress in that direction though, and I'll list them below: 1) Attention maps in CNNs can tell us what the net is usually looking at. 2) "Attentive Explanations" use attention…

Paper and discussion history. A mention is not an endorsement.

Attentive Explanations: Justifying Decisions and Pointing to the Evidence · HackerPapers discussions

Attentive Explanations: Justifying Decisions and Pointing to the Evidence — mentioned by deepnotderp