An Introduction to Deep Learning

blog.algorithmia.com

110 points·felix_thursday··5 comments

I found this panel video to be quite good discussion of DL. https://www.youtube.com/watch?v=furfdqtdAvc

> Ford predicts self-driving cars will see widespread use within the next five years. The U.S. government has gotten on board by issuing safety guidelines.

I would love to see what those guidelines entail. For example, how thoroughly do systems need to be retested after major and minor updates? And how are we going to enforce those safety guidelines, given the scandals with emission guidelines (which is a much less complicated territory) we've seen lately (e.g. VW).

Emission guidelines are less complicated, but I suspect more gameable. Also if systems break the guidelines, the result is likely to be detected by end users, to put it lightly.

Yes, but consider the dilemma: recalling or grounding hundreds of thousands of vehicles with a bug, fixing the bug and testing it for thousands of driving-hours (this will take weeks), and then patching those vehicles. Versus: making a quick fix and quietly uploading it.

Therefore, one thing I think should happen is that governments should forbid companies to upload fixes to cars directly. Instead, the uploading should be done by a government-controlled agency, and it should check that those uploads are at least X weeks/months old.

What I really want to see is the Linux for self driving cars. Ubiquitous, open, verifiable, contributed from multiple orgs/experts kind of system.

I really want udacity self driving car code base to be solid enough that I can convert my own car to self driving. Someday

Reply on news.ycombinator.com