03:18 < tham> Last night we mentioned about use gpu to train deep network

03:19 < tham> It would not be a trivial task to develop gpu part for mlpack

03:19 < tham> instead, I think it may be a good idea if we could let mlpack resue training results of other libraries

03:20 < tham> like caffe, opencv3 provide binding for it, but the speed is very slow

03:22 < tham> By now I am trying my best to study the implementation details of edgeboxes, selective search and googlenet

03:24 < tham> So I could provide some helps in the GSOC project

03:25 < tham> After that I think it would be a good idea to provide a method to reuse those pretrained model, what do you think?

03:25 < tham> Last night zoq mentioned about auto driving, I saw a video from youtube, it mentioned auto driving of today perform better than human

03:29 < tham> It would be very cool if all of us can let machines drive the cars for us

09:41 < zoq> tham: I think reusing trained models from other libraries isn't that complicated as long as we can replicate the structure of the model. I guess, all we need to do is to set the weights.

09:41 < zoq> tham: I guess, there are some challenges we have to master before we can actually use that machine to actually drive a car. There are also some interesting ethical decisions that aren't technically at all: https://www.technologyreview.com/s/539731/how-to-help-self-driving-cars-make-ethical-decisions/

13:51 < tham> zoq : I read the article you post, google claim their auto driving techinique could perform better than human, but not always

13:51 < tham> even the data show auto driving could reduce more accidents, there are ethical problem to solve

13:52 < tham> another problem I could think of is, would this technique make a lot of humans lost their jobs?

13:54 < tham> However, this is not a technicle problem, but more like philosophy issue

19:16 < zoq> tham: Right, I guess you could probably come up with the same issue when it comes to the computerization in general.