cr4zy

11 ms·

61.
by cr4zy 10y ago
Convergence on target speeds and steering is actually very quick (a matter of hours on a modern GPU / AlexNet / 227x227 RGB) once you have the data. Generating training data with the correct speed limit and providing speed limit b
62.
by cr4zy 10y ago
Yes, it would be great to have source! The modding community does make up for this in some ways. For example, mods provide access to things like finding memory addresses for things like steering angle that weren't available in the API
63.
by cr4zy 10y ago
It's surprising how scriptable the game is. For example, the car I train the AI with is set to the least aggressive driving style which stops for rabbits crossing the road :) Also, all violence against the player is disabled.
64.
by cr4zy 10y ago
We currently don't look at being shot at (although it's an interesting suggestion!) - the current reward function is basically this: distance_reward - off_road_penalty - speeding_penalty - slow_penalty - discomfort_penalty;
65.
by cr4zy 10y ago
So the training car has a max speed of 10 meters per second or 22mph which the net matches. Future datasets will contain the road's speed limit as well as throttle / brake targets required to reach that speed limit.
66.
by cr4zy 10y ago
Thanks! Yes, the game is highly scriptable[1], so you can test the types of scenarios you'd want. We've disabled violence against the driver, but could enable[1] it on-demand if it's useful for training. [1] http:/&#x2F
67.
by cr4zy 10y ago
deepdrive.io creator here - I'm actually not affiliated with the Berkeley project of the same name. There's also a DeepDriving at Princeton plus plenty of other (mostly perception) projects using GTAV, so it can be confusing. I&#x
68.
by cr4zy 10y ago
Robot rights make sense to prepare for a time when there is rough intellectual parity between humans and robots and the possibility of an adversarial relationship arises. I think it's unlikely, but still important, given the impact suc
69.
by cr4zy 11y ago
Looks like they're moving it to GitHub wiki: https://github.com/Microsoft/CNTK/wiki/Config-file-overview
70.
by cr4zy 11y ago
This is great news! I think distributed access, control, and contribution to the best AI's will help create 'safe' AI's much faster than any AI created in secret. One thing this does not address, and is something that Je
71.
by cr4zy 11y ago
Looks similar to the top project on this page from cs231n by Jon Gauthier: http://cs231n.stanford.edu/reports.html
72.
by cr4zy 11y ago
I highly recommend turning on page throttling in the Chrome Dev Tools sometime. You'll be amazed at even how slow even 4G seems.
73.
by cr4zy 11y ago
Another cool resource for learning to apply and implement Kalman filters is Udacity's AI for Robotics (focused on self driving cars) course by Sebastian Thrun. Apparently Kalman filters are how Google's self driving cars predict t
74.
Resveratrol and Clinical Trials (2013)(ncbi.nlm.nih.gov)
23 points by cr4zy 11y ago|2 comments
75.
Dependency solving with SAT for Anaconda(continuum.io)
1 points by cr4zy 11y ago|0 comments
76.
Starting up at a startup(blog.clari.com)
2 points by cr4zy 11y ago|0 comments
77.
Deepmind releases source for DQN(sites.google.com)
2 points by cr4zy 12y ago|0 comments
78.
DeepMind: Human-level control through deep reinforcement learning(nature.com)
2 points by cr4zy 12y ago|0 comments
79.
Plotly(plot.ly)
2 points by cr4zy 12y ago|0 comments
80.
Jetbrains C/C++ IDE – CLion(jetbrains.com)
1 points by cr4zy 12y ago|0 comments
81.
by cr4zy 12y ago
Hugely excited about OpenWorm! More great stuff about the robotic embodiment of OpenWorm: http://www.i-programmer.info/news/105-artificial-intelligenc... https://news.ycombinator.com/item?id=8745639 Ey
82.
A cheaper way to capture carbon with baking soda capsules(qz.com)
1 points by cr4zy 12y ago|0 comments
83.
by cr4zy 12y ago
Also check out this embodiment http://www.i-programmer.info/news/105-artificial-intelligenc... And some implementation details https://news.ycombinator.com/item?id=8745639
84.
Linear Classification Loss Visualization(vision.stanford.edu)
9 points by cr4zy 12y ago|0 comments
85.
Show HN: Happy Holidays from Clari(vimeo.com)
1 points by cr4zy 12y ago|0 comments
86.
Speculations Concerning the First Ultraintelligent Machine (1965) [pdf](webdocs.cs.ualberta.ca)
15 points by cr4zy 12y ago|1 comments
87.
Long bets(longbets.org)
4 points by cr4zy 12y ago|0 comments
88.
by cr4zy 12y ago
The second half of this video has an earlier talk that goes along with the slides: http://youtu.be/S9twUcX1Zp0?t=22m49s
89.
How neuroscience can help computer science – Yann LeCun(facebook.com)
7 points by cr4zy 12y ago|0 comments
90.
Neural Turing Machines by DeepMind [pdf](arxiv.org)
8 points by cr4zy 12y ago|1 comments

More ›