fchollet

14 ms·

151.
by fchollet 12y ago
> I'm sure we'll see the Circular Harmonic Transform for rotation invariant (up to a point) Deep Nets A fairly easy way to introduce rotation invariance in DCNNS is to perform random rotations on the inputs during training. Lik
152.
by fchollet 12y ago
How about random sleeps distributed so that the mean sleep time is always the same for every computation? (and much longer than any likely computation). timer.start(); computation(); time = timer.stop(); sleep(random_centered_on(mean-time))
153.
by fchollet 12y ago
I refer you to this post by the founder of Numenta, where he tries to inject some sanity into the "debate": http://recode.net/2015/03/02/the-terminator-is-not-coming-th...
154.
by fchollet 12y ago
> my biggest fear is that a powerful nation will use AI for their own advantage, mostly in the form of propaganda and mass manipulation through Internet. Chances are it is already happening. More advanced AI will give governments and c
155.
by fchollet 12y ago
Like every other technology, AI has risks and benefits. The main issue here, is that we are letting fear dominate the discussion, and that fear is not based on any supporting facts or evidence. Meanwhile, we are completely missing the discu
156.
by fchollet 12y ago
Here is a short recap of what the people who understand where machine intelligence is now, and where it is going, think of the whole "evil superintelligent AI" blogging trend we're seeing these days. These are some of the peo
157.
The Terminator Is Not Coming. The Future Will Thank Us(recode.net)
17 points by fchollet 12y ago|11 comments
158.
by fchollet 12y ago
I'm sure a number of people who have contributed to Torch are working at DeepMind. However, Torch has been around for much longer than DeepMind (about 12 years at this point). Two of the major contributors to Torch, Ronan Collobert and
159.
by fchollet 12y ago
Interesting that they're using Torch7. The code is pretty concise and readable, very cool stuff.
160.
by fchollet 12y ago
I don't think anybody would posit that it's unreasonable to entertain the idea as kind of far-fetched long-term possibility, much like encounters with alien life or faster-than-light travel. It's the fear-mongering that'
161.
by fchollet 12y ago
I think you nailed it. "AI doom" is a powerful psychological trope for people in technology, and much like the hypothetical rogue super-AIs, it has gone out of control.
162.
by fchollet 12y ago
> I hope you aren't a researcher Andrej is a brilliant researcher, currently doing his PhD. He has a bright career ahead. Maybe you should actually read his comment instead of dismissing it crudely, and likewise for the thoughts of
163.
by fchollet 12y ago
His thoughts seem quite sensible, although extremely abstract and theoretical. This is lightyears away from the Elon Musk / Bill Gates / etc. fear-mongering.
164.
by fchollet 12y ago
Shane Legg is known for being a co-founder of DeepMind, in a business role (as I understand). He's a complete nobody as a researcher (is he even an AI researcher? I would be surprised). The big names of deep learning have all taken a v
165.
by fchollet 12y ago
I have yet to meet a serious AI researcher who worries about AI ending the human race. At every AI conference I've been to lately, some guy would inevitably ask the same question to the speaker: "do you think we should be concerne
166.
by fchollet 12y ago
There seems to be some confusion between complexity, quantity of DNA, and intelligence as a measure of organism complexity. A larger quantity of DNA does not translate into a more "complex" organism, as amoebas prove. In fact, I w
167.
by fchollet 12y ago
What who you rather get out of your tax dollars, roads and parks and fast internet, or explosives and ammo? Both mean jobs for Americans all the same. It's the value of the output that's different. Why funnel efforts and capital i
168.
by fchollet 12y ago
You could try contacting directly active developers/startups who have released similar apps, they are the ones most likely to be interested (and they're definitely not checking flippa.com). I imagine it wouldn't take too long
169.
FAA Opens Skies to Drones in First U.S. Unmanned Flight Regulation(bloomberg.com)
1 points by fchollet 12y ago|0 comments
170.
by fchollet 12y ago
I wish I was a high-profile VC. Then I too would be writing essays exhorting people to take on high amounts of risk and to spend their 20s and 30s grinding through 60 hour workweeks, so I could grow my portfolio. Because, we're losing
171.
by fchollet 12y ago
The large majority of bots nowadays use Javascript, and are almost indistinguishable from humans if you don't look at their activity over time. Filling a form with coherent content is a reasonable proof of humaneness, but only if you&#
172.
by fchollet 12y ago
The accuracy reported is top-5 accuracy, meaning that a model is considered correct on a test image if it includes the expected label in its top 5 predicted labels. This does mitigate the multi-object issue quite a bit.
173.
by fchollet 12y ago
Genetic programming is an inefficient search method, and will require many evaluations of the cost function to optimize anything. In the case of DCNNs, evaluating an architecture can keep a modern GPU busy for days, so genetic algorithms ar
174.
Surpassing Human-Level Performance on ImageNet Classification [pdf](arxiv.org)
53 points by fchollet 12y ago|13 comments
175.
by fchollet 12y ago
> One layer Autoencoder - SVD done by a neural network Not necessarily. Any serious user of autoencoders would apply some kind of L1 regularization or other sparsity constraint to the coefficients learned, so that the autoencoder does no
176.
by fchollet 12y ago
We had a Lockitron on the main door at my previous office. I can't even tell you how many times I got locked out because the device would get stuck (only way to open the door in these cases was with the key, which only two people had).
177.
by fchollet 12y ago
In any case, I think that's a point very much relevant to the analogy: when you do more of the same, but at a scale a few orders of magnitude larger, you do get qualitatively different results. That's very much applicable to AI. A
178.
by fchollet 12y ago
Your "skyscraper" would need to have its top in geostationary orbit, or else it would be wrapping up around the earth as its top would be in LEO or HEO and thus would be moving at a different speed than its base. I.e. your skyscra
179.
by fchollet 12y ago
Teaching is made of two components: access to knowledge, and motivation. Today, access to knowledge is almost entirely irrelevant: pretty much all knowledge is freely accessible from anywhere, anytime. You just need the motivation to look i
180.
by fchollet 12y ago
These artifacts were found by random chance, often while plowing a field. Which means that the vast majority of such artifacts has never been found. So if we found a few already, it is quite likely that there are many more such identical di

More ›