highd

16 ms·

121.
by highd 10y ago
The VAE "coder" is modelling a distribution p(z|x), and the decoder is modelling a distribution p(x|z). I like these slides: https://home.zhaw.ch/~dueo/bbs/files/vae.pdf
122.
by highd 10y ago
One of the most popular ways of using techniques like this is the "Variational Autoencoder". I've been working on using some alternate distributions with them as of late - it's very interesting, and quite powerful.
123.
by highd 10y ago
Said another way - light follows the shortest path between two points. In flat spacetime, that is a line. In curved spacetime, it may be some other path - like a path traced by an ant walking on a sphere. These paths are called geodesics:
124.
by highd 10y ago
The idea is that the vertical axis of the spectrogram is basically already an hierarchical set of features (in scale/frequency). Then convolutions on that is a lot like how DenseNets combine hierarchical features. I agree it seems a li
125.
by highd 10y ago
Yes, that was what I was saying. Absolute value, 2-norm are fine thanks to subgradient techniques and theory, as well as their differentiability over the majority of the function - but you can imagine tons of non-differentiable models where
126.
by highd 10y ago
"if the best model is not differentiable, you should still use it." I'm not sure I would say that - neural nets are "near everywhere differentiable", for example. Without differentiability we're stuck with, for
127.
by highd 10y ago
Sorry you're getting down-voted, I don't think it's an unreasonable question. In the sense I think you're using it, "statistics" are really methods for dimensionality reduction - we take means, and medians and
128.
by highd 10y ago
Another pro tip - absolute error magnitude is the convex hull of non-zero entry count for vectors (l_0 norm in some circles). So in the convex minimization context (and for most other smooth loss terms in general) you end up with solutions
129.
by highd 10y ago
Heliocentrism makes different predictions, it's a totally different comparison.
130.
by highd 10y ago
That would have to be worth the increased confusion in communication between experts using these different languages. Furthermore, given that the physics of the "pilot wave" are not significantly different from that of the wave fu
131.
by highd 10y ago
i.e. at best, it provides no new information at the expense of learning an entirely new framework.
132.
by highd 10y ago
Important point from paper abstract: >The pathway is up to five times more efficient than the in vivo rates of the most common natural carbon fixation pathway. Pretty awesome depending on how tight that "up to" bound is. As an
133.
by highd 10y ago
To be clear your complaints only described a subset of ML - neural network approaches in particular.
134.
by highd 10y ago
This is primarily in regard to deep neural nets, where second-order methods are too expensive (O(p^2) in parameters, where p is on the order of millions). I'm not sure conjugate gradient methods are used much, due to the non-convexity
135.
by highd 10y ago
Blog spam pass through to the original: http://sebastianruder.com/optimizing-gradient-descent/ Useful reference, 6 months old. Side note, anybody aware of any implementations of the "learning to learn gradient des
136.
by highd 10y ago
Reported thrust to weight ratio is 360 times raw photons: http://www.wolframalpha.com/input/?i=(1.2+mN%2FkW)%2F(1%2F(3... ) I spent a little time evaluation some form of quantum resonant enhancement of pair production a
137.
by highd 10y ago
http://home.basu.ac.ir/~psu/Books/%5BRamamurti_Shankar%5D_Pr...
138.
by highd 10y ago
Nice! I've hacked some shell scripts together to do something similar, but this is way better!
139.
by highd 10y ago
Please don't suggest fringe views to people with no experience in the field.
140.
by highd 10y ago
That's not necessarily survivorship bias. Survivorship bias would be "Here's 10 successful companies, do what they do". If you analyze a pool of random companies and then consider what the successful companies do differe
141.
by highd 10y ago
Did the jet stream change in some way recently? Or are they just now taking advantage of it?
142.
by highd 10y ago
You're not trying to get absolute position from those sensors, are you? I spent far too long on that project. At least at the precision we needed, my conclusion was that you would want to acquire and process the images yourself - both
143.
by highd 10y ago
I would be pretty hesitant to start talking about TensorFlow and Deep Learning before confirming, for example, at least a rudimentary understanding of Linear Algebra.
144.
by highd 10y ago
Ouch - worst of both worlds!
145.
by highd 10y ago
Somewhat off topic - has anyone used a desk bicycle (pedals) or treadmill for a significant period of time? Being able to do that for an hour every workday would be an amazing health improvement basically for free (in terms of time), but I
146.
by highd 10y ago
That's my thinking - so this should have crazy surface area. You could make some really awesome aluminum parts. Create a dense exterior with some interior "strut" infill structure and a very porous fill and put plumbing fixtu
147.
by highd 10y ago
I wonder how effective the micro-porous aluminum is at transferring heat from a cooling fluid to a heat sink. That might be useful even in active, non-evaporative cooling systems.
148.
by highd 10y ago
Cool - that's TV on RGB. There may be improvements on the channel selection in the literature in terms of "selectivity" for real images - i.e. HSV or something else. That might be an easy change that would bump your performan
149.
by highd 10y ago
Which is NP-Hard in general, but this instantiation of it can be made significantly easier than a random instantiation (in clock time, anyways). You could pick a similarity metric such that adjacent columns have some bound on worst-case sim
150.
by highd 10y ago
It seems like the spec supports digital and analog audio data. That seems like a potential nightmare for consumers - your headphones may plug in to you phone, but they don't have a DAC, and neither does your phone. Or what if they both

More ›