highd

14 ms·

31.
by highd 9y ago
The result is in the large sample limit, which is pretty much never the case for high dimensional datasets like the ones most popular for ML these days (images, audio, text). It doesn't mean what the parent thinks it means.
32.
by highd 9y ago
So terminal UIs are coming back?
33.
by highd 9y ago
I highly recommend switching to the conda distributions and package ecosystem. A lot of those issues have gotten a lot better for me since I did that.
34.
by highd 9y ago
"that have entered our diet far too short a time ago to have had evolutionary impact"
35.
by highd 9y ago
Our diet is full of fruits and vegetables containing thousands of complex chemicals, indistinguishable from "synthetic" chemicals, that have entered our diet far too short a time ago to have had evolutionary impact. If one wanted
36.
by highd 9y ago
I used to use this a bunch when calculating gradient descent algorithms, but I've since moved to using libraries with auto-differentiation (numpy with autodiff or theano) and it's a huge force multiplier for my work (not to mentio
37.
by highd 9y ago
It seems like a more comparable reinforcement learning thing to do would be to combine the entropy criterion with a known reward when available in some way and then do Q learning on that without the simulation requirement. Then in cases whe
38.
by highd 9y ago
Yes - the issue is that the work is currently presented as requiring "no training", but it has simply relocated that problem to constructing a perfect simulation of the environment. It then uses the fact that current benchmarking
39.
by highd 9y ago
I'm referring to posts like this: https://entropicai.blogspot.fr/2017/06/openai-first-record.h...
40.
by highd 9y ago
I understand. It appears that constructing the problem this way is a very unfair way to measure if this idea works compared to other reinforcement learning approaches. If you can simulate the system perfectly you can always just simulate k
41.
by highd 9y ago
In that case, aren't the big benchmark gains being claimed mostly the result of changing the problem to allow a perfect simulator in the system? The original author is claiming benchmark results either with a perfect simulator or with
42.
by highd 9y ago
The original work. What is the optimization problem being solved precisely? What exactly is done prior to submission to the OpenAI gym? What data does the system have access to prior to submission and during runtime?
43.
by highd 9y ago
I've been trying to parse this body of work - there doesn't seem to be a writeup on the exact implementation, just that it's using "Causal Entropic Forces". They have this writeup on the optimization implementation
44.
by highd 9y ago
Yep, this is a current area of research for content generation. I think most current approaches build some transform to a latent space and then compare generated images with their nearest neighbors in the training set. If they're ident
45.
by highd 9y ago
SEEKING WORK | Boston | Remote or local Algorithms / Machine Learning / Optimization Problems / Statistics as well as the full stack infrastructure to build a visual interface around it (generally Flask + D3.js, occasional Re
46.
by highd 9y ago
My thinking is this is a good GAN problem. L2 norm will have these bad trivial upscalings as local minima - since L2 in time domain is the same as L2 in frequency domain, you can think in the frequency domain that it basically has this big
47.
by highd 9y ago
Great catch. This is a super important factor. The real concern in science is how easy it can be to hide the fact that so many possibilities were tested, precluding this observation. As another example, there are GWAS studies where people t
48.
by highd 9y ago
True for graphics, though it would be nice for HPC. See https://news.ycombinator.com/item?id=14508928 Though it may be that expensive high-end VR could be the luxury home theater of tomorrow, and you put 4 GPUs in a box to
49.
by highd 9y ago
I think this might be another solution, using the capacitance of the diode to make a resonator and then driving near resonance. Here V(n001) is some pin voltage and V(n002) is the voltage across the diode: http://imgur.com/a
50.
by highd 9y ago
Wow, thanks!
51.
by highd 9y ago
Still pretty good - thanks!
52.
by highd 9y ago
That sounds amazing - do you know of any in particular? I can't seem to find any motherboards with 8 PCIe slots. Or do they need some sort of additional splitter?
53.
by highd 9y ago
I feel like you should be able to do better than $15K for $7.2K worth of graphics cards.
54.
by highd 9y ago
I've been evaluating this space a fair bit recently. If you want to optimize FLOPS/$, especially for research-workstation sorts of setups, there's unfortunately not a lot of options for getting more than 6 GPUs in a motherboa
55.
by highd 9y ago
I would say since ResNet beat state of the art on ImageNet in 2015. It's not standard, but that's what state-of-art can take.
56.
by highd 9y ago
Please don't be a jerk on HN. 5 layer CNNs are not best practice anymore, and it's not a fair comparison. Your flippant attitude is unwelcome.
57.
by highd 9y ago
True - it would be interesting to use, say, the coding layer of an autoencoder trained on one of the many face datasets and then training only on the last few layers to fit their data.
58.
by highd 9y ago
A 5 layer CNN is absurdly shallow, so this isn't particularly surprising. I routinely work with 150+ layer CNNs - that's fairly standard practice if you want high-quality results.
59.
by highd 9y ago
In particular, if they're so confident about the absurdity of the critiques, they should publish those as well. Maybe dentists have some real concern that these inventors are dismissing out of hand - the best way to dispel that, if fal
60.
by highd 9y ago
Proof-of-Stake will be released on Ethereum soon and will greatly reduce the mining overhead, FYI.

More ›