Hacker News Reader

mddaon Academic Torrents: A distributed system for sharing enormous datasets

PostDocs may have a bad rap, but 'subhuman' is going a bit far...

mddaon A Surveillance Master Dissects a Murder from the Sky [video]

Talk about attention to detail : every page is marked "PSS Proprietary Infomration"

mddaon Facebook AI Research Team Open Source DeepMask and SharpMask

Because vonnick has a habit of plugging deeplearning4j in every deep learning-related thread. IMHO, that's somewhat to be expected/laudable here on HN - particularly since deeplearning4j has an open-source framework, and has published interesting and informative blog posts. OTOH, they're ours-is-more-serious because java-is-enterprise thing is a little out of mainstream, so having OpenAI and deeplearning4j being highlighted in a way that seems to equate them isn't as helpful as it appears.

mddaon Otto joins Uber

Pressure on two sides of a surface don't have to be equal. The overall forces do, but there's an additional component being applied by the surface itself. An alternate way of looking at it : consider how the space station internal pressure differs from space.

mddaon What Danes consider healthy children’s television

As an early movie recommendation, I suggest WALL-E. We started with it at ~2ish, and every time we watch it together, there's more depth revealed (currently at 5.5years old).

At first, it's about a robot and another robot doing stuff. Then they're on a journey. Then the desire for friendship. And hospital. Then (gradually) the idea of different motivations comes in. Perhaps some ecology. Then conflict, etc.

All the while, my daughter is delightfully more interested in robots and space travel than standard 'pink princess' fare.

Of course, YMMV.

mddaon What Danes consider healthy children’s television

Easily ~Youtube-able option with lots of episodes : "Kipper the Dog".

Also 'findable' : The Clangers (charming 1970s show from the UK); Mr Men; Shaun the Sheep; Mr Benn.

All of these are 'gentle' but fun, and not really educational per-se. Even though you'll end up watching all of these again-and-again, Shaun the Sheep remains entertaining for parents too.

* https://en.wikipedia.org/wiki/Kipper_the_Dog

* https://en.wikipedia.org/wiki/Clangers

* https://en.wikipedia.org/wiki/Mr._Men

* https://en.wikipedia.org/wiki/Shaun_the_Sheep

* https://en.wikipedia.org/wiki/Mr_Benn

mddaon Making Sense of Everything with words2map

The paper (with fewer typos) was actually accepted into 'ICONIP' in Japan in October - so I'll definitely have code on GitHub by the end of the summer. Currently my Theano implementation is buried in typical exploratory kind of code, which just needs to be stripped away to make something functional from GloVE->Sparse in one command.

The NNSE paper has associated code already, but I found setting the sparseness preference parameter was very hit-and-miss, which is why I preferred the explicit sparse-by-percentage measure in my work.

mddaon Making Sense of Everything with words2map

Embeddings can be compressed even further than that : https://arxiv.org/abs/1511.06397 (disclaimer: I'm the author)

mddaon Making Sense of Everything with words2map

It is a genuine message. It is also simple. But it isn't actually true. Computers are not able to learn like humans : the learning process is entirely different. And what they learn is different - even though it can be visualized in a way that makes it look 'humanlike'.

These techniques are impressive, and yhat is demonstrating that they are very capable. It's just that I feel a little sad that the 'AI pitch' is being turned on, when the 'really good tech' is a much more valid way to understand what they're doing.

mddaon Making Sense of Everything with words2map

"We are now at a point in history when algorithms can learn, like people, about pretty much anything. " seems pretty disingenuously worded.

One infers from a quick read ~"Algorithms are now like people, and can learn about anything." But careful parsing of the commas shows that the sentence is true, but in the precise sense that "People can learn about anything. Now, algorithms can also learn about anything." - and the extent of learning/understanding is not being compared.

Perhaps I'm nit-picking, but this statement appears to have been constructed to support an AI pitch, and is literally true, but no 'actual AI' is involved (and no-one is actually claiming it is... unless you /want to believe/).

mddaon Singapore will pay startups to solve its problems

The government actively pursues areas in which Singapore might have a comparative advantage - and would discuss it with you in precise detail... They would wear a "technocrat" badge proudly.

btw : Is this "Correct-by-Construction" seminar [1] relevant?

[1] https://mysoc.nus.edu.sg/~cmsem/SemPDF/sem_12396.pdf

mddaon NVIDIA Announces the GeForce GTX 1000 Series

(nitpick) actually 10 TFLOPs is a more appropriate target. Doesn't make sense to aim to cripple your performance by a factor of 822 for marketing reasons...

mddaon RISC-V Offers Simple, Modular ISA

This seems like an ultra-basic question (sorry). On the VectorBlox/orca github page, it mentions that the core takes ~2,000 LUT4s. Are those numbers apples-to-apples with the 22,320 LEs given for the Cyclone IV board mentioned earlier[1]?

If so, then (naively) could one pack ~10 on that single FPGA? Or does the 'packing overhead' become a big problem? Or does the design use more (say) multiply units pro-rata, so that they become the limiting factor?

[1] https://www.adafruit.com/product/451

mddaon Image Analogies using Neural Networks

This makes me a little sad, though : "University of Tübingen has a pending patent application for the Neural Art technology."

mddaon Singapore witnesses the rise of the entrepreneur

New rules were instituted, requiring jobs to be advertised locally first : [1]

In addition, as a EP-holding business-owner, I was 'asked' to explain my plans to employ Singaporeans/PRs in the coming year.

OTOH, I don't see this as entirely unreasonable : After all, the USA does the same (for E-2 investor visas, for instance). Moreover, the Singapore visa turn-around time is only a few days (after an online application), whereas the US visa process makes one long for the friendliness of the DMV...

[1] http://www.mom.gov.sg/employment-practices/fair-consideratio...

mddaon Singapore witnesses the rise of the entrepreneur

One issue that seems to resonate with people in Singapore is that while in the rest of the tech world :

Developer == Essential (e.g. : CoFounder)

in Singapore :

Developer == programmer == glorified typist == someone whose job can be outsourced.

This gets in the way of attracting people to programming in the first place.

The environment is also disappointingly oriented towards local graduates aspiring to become Trainee Managers at Multi National Companies (MNCs). It's not a money thing IMHO - more of the status associated with working at a recognisable brand-name firm.

Clearly, there are exceptional people around, but there's an additional (surprising) hurdle for local tech talent here.

mddaon Deep Residual Learning for Image Recognition

This Microsoft Research's approach, that romped to first place in the recent ImageNet challenge [0].

What's neat is that the technique is an almost comically simple way to add extra layers to a network. It's commonly accepted that deeper networks can learn better, but they get very unwieldy/difficult to train as they get deeper.

Roughly speaking (and please correct me if I'm off-base), the paper's technique is to slot in additional layers that that are initially 'identity+', where the new layer then gets trained to hone in on the differences from 'identity'. This training on residuals alone is more stable, since answers near each '~0' starting point are simply as good as the original network - any improvement is a pure win.

So... their winning network has a breathtaking 152 layers (and then ensembles a few of them together).

[0] http://image-net.org/challenges/LSVRC/2015/

mddaon Genetic Algorithm 2D Car Thingy

I researched GAs / GPs back in the mid-90s (but then went in a different direction).

Is there a paper/presentation that embodies the current best practices/thinking that you could recommend? I'm not trying to be lazy, it's just that there is clearly a lot of retro thinking among the top search results on the net, and it's difficult to separate the wheat from the chaff...

mddaon The Musicians Behind One of the Most Sampled Songs in History Finally Got Paid

The words "I" "have" "a" "dream" take less than six seconds to say, so surely that information content has been hit upon by others in the past... As someone else has posted, there's a great documentary about the whole sampling/resampling scene that explains a bit about the attractions of this particular sonic segment : Suffice to say, there's a lot more to it than just a 1-2-3-4 pattern.

mddaon Fedora 23 released

Suppose I just want to use my Nvidia card in GPU mode (i.e. monitor only connected to Intel integrated video), does the X.org version still impact me? Or does the 'nvidia' proprietary driver somehow pull in Xorg dependencies even when not displaying anything?

mddaon Kexi Project – an open-source visual database applications creator

It's proprietary, but has a long history as an MS-Access (and other DBs) app maker : http://www.alphasoftware.com/ms-access-mobile.asp

mddaon Deep Learning Courses

Theano has the infrastructure in place for OpenCL, but not all 'operations' are implemented, which (for any decent calculation) means that it's a no-go [1].

Unfortunately, the focus at the Montreal lab that has a huge influence on its development seems to be (a) 'blocks' for a high-level DNN environment (which is very cool) and (b) CUDA-to-the-max (which is understandable, given Nvidia actively seeds research labs with freeby cards, and -- as evidenced by the article -- is putting a lot of effort into supporting deep learning).

rant start:

It's a shame that OpenCL doesn't get more love. Just the other day there was a cool Clojure GPU project (based on OpenCL) announced on HN. One of the comments was 'will you be building this for CUDA too?'. Rather than pressure open source writers to support closed systems, it would be better to pressure Nvidia to provide up-to-date OpenCL drivers. Newer Nvidia cards are at OpenCL 1.2. And the (somewhat old) OpenCL drivers are always there in an Nvidia install. But does Nvidia ever talk about that : No. It's entirely in Nvidia's interest to encourage everyone to talk CUDA-only. But on a GFLOPs/$ basis, and for the cause of Free, CUDA isn't the right way to go.

rant end.

[1] https://github.com/Theano/Theano/issues/2936

mddaon The tech boom may get bumpy, but it will not end in a repeat of the dotcom crash

Yup : and they even mock themselves for having done so : http://www.economist.com/node/9465026

mddaon Ask HN: What problem in your industry is a potential startup?

https://en.wikipedia.org/wiki/Application_lifecycle_manageme...

https://en.wikipedia.org/wiki/PTC_%28software_company%29

https://en.wikipedia.org/wiki/Alternative_terms_for_free_sof...

mddaon Show HN: A Python with Hindley-Milner-like type annotations, compiling to C

FWIW, the geany editor has a GeanyPy plugin that enables you to write other plugins entirely in Python.

mddaon Nvidia Pascal GPU Architecture to Provide 10X Speedup for Deep Learning Apps

FLOP-wise that makes sense. But for deep learning, the big deal is in the 12Gb GPU-local memory, which has enormous bandwidth (and can store more of your dataset / parameters at once). The largest concern with GPU processing is keeping the GPU adequately fed with data - and avoiding round-trips of blobs of data with the CPU helps a lot.

mddaon Nvidia Pascal GPU Architecture to Provide 10X Speedup for Deep Learning Apps

I was at an Nvidia presentation where they made a big deal about just being 'on-silicon' rather than having to go through copper connects (to a bus external to the chip). Each join, corner, etc adds to capacitance on the wire, which then leads to delays (alternatively power consumption, and heat).

mddaon Aetherial Symbols

The 'big activity vector' language is partly a reference to the Vector Word Embedding stuff (see [1] for an explanation). The surprising thing about that is that it is possible to learn an embedded of individual words in a multi-dimensional vector space, such that (for example) vector(Queen)-vector(Woman) ~= vector(King)-vector(Man). Which is to say that there's a general 'royalty' direction within the vector space - and all this can be learned purely from seeing large amounts of English text (no 'traditional' supervised training). Perhaps a 'god' could identify the meaning of each direction in the space (or region of words), but the big Machine Learning labs (Google, Baidu, Montreal, Stanford, Facebook, etc) are proving that purely manipulating the vectors in the abstract works really well.

In addition to the 'word vectors' as inputs, the RNNs illustrated are also iterating over an internal state (flowing from left to right though the same network for each new word) - and this internal state is also an embedding of some kind. But it's going to be very difficult to decipher what each dimension here represents, as it's being built purely as a function of the input word vectors, its own previous state and a NN with initially random weights.

Now, although actual 'brain experiments' have shown that individual neuron (or local clusters) apparently light up when particular thoughts are had (alternatively, cause thoughts to be had), each cluster seems likely to be just one aspect of (say) 'dogginess'. So, one area will correspond to the smell of dogs, others to wet noses, others to being outdoors (i.e. all aspects of the overall 'dogginess' concept) - but these things will all overlap in multiple ways with other concept 'vectors'. Which is how huge spaces of ideas are searched in parallel, rather than sequentially (using, say, an is_doggy_quality symbol).

There are also parallels here with the Numenta Sparse Distributed Representations [2].

Overall, this presentation seems to be probing at the frontier of what works, and how to leverage that up into something that's more about 'general thinking' rather than pattern matching. It also appears to be a thought-piece, rather than a conference presentation (though, of course, Hinton deserves to be heard on just about anything in NNs, IMHO).

[1] http://colah.github.io/posts/2014-07-NLP-RNNs-Representation... [2] https://github.com/numenta/nupic/wiki/Sparse-Distributed-Rep...

mddaon Aetherial Symbols

Perhaps it's not getting votes because the title is so non-descriptive...

"Hinton's internal presentation on AI and Deep Learning" might attract the attention it deserves.

(I'm not really advocating a title change/resubmission, but if someone's reading the comments before going to Google Drive, it may be more of a hint about value/bandwidth)

mddaon An Analysis of SpaceX’s Falcon 9 Crash Landing

You can't count position and velocity (i.e. d(position)/dt) as separate degrees of freedom. If you insisted on doing so, then you should also count rocketthrust and rocketthrust changes through time.

IMHO, the constraints here (as an armchair engineer) are more that the 'big rocket' isn't very responsive to requested changes in thrust, whereas the 'little nose rockets' may be responsive, but very weak compared to the mass of the thing they're trying to control.