HNHacker News
TopNewBestAskShowJobs

ssfrr

463 karma · joined February 3, 2014

spencer@heavisideresearch.com https://heavisideresearch.com
submissionscomments
ssfrr··on Cambridge University releases an app that improves concentration akin to Ritalin
The game looks like a gamified version of the test they're using to measure concentration. It seems likely that the subjects are getting better at that specific task through practice, but I'd be skeptical that the results would generalize to other tasks requiring concentration.
ssfrr··on Neural Ordinary Differential Equations
> as far as I know we were the first people to release an open-source set of differentiable solvers using reverse-mode adjoint sensitivities

Would the DifferentialEquations.jl work in the Julia community qualify? As I understand it for the pure Julia solvers you can run them through autodiff to differentiate with respect to the parameters. Chris Rackauckas has talked a lot about how cool it is that he can take his solvers that were not written to be AD-aware and use somebody else’s AD package that’s not specialized for diff eqs, and combined you get differentialable diff eqs.

ssfrr··on LibrePCB: First stable release published
They make high-precision low-resistance resistors specifically for current measurement. Your standard cheapo zero-ohm resistor though is more for creating optional connections that you can configure at assembly time.
ssfrr··on Show HN: A realtime node-based audio synth based on LLVM
Thanks for replying, I'd definitely be interested in more architectural details.

3ms to complete codegen is orders of magnitude faster than I expected. Does that include everything necessary from making the change through actually synthesizing samples?

ssfrr··on Show HN: A realtime node-based audio synth based on LLVM
Looks super slick, thanks for sharing!

Can you give a little more detail in how this compares architecturally with something like PD? I imagine a tradeoff of generating LLVM MIR and compiling it is that you can get better performance, but don't get the interactive experience of hearing changes immediately as you tweak the object graph.

Does the DSP graph operate on blocks or sample-by-sample?

ssfrr··on Artwork created by AI sold for $435,000. Algorithm engineer not credited.
That's a pretty poor example. Not all tools are equal, and one of the important ways to describe a tool is by what degrees of freedom it affords the user, and what aesthetic decisions are baked into it. A paintbrush has a lot of degrees of freedom and makes few assumptions. Even for tools that do a lot for you, finding unexpected/unintended degrees of freedom or subverting those aesthetic decisions is one way that artists innovate. From the article it seems that most of the aesthetic decisions in the resulting work were baked into the tool, not made by the artist.

That said, a huge part of an artist's work is contextualizing their work and building a story around it, so in some ways if these guys were able to successfully convince people that it was worth $435k then good on them. I just don't expect that to last. When video equipment was inaccessible to most, simple video art was innovative and new. Now that the tools are easy to get and use the work needs to be more sophisticated.

ssfrr··on Oboe: A C++ library for low latency audio in Android
Check out the Bela project[1], which has back-ends for a few different audio packages (e.g. PD and SuperCollider) that totally bypass the kernel and handle the audio callback directly (maybe from a bare-metal ISR, not exactly sure). That’s how they get sub-ms round-trip latencies.

[1]: http://bela.io

ssfrr··on Japanese Medical School Deducted Points from Exam Scores of Female Applicants
I think deciding how best to counteract systematic bias at earlier stages of the pipeline is difficult, and I don't think we should accept widespread collateral damage. That said, I don't think we should use that as an excuse to blow up the whole idea and ignore the bias (I'm not saying that's what you're advocating, just expanding my point).
ssfrr··on Japanese Medical School Deducted Points from Exam Scores of Female Applicants
I don't think it's a stretch to say that life is harder (on average) for minorities.

Ignoring the role of racism and sexism to try to put all the blame on socio-economic background is a gross oversimplification. The NYtimes had a really nice infographic [1] showing the racial disparities in class mobility. tl;dr - white folks who are born rich tend to stay rich more often than black folks, and those who are born poor are more likely to move upwards.

[1]: https://www.nytimes.com/interactive/2018/03/19/upshot/race-c...

ssfrr··on Japanese Medical School Deducted Points from Exam Scores of Female Applicants
I don't think it's really relevant to my point above, but no, I think on average asian americans face more racism than white americans.
ssfrr··on Japanese Medical School Deducted Points from Exam Scores of Female Applicants
Except all other things are not equal. Someone who has achieved at the same level while also fighting sexism/racism has had to work harder and be better to get there. They’re playing on hard mode.
ssfrr··on Why I Ripped the Same CD 300 Times
That assumes the noise is uncorrelated though (which the noise on a camera sensor generally would be). If the noise is correlated then averaging doesn't help.
ssfrr··on Microfilm Lasts Half a Millennium
That only covers the errors due to aliasing. There's still going to be additive noise introduced during the transfer that will build up each time the content is transferred.
ssfrr··on My favorite things that are coming with Julia 1.0
Whether elementwise or linear algebra operations are more common depends on your use case. I’m personally super happy not to need to write np.dot(A, b) when I could just write A*b, or a’b for an inner product, which matches the math and will do the right thing with complex numbers.
ssfrr··on My favorite things that are coming with Julia 1.0
Hopefully I can clarify some of the "maybe it is, maybe it isn't" in this post - I've found the Julia community to be almost universally full of helpful and friendly folks, mostly scientists and/or people with a language nerd bent.

I'm not sure if there's a fair way to quantify drama in a community, but it hasn't been a big issue for me relative to all the positive interactions I've had with other julians.

ssfrr··on My favorite things that are coming with Julia 1.0
Better cache access patterns are one of the reasons Julia's dot-broadcasting[1] is super cool. If you have big vectors `a` and `b`, the expression `sin.(a .+ exp.(b))` will do a single pass over your data, calling `sin(a[i]+exp(b[i]))` for each element, rather than creating big temporary arrays for the intermediate expressions and looping through multiple times.

Then because putting all those dots can be unwieldy, there's the `@.` macro which puts a dot on all your function calls.

[1]: https://julialang.org/blog/2017/01/moredots

ssfrr··on Show HN: Write every day, measure your progress, achieve your writing goals
FWIW I found your feature descriptions pretty tight and compelling. It was cool to read through what's available and then pop to the top of the screen to see it in action (particularly the typewriter and blur modes).

The main thing I didn't get a feel for was how this works as your document gets longer, or you're working on multiple documents. If you're writing a book does the whole thing end up being in the same text area and you just scroll around? Is there a save/load?

ssfrr··on Live Coding in Sporth: A Stack-Based Language for Audio Synthesis [pdf]
I think the issue of control rate in PD is often confusing for folks, because there isn’t any “control rate” that is fundamental to the language. There are audio rate signals that are processed every block, and there are messages that get handled ASAP when they are received, and are decoupled from the audio rate graph. These messages could be once an hour or once per millisecond. The control rate thing comes up because most PD objects (e.g. osc~) only update their parameters on block boundaries, which creates timing jitter (variable delay between when a message is received and when it takes effect). You could definitely have an audio-rate object that took into account the precise timing of the messages it received and scheduled the parameter changes a constant amount of time in the future, which eliminates jitter at the expense of a fixed latency. The vline~ object is useful for generalizing this approach to any object that can take audio-rate parameters, but there’s no reason you couldn’t make a vosc~ object that had the functionality built in.

This is different from supercollider (and I think sound), which have an explicit control rate.

ssfrr··on Show HN: I wanted better offline music on the iPhone, so I made Doppler
If all the music you want to listen to is available via streaming services than great, but I have a bunch of MP3s that aren't (things ripped from old CDs, music from my friends, music from anti-streaming artists).
ssfrr··on Show HN: I wanted better offline music on the iPhone, so I made Doppler
+1 - I have a couple Chromecast Audios that I use as a poor-man's Sonos and it works pretty well, but the Spotify iOS app's Chromecast streaming is pretty half-assed. It can only adjust volume when the app is in the forground, and sometimes not even then, sometimes forgets it's streaming and loses sync.
ssfrr··on Ask HN: What happened to the semantic web?
The flight info example is one of the places where semantic web tech went mainstream. Those flight emails have embedded metadata in JSON-LD (linked data) format and Gmail uses it for more specialized display[1].

[1]: https://developers.google.com/gmail/markup/reference/flight-...

ssfrr··on Why are successful professionals still working 70 hours a week?
This comment really baffles me, particularly the Office Space reference. Generally I see that movie invoked specifically to reference the meaninglessness of most white-collar work. TPS reports and pieces of flair are supposed to be absurd examples of things that employers to do dehumanize and control employees - the opposite of meaningful goals.

I think most people want meaningful (purposeful?) lives. Some find that almost exclusively through their work, some through their nonprofessional activities, some through their families. Most are probably a mixture. I think people often find happiness through figuring out what's important to them and organizing their life around that.

ssfrr··on Keyboard latency
Don't confuse throughout and latency. The mouse may be measuring and sending data every 1-2ms, but that doesn't say anything about the latency before the data is handled.
ssfrr··on MILA and the future of Theano
I don't have a lot of experience, but I recently ported an RNN model from lasagne/theano to keras/tensorflow, and the latter combination was about 4x slower. Not sure whether it's keras or tensorflow causing the slowdown. The API is nice though.

I also tried to flip the keras switch to run on top of theano instead but it had issues that I didn't have time to fix, so I just stuck with the original lasagne/theano stack.

ssfrr··on I don’t know who the Web Audio API is designed for
PortAudio is MIT-licensed[0] and seems like a decent example of the primitives you need for audio.

Broadly low-level audio APIs are divided into 2 categories:

1. Callback-based - every time the underlying system has a new block of audio available and/or needs to be supplied with a new block, it calls your callback, which reads input data, does whatever processing you want, and writes output data

2. Stream-based - Inputs and Outputs are represented by streams. You can read from the input stream to record and write to the output stream to play back.

Both types of API can be used for low-latency audio, but you generally introduce a buffer of latency when you need to convert between them.

Portaudio lets applications choose which API they want to use.

[0]http://www.portaudio.com/license.html

ssfrr··on A new way to block ads but still support content: buy them yourself
I think the interesting thing here is that they're using the existing ad auction systems, but you're bidding against the ad companies. So it doesn't require the sites to opt-into a separate system like Contributor does.
ssfrr··on Prototyping a 3D light field display video projector array
Even for a single user you need to add head tracking to get a perspective that shifts as you move your head (though that's pretty doable with consumer stuff these days). For a multi-user system I think you need something more like this.

It could be interesting to do multi-user individual-perspective shutter-based 3D, but you'd need to run at a really high framerate and you'd probably have brightness issues as each eye would only be receiving light for some smaller fraction of time.

ssfrr··on Prototyping a 3D light field display video projector array
If you're interested in that you may also like Matt Hirsch and Gordon Wetzstein's more recent work from the MIT Media Lab (they cite this paper):

http://web.media.mit.edu/~gordonw/CompressiveLightFieldProje...

Similar concept optically (as far as I can tell, it's not really my field) but solves some of the practical implementation issues (particularly the one-projector-per-pixel problem).

ssfrr··on Blown-up brains reveal nanoscale details
It's nice to see some of the less flashy and HCI-focused Media Lab research getting some coverage. There are some great things going on in Ed's lab.
ssfrr··on Why we need to raise Logic’s price
It's great to see a company so transparent about their challenges and what they're doing to address them.

I can also whole-heartedly recommend the Saleae logic analyzer, it's saved me literally hundreds of hours over the past few years, and keeps getting better.

← PreviousPage 4 of 5Next →