HNHacker News
TopNewBestAskShowJobs

romaniv

2,750 karma · joined February 2, 2011

submissionscomments
romaniv··on Google praises 86-year-old for polite internet searches
Think about how much easier the whole thing would be if there was one button (one!) that switched between "entry" and "command" modes when you're holding it.

Alternatively, it could be some kind of sound that's not common to English. In fact, your could make up a whole command structure based on sound "gestures". It would sound funny, but the ease of use would trump those awkward verbal commands. The sounds could mimic natural sounds for mnemonic ease.

romaniv··on Improved Techniques for Training GANs – OpenAI's first paper
This is one of many examples why claims regarding near- or super- human performance in AI papers need to be taken with a good amount of salt. CIFAR-10 is great for experimenting with your algorithms, but it's a horrible dataset to do any kinds of human-to-machine performance comparisons.
romaniv··on CS188 Intro to AI – Course Materials
Winston is, indeed, an amazing lecturer. And he has extremely interesting side comments.

In one of his lectures there he pointed out that research papers with a single good idea are more likely to succeed (be cited and discussed) than a paper with multiple good ideas. After thinking about it for a while, I realized that this was the reason some of my personal projects failed to gain traction. It's a frustrating realization, but at least he has some concrete, easy to implement advice on how to overcome this issue.

His comments on the history of AI field are also very interesting.

romaniv··on Computer Vision Research: The deep “depression”
I don't think this an adequate analogy. SVMs never completely captured the field. There were also random forest, boosting and other promising algorithms that delivered results "in parallel" (within the same time-frame).

Also, the attitude that ANNs are "all we need" was there even before deep learning delivered all the recent SOTA results. I remember Patrick Winston commenting on that in one of his lectures in 2010(!).

romaniv··on Computer Vision Research: The deep “depression”
In practical terms, performance without understanding can lead to highly surprising/counter-intuitive results when algorithms are applied to real-life problems. This doesn't matter much if you're doing movie suggestions or something like that, but it does matter in many other areas that could benefit from AI/ML.
romaniv··on Computer Vision Research: The deep “depression”
I had very similar thoughts recently. Really glad someone took time to express them properly.

Even on this very website... I feel there is an immense bias in favor of anything "deep" and "neural" when it comes to AI. Can't recall any recent AI papers that made it to the front page without having those two words in the title.

And please, don't tell me there is nothing interesting going on in the field outside of deep learning. Even when an approach doesn't beat SOTA in terms of error rates, it can still contain valuable ideas or have interesting properties.

romaniv··on React Tutorial: Cloning Yelp
As far as DRY, the worst aspect of no-dependencies approach is dealing with AJAX. The rest of it is actually not that bad. After some experimentation, I separated the functionality into "behaviors", which are fairly reusable across different pages and projects.
romaniv··on React Tutorial: Cloning Yelp
>I'm not blaming react, everything else in the javascript space is just as bad right now. There's at least 4 different ways of loading modules (UMD, AMD, commonJS, es (babel) modules), 2 of which (require(), import .. from) are used within this example.

Personally, I got so frustrated with ever-increasing complexity of tools and growing networks of dependencies that whenever I can I try to package my JS code in self-contained modules/files that depend on nothing -- not even on one another. Needless to say, it's not an approach that's very popular, but it works remarkably well for websites of moderate complexity. One thing I'm sure of: using those modules feels far more satisfying than using a framework.

romaniv··on E Ink announces a full color electrophoretic ePaper display
Also, it would be cool not to have a "power saving" mode engaging and disengaging, taking time.
romaniv··on Pastejacking
The full reach of this issue might not be limited to just text. It might be possible to mess with scripts in programs like Word by abusing rich text, macros and styles:

https://www.youtube.com/watch?v=LoORMRbptTg

romaniv··on CSS coding techniques
BEM is extremely verbose:

  <div class="block block__element block__element--modifier"></div>
Is there some good reason people write that instead of just the following?

  <div block="element modifier"></div>
Unless I'm missing something, with the second approach you get the same specificity, higher flexibility and shorter and more readable markup. Moreover, you end up using selectors the way they're intended to use.
romaniv··on Preparing for the Future of Artificial Intelligence
It's hard to tell these days. Many people today fully accept the idea that human should adapt themselves to the existing machines and technologies, rather than design/adapt those machines and technologies to human needs.
romaniv··on Preparing for the Future of Artificial Intelligence
>There were expert systems that worked well and they became profitable businesses, but othervice there was little to show.

One could similarly say about our time: "there were neural network applications that worked well, but otherwise there was little to show". What is the fundamental difference between what is going on now and what was happening before the previous AI winter?

I feel that people constantly misrepresent how impressive expert systems seemed back in their heyday. They had a lot of practical applications and they could do some very cool tricks.

Interestingly, the highly impressive accomplishments of SVMs, random forests and boosting went mostly unnoticed, precisely because of the AI Winter.

romaniv··on An AI First World
Unfortunately, the ways AI researchers measure and compare their results to human performance are often rather weird and favor algorithms via.. let's say "unusual" experiment setup.

For example, face recognition performance is often measured by comparing two random faces taken from a sample set. In real life, the "problem" is almost always to match a random face against the entire set of faces you remember. Performance in these two tasks is not at all equivalent.

I've seen similarly weird problem formulations in many other AI research papers.

romaniv··on Future Screens Are Mostly Blue
I would buy a Japanese flip/smart phone in a second if it was properly supported outside of Japan. From what I heard, they have weird rules about warranty and even software updates outside of the country.

It's really annoying to be forced in the direction of fake technological "progress" that's mostly a result of extensive marketing. There is no fundamental reason why a smart phone can't have proper keys for calling.

There are couple of Android flip phones on the market, but since Android UI is designed 100% for touchscreens, I am pretty sure the experience of using them will be horrible.

romaniv··on SQL Tricks
"SQL is the worst form of database querying, except for all the other forms."

IMO, if you think of Prolog(s) as a database(s), that model is far more enjoyable to work with than SQL.

romaniv··on Have You Tried Using a Nearest Neighbor Search?
Yes, but how different would those 7K frames would really be? Same lighting, same background, same surrounding objects, the exact same condition of the vehicle's interior and exterior, same quirks of the camera's color profile, etc, etc. It would be an interesting experiment to actually try this, but I have a feeling the results wouldn't be all that good. Point being, you probably wouldn't get most of the benefits of deep learning and you might as well use the same approach the author used.
romaniv··on How Far to AI Utopia?
I thinks that's a very good question.

I can name countless cases of "semi-trivial" AI being used to improve people's lives. Tree/graph searches, linear regression and clustering algorithms are used all over the place: in science, engineering and medical field.

However, when it comes to more sophisticated algorithms I am hard pressed to point out strongly positive use cases. I.e. something that actually make a positive impact for large groups of people, isn't a gimmick and does not involves massive privacy violations.

Hm, I guess handwriting recognition is used by postal services. There are a lot of expert systems running behind the scenes in some industries (stuff that involves logistics, for example). Voice recognition helps disabled people to interact with computers. Machine translation can be useful at times. Although I've never been a situation when it made any significant difference in my life or in lives of anyone I know.

Stuff like face detection in cameras or voice recognition in phones and cars seem cool, but nothing would really change if they weren't available.

The most impactful uses of fancy AI that I know involves marketing and surveillance, and I don't consider either of those uses as a positive thing.

romaniv··on Tinker with a Neural Network in Your Browser
>Before deep learning people would manually design all these extra features sin(x_1), x_1^2, etc.

It's probably worth pointing out that this is true for ANNs, but there were (and are) other "shallow" classifiers that can handle swiss roll problem without manual parameter encoding. SVMs, for example.

http://cs.stanford.edu/people/karpathy/svmjs/demo/

romaniv··on Have You Tried Using a Nearest Neighbor Search?
> They're only 7k images.

I feel like all those deep learning papers distorted people's perception of scale. If you need to take those 7k images by hand because your application domain is obscure and they aren't available in an existing dataset, that's way beyond feasible.

romaniv··on It's time to dispel the myths about nuclear power
The consequence of a nuclear accident is much greater than the consequence of a wind power accident

Accidents are not the only things that cause damage, though. Pollution kills people too. Stuff like that should be factored in when considering risks.

http://www.washingtonpost.com/wp-dyn/content/article/2008/03...

romaniv··on When Is the Singularity? Probably Not in Your Lifetime
I was told that driverless cars and household androids would be commonplace by the year 2000. Different people predicted different things depending on the decade you're looking at.
romaniv··on When Is the Singularity? Probably Not in Your Lifetime
>That being said, I think we are finally at the dawn of "useful AI"

Were AI systems that outperformed humans in solving complex differential and integral equations useless? Were systems that diagnosed diseases better than doctors useless? Were systems that handled logistics at scale never possible before useless?

All of those were created in the past (60s, 70s, 80s).

The problem with AI was never the lack of results. It was the hype that far exceeded the results, however useful and practical they were.

romaniv··on Deep Networks with Stochastic Depth
I'm reading Delip's followup post[1] and it reminds me how much of ANN stuff is till pretty much alchemy.

[1] http://deliprao.com/archives/137

romaniv··on Music Language Modeling with Recurrent Neural Networks
In my opinion, anyone who works on music generation should take a look at Karma (http://karma-lab.com/) for a baseline of what can be achieved by simple math and plain old programming. Probably not particularly interesting from programmer's perspective (it's closed-source and to the best of my knowledge doesn't use anything fancy), but the end results are spectacular and used in real music.
romaniv··on Microsoft Cognitive Services
I have a bittersweet feeling about all these APIs. On the positive side, this is something we can take and use, possibly saving ourselves tons of time on some routine tasks. On the negative side, this particular flavor of AI-as-a-service from a big company like Microsoft means that algorithms will be reduced to a black box which you don't really understand or control in any meaningful way.
romaniv··on Houdini: Maybe the Most Exciting Development in CSS
Maybe we should just transpile Flash into ASM.js and be done with it.

And yes, I am being sarcastic here. I understand the motivation underpinning such projects, but the approach of introducing yet another complicated API with dubious cross-browser support is not something that will move the web forward in a meaningful way.

romaniv··on Tay – Microsoft A.I. chatbot
>It's not Microsoft's fault that the public thinks of Terminator robots when hearing "AI"

Large corporation definitely could do better when it comes to explaining the limits of their AI/ML technologies and putting those technologies in perspective (especially historic perspective). Scaling down on hyperbole, buzzwords and personification would help as well.

romaniv··on Akin's Laws of Spacecraft Design
It is remarkable that most organization try to "fix" this by teaching everyone to present (i.e. sell ideas) better, rather than teaching people how to better listen and evaluate new ideas.
romaniv··on Some Starting Points for Deep Learning and RNNs
Are you speaking about just the code samples or commenting on the general approaches used in the book?
← PreviousPage 13 of 30Next →