HNHacker News
TopNewBestAskShowJobs

javierluraschi

402 karma · joined March 23, 2014

Founder at hal9.com working on cloudish.ai

Previously RStudio, The Linux Foundation, Microsoft Research, and SAP.

x.com/javierluraschi github.com/javierluraschi linkedin.com/in/javierluraschi

submissionscomments
javierluraschi··on NNCP: Lossless Data Compression with Neural Networks
No, I meant this comment: "His relatively simple LSTM-based neural net compressor is 10x faster than one of the more complex state of the art algorithms"
javierluraschi··on NNCP: Lossless Data Compression with Neural Networks
I mean, regarding algorithm speed.

From the conclusion:

“We presented a practical implementation of an LSTM based lossless compressor. Although it is slow, its description is simple and the memory consumption is reasonable compared to compressors giving a similar compression ratio. It would greatly benefit from a dedicated hardware implementation.”

I only skimmed through the paper, but looks to me like this might be a 10X improvement over other LSTM compression methods, not state-of-the-art compression.

That said, I’m far from being an expert in compression and was hoping from someone in HN to explain how relevant this is in their discipline.

javierluraschi··on NNCP: Lossless Data Compression with Neural Networks
Where is the reference to the 10X improvement? As in, which benchmark are you comparing against? Thanks!
javierluraschi··on Lego Mosaics Using R
Got your back! I've been working on a JS voxels side project which I just made available for R: https://github.com/javierluraschi/voxels

I still need to add support to add voxels from R and retrieve them, but should be straightforward, the hard part was putting together the rendering engine.

javierluraschi··on StackOverflow Is Down
"Stack Overflow is currently offline for maintenance Routine maintenance usually takes less than an hour. If this turns into an extended outage, we will tweet updates from @StackStatus or post details on the status blog."
javierluraschi··on Meritocracy doesn't exist, and believing it does is bad for you
What’s the alternative?
javierluraschi··on sparklyr 1.0: 40x speedup for R on Spark
"Sparkly R"
javierluraschi··on Ask HN: Which tool do you use to create beautiful diagrams?
I like using http://nomnoml.com, mostly, since its intuitive syntax allows you to create nice diagrams in only a few seconds.

I built an R package to easily support nomnoml in R Markdown: https://github.com/javierluraschi/nomnoml

javierluraschi··on Software development requires servant leaders
I think the difference in software engineering is that the cost and time of digging a tunnel is almost zero. When you write code that "digs a tunnel", writing the design and code are the hard part, distributing the compiled code is usually predictable and simple. So I think software engineering is closer to the planning phase in civil engineering, where you need to go research the mountain, plan multiple routes, deal with regulation, etc. I bet it is also hard to estimate in civil engineering how long it's going to take to complete the plan to build a tunnel.
javierluraschi··on Ask HN: What should the technical design for a front end feature include?
I would say: Controls, events, views, view-data bindings, data-server bindings, state, authentication, telemetry and error handling.
javierluraschi··on AI Has a Hallucination Problem That's Proving Tough to Fix
Intially, kids don't learn language that fast, they spend a whole year getting samples from parents where we try-and-try-and-try over and over to get them to say something, so there is a high sample rate going on for sure. However, it is also true that humans learn faster at some point by using self other tools, like consciousness. Not sure how exactly that works on a toddlers brain, but on mine, if you ask me to remember a phone number, I will repeat it in my head several times and try to make associations, those higher-level learning processes seem to be the algorithms that we are missing to discover and implement successfully.
javierluraschi··on Show HN: Embedding Deep Learning models into your Slides using R
I had to look at the MNIST images and then try to draw the digits in the same way, the easiest ones are zero and one.
javierluraschi··on Show HN: Embedding Deep Learning models into your Slides using R
A lot of us post hacks under 'SHOW HN'. I honestly don't even know what's the practical application of using deep learning models within one's slides, take it for whats worth.

What is more interesting to me is that one can pull together this demo pretty easily using all the new tools that have just been made available in R. Those tools are much more interesting than this post.

Regarding, 'which is compounded by the nature of the former', the argument from the comment seems to be that creating a layer of something makes things worse, this is not necessarily the case. Abstraction layers are one of the most powerful and reliable tools in software engineering. For instance, I would much rather write a layer of C++ code that wraps assembly code. Others can choose to create layers over C++ and so on, it's all good.

javierluraschi··on Show HN: Embedding Deep Learning models into your Slides using R
This model is based on MNIST and trained over a few dense layers, there are much better ways to train this model. It would be interesting to build a web app that allows collection of a dataset of drawn digits over an HTML canvas that we can later train with a more sophisticated model, but that stretches a bit further than the time I have available.
javierluraschi··on Show HN: Embedding Deep Learning models into your Slides using R
Are you referring to the 'keras' R package as a hack? From your comment, it rather sounds like a disagreement in design principles, not a hack.

If you are referring to the act of mixing "rmarkdown, keras and kerasjs" to run deep learning models inside a slide as a hack, that would make much more sense.

Could you please clarify?

javierluraschi··on Show HN: Embedding Deep Learning models into your Slides using R
You can see all the slides under rpubs.com/jluraschi/deploying-tensorflow-rstudio-conf

Worth mentioning that this model uses a simple feedforward network with a few dense layers trained over the MNIST dataset; therefore, digit recognition classification is not very accurate.

javierluraschi··on New institute aims to make Toronto an ‘intellectual centre’ of AI capability
Is not that meaningful to compare salaries without considering cost of living.
javierluraschi··on Show HN: Math Worksheets for Kids
I was thinking of locking the UI to keep kids focused on finishing the worksheet before they ask for help to get a new one. But I agree it's currently hard to get back to get grid, I'll add support for navigating back to the grid to using the back button.
javierluraschi··on Show HN: Math Worksheets for Kids
You can now draw on phones and tablets by clicking on the worksheet, quite useful while on the go with kids or to save some paper.
javierluraschi··on Show HN: Math Worksheets for Kids
@tar11 thanks for the comment. The repeating questions are by design since I found out my kids get really motivated by starting with a few easy ones, ones they got the worksheet started, it's harder for them to stop. As the worksheets get harder, duplication becomes less common.

I'll look later today into adding a tablet version that will allow them do draw over to avoid printing them. Give it another try tomorrow.

javierluraschi··on Ask HN: Why are we building self driving cars to navigate based off human signs
1958 "self driving" car: https://www.youtube.com/watch?v=xKOdux6Gjno&feature=share - There is really nothing stoping someone from taking a different approach to driverless cars. Smart highways is not a bad idea at all. In the same way we have a carpool lane, one could provide a driverless interstate lane that is faster and more efficient by equipping this lane with a guiding system. #hyperlane
javierluraschi··on Ask HN: What do you use to make your presentations for talks?
http://rmarkdown.rstudio.com/revealjs_presentation_format.ht...
javierluraschi··on Ask HN: How to measure if it is “worth it” to accept a job with less coding?
I would say, life is too short, try doing what you love te most. Also, beware that making more money will encourage you to spend more money so you will feel the need to make more money, again. Maybe a better question is: Who do I aspire to be like and which decisions can get me there?
javierluraschi··on AlphaGo beats the world champion Lee Sedol in first of five matches
You can take a look at their paper... http://airesearch.com/wp-content/uploads/2016/01/deepmind-ma...

Basically, the ("lots of hardware") distributed implementation gets ~3100 points in the Elo rating against ~2900. ~2900 is still sufficient to win against Fan Hui. So I would say, that yes, this algorithm has most of the merit here.

javierluraschi··on AlphaGo beats the world champion Lee Sedol in first of five matches
What does a lot of hardware means? The human brain has about 100B neurons with about 100+ dendrites per neuron, while AlphaGo has about 1K CPUs with about 2B transistors per CPU.
javierluraschi··on Human-level control through deep reinforcement learning
Here is the marketing side of this publication in which Google scientists (aquihired from Deepmind) have developed a way to outperform humans in Atari games: http://m.phys.org/news/2015-02-hal-bests-humans-space-invade...
javierluraschi··on Human-level control through deep reinforcement learning
I think qlearning is really interesting, I posted yesterday a simple implementation/demo in Javascript of qlearning. This paper goes way beyond qlearning by deducing states based on a deep neural network from the actual game rendering, really cool. Regardless, as a first intro to qlearning I had fun putting this together https://news.ycombinator.com/item?id=9105818
javierluraschi··on Show HN: AI QLearning Robot Simulation in JavaScript and WebGL
Do you have a library for your project? It sounds really interesting. I think this space is interesting enough that would benefit from a GutHub project to collaborate.
javierluraschi··on Show HN: AI QLearning Robot Simulation in JavaScript and WebGL
I've gone as far as ~1000 food ~100 poisson blocks to make sure it's on the right track. Speeding this up is a good idea, I'll make some time to add this and post an update. You can ping me on twitter @javierluraschi.
javierluraschi··on Show HN: AI QLearning Robot Simulation in JavaScript and WebGL
Currently, it sees a board of 3x3 states, as in:

_ _ G

G B B

_ _ G

A next step would be to use real computer vision over a 3D camera in the environment aligned to R2D@ and a deep neural network to interpret pixels instead of states. DeepMind has a paper describing this approach: http://arxiv.org/pdf/1312.5602.pdf

← PreviousPage 6 of 7Next →