legel

12 ms·

91.
Mesh-TensorFlow: Model Parallelism for Supercomputers (TF Dev Summit ‘19)(youtube.com)
3 points by legel 8y ago|1 comments
92.
by legel 8y ago
Alan Kay’s YC Startup School lectures “How to Invent the Future” really influenced me, wonderful humanitarian wisdom: https://www.startupschool.org/videos/11 It was also a privilege conversing with him via email about
93.
A New Test for the Leading Big Bang Theory(quantamagazine.org)
2 points by legel 8y ago|0 comments
94.
WeGrow: A School by WeWork and BIG(wegrow.com)
3 points by legel 8y ago|0 comments
95.
Kencko wants to help you eat more fruit and vegetables(techcrunch.com)
4 points by legel 8y ago|1 comments
96.
by legel 8y ago
Looks neat and wonderful. It’s always impossible to compress all structure from high dimensions to 2D or 3D via something like PCA or t-SNE, and the focus on “geometric” insight is also encouraging. Cool/appropriate name too, nice choi
97.
by legel 9y ago
"GDPR doesn't penalize or prevent innovation. It just forces it out into the open." This is a nice summary.
98.
by legel 9y ago
This is an important update for practitioners: beware of Adam, it has fundamental flaws. In practice, carefully tuned learning rate schedules for basic SGD, as discussed in this, has gotten me much better results on very large very complica
99.
by legel 9y ago
This looks like a gem, thanks! Under my elbow as I type is a small introductory book called Topology for Computing, which I’d recommend to anyone trying to investigate such links, for its concise clarity and elegant figures. https:/&#
100.
Information, Physics, Quantum: The Search for Links (1989) [pdf](cqi.inf.usi.ch)
64 points by legel 9y ago|17 comments
101.
by legel 9y ago
Having done a lot of data mining and analysis of Wikipedia (scanning its entire corpus several times, machine learning new representations, studying traffic and article growth stats, studying its ontological structure) I would like to make
102.
by legel 9y ago
Thanks for the edit, I was wondering how that would make sense otherwise. More generally, thanks for clearly communicating what's really going on! Math can only be extremely beautiful if it is actually understood.
103.
by legel 10y ago
I think it depends on what type of researcher you are. In every field there are always authoritative leaders who are comfortable writing "survey papers", which is perhaps most comparable to what the "research distiller"
104.
by legel 10y ago
The post is clear that A.I. startups in this vertical will be given special resources, but not clear if there will be more startups selected specificially for this?
105.
Changing Grid: Exploring the Impact of the High Line(streeteasy.com)
1 points by legel 10y ago|0 comments
106.
Ask HN: Best way to represent HTML for deep learning its structure and content?
2 points by legel 10y ago|0 comments
107.
(No) Roads and “Sidewalk Labs”(medium.com)
3 points by legel 10y ago|0 comments
108.
The Guardian view on machine learning: people must decide(theguardian.com)
2 points by legel 10y ago|0 comments
109.
by legel 10y ago
This is a deeply appreciated update on the state of the art from Christopher Olah and the Google Brain team, including great insights into the nature of engineering attention. I'd be curious to understand more about how making so many
110.
by legel 10y ago
I just published technology that some researchers called an 'AI pitch' about 'magic AI does everything' ( https://news.ycombinator.com/item?id=12167097#12171264 ), even while they graciously pointed out so
111.
by legel 10y ago
~6% sparsity + the way that you think about data representation is very interesting. I will have to have a shot at running the autoencoder over the Mikolov 3m word corpus. My goal is to get the first 1m words compressed at under 100 MB zipp
112.
by legel 10y ago
This is an interesting approach. The idea of overlapping clusters is appealing, you're probably familiar with the LDA family of algorithms that provide nice data structures for this.
113.
by legel 10y ago
Thanks for the 'really good tech' comment, I'll take that on HN any day. I'm sure we could agree on many structural differences in human and machine learning, and in that sense, I'd agree that it makes sense to be c
114.
by legel 10y ago
Wow - 3 bits per element. Wonderful work with the quantization, I'd love to implement this. How would you recommend I proceed?
115.
by legel 10y ago
The message is genuine and simple: computers can now learn like humans.
116.
by legel 10y ago
That sounds awesome, and technically achievable in various capacities. Something a colleague thought of recently that may be of interest is hierarchical mapping, particularly if the number of personalities is great - so, e.g., having a map
117.
by legel 10y ago
Thanks, shoten looks wonderful and exciting. You're welcome to reference/reuse words2map as you wish, while the authors of word2vec, t-SNE, HDBSCAN, and others deserve all the credit. :)
118.
by legel 10y ago
Thanks for pointing the memory optimizations out. The Google model trained on 100 billion words unzips as a 3.4 GB binary, which often is a non-starter on smaller servers, while these vectors indeed come in at less than 100 MB. Mainly this
119.
Show HN: Best way to distribute big, complex Python projects regardless of OS(stackoverflow.com)
4 points by legel 10y ago|0 comments
120.
by legel 10y ago
The most profound thing I know about cities + public policy: Google hosted a talk in 2014 by MIT Professor Sandy Pentland called "Social Physics: How Good Ideas Spread". His research concludes that our connectivity - free and fas

More ›