roye

one more, from google: https://ai.googleblog.com/2020/07/announcing-scann-efficient...

Academic example but I'm sure it's made it's way to this industry as well: in bioinformatics, this has proven tremendously useful for approximate string matching to do things like identify microbes in sequenced microbiome samples - cf https://genomebiology.biomedcentral.com/articles/10.1186/s13...

how do you compare with substack?

I wonder how driverless cars will effect this. Similarly I'm wondering if ride-sharing apps already are. Single cars picking up 4 adults that would otherwise be driving could potentially hugely reduce the number of cars on the road, making commutes shorter for all involved.

Long reads from PacBio, Oxford Nanopore, and 10x are also exciting. This new tech coupled with HiC for scaffolding and single cell sequecing brings up the possibility of complete knowledge of your genome, the collection of strains in a metagenome, or all the types of cells in a tissue/cancer sample.

Thinking along these lines I also found this: http://reversedictionary.org/

I saw stars indicate difficulty; do colors also have some meaning?

there's also a cornell-technion joint venture called the "Runway startup postdoc" aimed for new PhDs to transition to being founders: http://tech.cornell.edu/programs/startup-postdocs

strange that Lytro isn't mentioned anywhere, given they developed a product that works similarly.

The more relevant continuation of Heng Li's posts is Pall Melsted et al's work on GFA: https://github.com/GFA-spec/GFA-spec

This is pretty close to my area...

The main one is vg, which I think Benedict Paten is actively supervising, along with Richard Durbin. The main developer is Erik Garrison. I'm pretty surprised it wasn't mentioned by name: https://github.com/vgteam/vg

Actually there are new approaches to study exactly that, and they are developing rapidly, cf: https://en.wikipedia.org/wiki/Chromosome_conformation_captur... Basically, this problem, along with a long list of other applications is being attacked by deep sequencing. The way it works is that you apply a treatment to DNA that 'glues' 3D contacts in place and covers the glued segment, apply restriction enzymes to cut out what's not covered by glue, get rid of the glue, sequence all that's left, and then map the remaining sequenced fragments to the reference genome. The output is the relative tendency of different regions to come into contact with each other.

It could be that the standard for moonshots changed with the formation of Alphabet. That's the impression I got from yesterday's press coverage.

The mention of groups of robots learning together by exposing themselves to the real world and sharing lessons is very interesting to consider. I wonder if this is inspired by ants searching for food and leaving scent trails. Are there success stories of this approach in robotics already?

Sounds like an accurate description of a ponzi scheme.

I don't think their intent is to replace existing database servers. My impression is that they have copied the data onto their servers to make analysis on their cloud service more attractive. For example, the fact that the 1000 Genomes data is there saves researchers from having to go to the trouble of uploading it themselves.

same here, had to watch some of those lectures at 2x speed

Outside of DBs, indexes have shown themselves to be extremely useful in string problems in bioinformatics (my area of research). The modern workhorses of this trend are the Burrows Wheeler transform + FM index (together) and Bloom filters (also mentioned in the link). These have been applied to sequence alignment, de novo assembly of genomes, and compression of sequences. I posit the same bag of tricks can be applied in the NLP/machine learning settings, but I know less about how commonly they have been.

There's a good set of notes touching on this stuff here: http://www.langmead-lab.org/teaching-materials/

Twitter is actually becoming quite popular in science. It's a great tool to hear of new papers, find out about new findings reported in conferences you can't attend, and communicate in a public forum. Don't think you would have had much of that in MySpace...

It seems like all the hash implementations I've seen have an insertion rate of ~Million/sec. I wonder if it's possible to get at least an order of magnitude faster (single threaded). Are we close to the input string parsing at this rate? Would sorting/caching help a lot?

I found this set of lectures to be really nice: https://www.youtube.com/playlist?list=PLT_GI2oINs-CRuwMGDa5_... Wish I had time to get through them all...

Despite the title, I found the details regarding rodent brains and the kiss and spit mechanism much more engaging and convincing than the discussion of effects on humans.

I didn't get much from the linked blurb, but the title made me think: if you sample training data exhaustively and then run a learning process, could you then take the training data maximizing algorithm performance to then better train humans at the same task?

That's what struck me about it as well. I think things like this along with body/health tracking will be the true IoT killer apps

sounds like Cython. Are any such optimizations planned for 3.5 (or beyond)?

This was discussed two weeks ago (https://news.ycombinator.com/item?id=8779532). The title was changed at the source since it was originally too bombastic.

I was also thinking along these lines as I read, and there have been plenty of articles posted on HN on the diminishing returns of working too long. Basically it boils down to people getting tired, and especially in creative pursuits, it helping to get away from your project and let your mind wander sometimes.

What I expected the author might say is that with many accumulated hours you become more efficient such that over time your output per hour grows, but his view seems quite different.

Nice development for trivial patents, but it makes me wonder if this could lead to people having to defend their (e.g., algorithm) patents by appeal to computational complexity or the physical constraints of human vs. computer memory, etc.

You reminded me of this: https://news.ycombinator.com/item?id=7771877. (Earlier post of mine, but on topic...)