thesz

53 ms·

841.
by thesz 5y ago
>This proves less dense networks can be more intelligent. Case in point: https://openai.com/blog/block-sparse-gpu-kernels/
842.
by thesz 5y ago
If you have an SQL table with a couple of indices, you have B-tree with random inserts/updates. For truly random data such as scale free graphs, B-tree shows quite non-linear performance decline, as if somehow it started to have O(N^2)
843.
Bluespec hardware design language and simulation tools(github.com)
1 points by thesz 5y ago|0 comments
844.
by thesz 5y ago
I believe that you should have a license for ARM ISA to implement it and sell chips. I also know that AMBA, with AXI and other popular bus designs from ARM are free to use without ARM CPU in the system and should be separately licensed when
845.
by thesz 5y ago
300ns is a latency of PCIe, not DRAM. DDR4 CAS latency is 15 clocks, which is about 9ns for highest clock frequency. Add to that cache latencies and memory controller latencies and you easily can get about 25-50 ns of latency and more. And
846.
by thesz 5y ago
Exactly. CGA has special point e_{inf} which is infinitely far away from any other regular point (with regular euclidian coordinates) and linear movements are rotations around that point.
847.
by thesz 5y ago
I cannot help but add a reference to a conformal geometric algebra here: https://en.wikipedia.org/wiki/Conformal_geometric_algebra Why? Because it describes movement as rotation from the infinitely far away point (a sp
848.
by thesz 5y ago
Artificial sweeteners also prompt insulin rise and promote insulin resistance: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7014832/ So, yes, if it is not water, with or without CO2, it should be ingested wit
849.
by thesz 5y ago
I guess an old Soviet anecdote is very relevant here: US citizen: I live in a free country! Nobody will prosecute me if I go next to White House and shout out loud "<name of US president>! You are an old stupid son of a bitch!&qu
850.
by thesz 5y ago
But they have significant decrease in heartbeat - about 5 beats per minute. For me, the difference between 57 and 62 beats per minute (with all other conditions being more or less equal) is a difference between me being healthy and having a
851.
by thesz 5y ago
Depends on the compute tasks. Even OpenMP has "dynamic" scheduling pragma for loops, in the case each loop iteration takes different time to complete. "async" is a way to address that, sorta.
852.
by thesz 5y ago
Tcl vaues do not have null and they are immutable - value assigned to a variable cannot change unless variable gets reassigned. Lists are immutable, for one example, even if they can be shared. These two properties make Tcl much more suitab
853.
by thesz 5y ago
This is what was done in the EROS [1] (extremely reliable operating system) UIs - it was not possible for user window to be rendered completely undistinguishable from system windows like, you knew it, a password prompt. [1] https:/&#x
854.
by thesz 5y ago
Please bear in mind that 3nm is an "alpha" - a multiplier characterizing the process. If you took your time and find out difference between processes with different "alphas" you will find that transistor density is nowhe
855.
by thesz 5y ago
These 1-bit-per-coefficient neural nets need pretty good floating point implementation to train them. From what I remember, they are trained with rounding - basically, a floating point weight gets rounded to -1, 0 or 1 and computed floating
856.
by thesz 5y ago
Compare Fourier (O(NlogN)) and wavelet (O(N)) transforms. Both can be used for similarity (correlation) search through projections.
857.
by thesz 5y ago
Rank based correlation coefficients: https://en.wikipedia.org/wiki/Correlation_coefficient#Rank There are several such algorithms. Also, there is a comment with a link to paper comparing this correlation coefficient to
858.
by thesz 5y ago
Yes, exactly. There are so many to choose from. ;) There are several discussions here in HN about PRNGs of different kinds and people still want to invent new ones. Why? Because old ones do not satisfy them in full. Returning to the problem
859.
by thesz 5y ago
Your point is very valid one. I was not able to find out how to keep ranks updated efficiently.
860.
by thesz 5y ago
I should point that Pearson's correlation coefficient can be updated on-line, the storage complexity is O(1). The storage complexity for sorting is O(N). Thus, this particular algorithm will not fit into, say, traffic analysis framewor
861.
by thesz 5y ago
You can update Pearson correlation online, you cannot update online OP correlation coefficient. Pearson and other correlation coefficients are linear, O(n), sorting would incur logiarithmic multiplier O(NlogN). Thus, it is not computational
862.
by thesz 5y ago
Here it is: https://mailman.science.ru.nl/pipermail/clean-list/2002/0021... The code I've asked contains two calls instead of one, aside of that it was quite simple and would allow at least Russian users
863.
by thesz 5y ago
As I tried both Clean and Haskell, for me Clean is "worse Haskell". As one example, Clean did not support non-latin1 alphabets in the UI library and my attempt to somehow correct the situation was in vain, the next release of the
864.
by thesz 5y ago
I think that if China has technology to break any [1] encryption with 256 bit key, they should have technology [2] to break encryption with longer keys. [1] Article does not specify encryption algorithms, so we should assume "any encry
865.
by thesz 5y ago
Mother's exercise during pregnancy enhance baby's brain development: https://www.newscientist.com/article/mg22029433-000-exercise... I've checked out my year averages for 2019, 2020 and 2021, they were 5
866.
by thesz 5y ago
Unless there is an entity or coalition of goal-aligned entities with 68% of stakes or 51% of hashing/PoW power. Then one suddenly finds oneself with a pretty much mutable world. It appears it has been shown many times [1] that such att
867.
by thesz 5y ago
Accesses can be distinguished too - read and write accesses are different. You can see this in [1] - there are Reader and Writer monads, where Reader is used for effectively unchangeable parameter to a monadic context (for example, socket)
868.
by thesz 5y ago
And you should use the type "Log m => String -> m ()" instead of general IO. This way, for one example, you can log in some State-derived monad, keeping logs for caller to extract and use. It is as safe as pure function. And
869.
by thesz 5y ago
Both Dali and Edison did not have families and children to rise. They were multimillionaries also. If you are also childless and multimillionaire, you can experiment with what essentially a sleep deprivation, having hallutinations that are
870.
by thesz 5y ago
If I may, I would like to tell you a couple of anecdotes from a colleague of mine and myself, which lead me to notice what is going on (and made fun of up there). That former colleague of mine, being student, once had to make presentation f

More ›