127 Matching Annotations
  1. Last 7 days
    1. This year my intention was to have fun, to not plan anything in advance and to just see what would emerge organically.

      I believe you were able to do that precisely because you were prepared. The more you prepare, the more beneficial letting go of the plan is .

  2. Oct 2022
  3. Jul 2022
    1. if we've looked at the synthetic a priori of time he also addresses the synthetic a priori of 00:15:46 space and one very interesting distinction he draws is between animals who have semi-circular canals in their heads somewhere and animals which don't now we do fishes do 00:15:58 limpets don't for an animal with semicircular canals the word this the nature of space is to have right left up down forward backwards it's to be suspended or the 00:16:11 head to move through a volumetric space that we is so familiar with us we think of that as space as simply existing an animal not so endowed with no 00:16:24 semicircular canals encounters space in an entirely different way and he discusses how space arises for something like a limpet or a paramecium so he has taken the synthetic a prioris 00:16:36 of time and space and reconsidered them in a manner appropriate to the 1920s and 30s paying keen attention to the structures and processes of the body 00:16:48 this is work that we still need to do

      Uexkull also studies how animals synthetic apriori sense of space differ between species. Humans and other animals have semi-circular canals in the ear and this helps then determine forward/backwards, up/down and left/right of volumetric space. Limpets and paramecium do not have such a semi-circular canal and therefore do not sense "3 dimensional space" the way that we do.

    2. he's built a little toy he's built a treadmill for a snail isn't that wonderful he's a genuine scientist he's doing lots and lots of experiments with 00:11:27 animals they're very creative experiments in this case he's built a treadmill for a snail so the snail is held by a vice on a rotating ball and the snail is then um approached by 00:11:40 the investigator who chucks it under the chin like this and if you chuck the snail under the chin like this the snail will recoil not surprising i 00:11:52 would recall it as well but as you speed up the frequency of these chucks under the chin at about five hertz once you pass a frequency of about five chucks per second 00:12:05 the snail's behavior changes remarkably instead of being perturbed and trying to withdraw it tries to crawl onto onto something the qualitative nature of what's happening 00:12:18 to the snail has changed for the snail and its response or it's it's um sense making is altered and it's tried it perceives seems to perceive now a constant surface onto which it might 00:12:32 crawl now that might seem strange to you but i'll remind you that if we flash a light for you five times a second you'll see a light flashing and if we speed up the interval 00:12:44 then we shorten the interval between flashes to make them faster there comes a critical point at about 20 site flashes per second where you no longer perceive individual flashes but you 00:12:54 perceive a continuous light something like this underlies the magic that happens with moving pictures as well where you know that the action you see in the cinema is a bunch of projected still pictures 00:13:09 but they um have this character of continuous movement for you likewise in sound if we play that for you you hear a bunch of disconnected claps but if we shorten the 00:13:22 interval between the claps and speed it up there comes a point at which it changes into a continuous low pitch and that happens again about 20 hertz at about 20 cycles per second now for this snail 00:13:35 that border is at a different place it's at about five cycles per second what this shows is quite profound remember kant's synthetic a prioris 00:13:46 time for this snail is different than time for you the time that arises as a function of the body of the snail has this border at about five hertz where you have one at about 20 hertz 00:14:01 his basic insight is that worlds arise for snails that are not commensurable with worlds that arise for humans with which are not commensurable with worlds that arise for earthworms 00:14:15 the notion of an umvelt we get to determine a minute is used to describe this bodily specific arising of a world together with time and space 00:14:28 now i said it's rather weird to think of time being fundamentally different for an animal of a different constitution but i'll remind you that we can use our cinematic tricks to make ourselves aware of our own 00:14:41 limitations on the left there through high-speed photography we managed to make perceptible an event which we cannot otherwise see the event that you see there with the splash is 00:14:53 perfectly real but we can only make it manifest through high-speed photography similarly whoops there are processes going on around us 00:15:05 that we do not perceive and we can use time-lapse photography to make those to speed them up so that they become perceptible to us something like the blooming of a flower or the battles intricate battles fought 00:15:16 between brambles and hedges these make us aware that we perceive time as unfolding at a rate dictated by our own metabolism and bodily processes so this idea that time and space 00:15:32 are considered very different from count but very much tied now to the body this is quite radical

      Here, the speaker, Fred Cummins, introduces us tto the synthetic apriori concepts of time explored by Uexkull.in his clever snail experiment.

      By holding the snail in place on a rotating vertical wheel and stimulating the neck of the snail by touch, by speeding up the frequency of touching the snail to about 5 Hz, Uexkull was able to produce a different behavior in the snail. The snail was no longer withdrawing its neck into its shell, but tries to walk instead.

      Cummins compares this unique sensing of time unique to the snail with that of humans. We perceive individual sounds and individual images as distinct as long as they occur at a frequency below approximately 20 Hz. When the frequency rises above this, we perceive it as continuous. This is how we digitize audio (moving sounds) as well as video (moving pictures), creating the illusion of continuous motion.

      So we, in effect CONSTRUCT the sense of time and motion. Jay Garfield talks about how we also construct aspects of reality such as color: https://hyp.is/go?url=http%3A%2F%2Fdocdrop.org%2Fvideo%2FHRuOEfnqV6g%2F&group=world

      Color and time are constructed based on the organisms specific perceptual structures. The snail constructs time differently than a human does.

  4. Apr 2022
    1. Alex Sigal [@sigallab]. (2021, December 7). We have completed our first experiments on neutralization of Omicron by Pfizer BNT162b2 vaccination elicited immunity Manuscript available at https://sigallab.net and should be available on medRxiv in the coming days [Tweet]. Twitter. https://twitter.com/sigallab/status/1468325159501287434

    1. ReconfigBehSci. (2021, February 1). @MaartenvSmeden @richarddmorey 2/2 Having conducted experiments on lay understanding of arguments from ignorance, in my experience, people intuitively understand probabilistic impact of factors, such as quality of search, that moderate strength. Rather than build on that, we work against it with slogan! [Tweet]. @SciBeh. https://twitter.com/SciBeh/status/1356228495714746370

  5. Mar 2022
  6. Jan 2022
  7. Oct 2021
  8. Sep 2021
    1. Just for fun, I did a little experimenting at home to show how some of these different types of cement hold up. I started by cementing a bunch of materials together with a bunch of different types of cement. I waited 24 hours, then cut each one roughly in half, down the middle.
    1. “From the culture’s point of view, Adler was a dead white male who had the bad luck to still be alive.”

      This is a painful burn by the writer Alex Beam.

      Perhaps worth modifying for Donald J. Trump?

      From the perspective of the American experiment and the evolution of democracy, Donald J. Trump was a dead white male who had the bad luck to still be alive."

  9. Aug 2021
  10. Jun 2021
  11. May 2021
  12. Apr 2021
  13. Mar 2021
  14. Feb 2021
  15. Jan 2021
  16. Oct 2020
    1. He has failed to do that

      According to Erza, the results of the experiment were only half of those that were expected. It may be possible of course that this is all that took place during that night. Someone else could have been present during Franklin's 'sleep walking' phase, just like how Franklin was observed during this experiment by Erza, and took the stone when Franklin tossed it to the ground? Or perhaps this outcome goes to show a more realistic point of view, in that it would have been actually impossible to replicate the whole night entirely, and that even the fact that they were able to reproduce it to such an extend is in itself an impressive accomplishment.

  17. Sep 2020
    1. Leicester UCU en Twitter: “Universities are conducting an experiment, an experiment that involves human beings (university staff and students) and a life-threatening virus. But experimental subjects must give informed consent. (That's basic research ethics.)” / Twitter. (n.d.). Twitter. Retrieved September 26, 2020, from https://twitter.com/leicesterucu/status/1309107917879156737

  18. Aug 2020
  19. Jul 2020
  20. Jun 2020
    1. he DBMS performs the best on both the low-contention and high-contention workloads with the Oracle/MySQLand NuoDB configurations. This is because these systems’ stor-age schemes scale well in multi-core and in-memory systems, andtheirMV2PLprotocol provides comparatively higher performanceregardless of the workload contention. HYRISE, MemSQL, andHyPer’s configurations yield relatively lower performance, as theuse ofMVOCCprotocol can bring high overhead due to the read-settraversal required by the validation phase. Postgres and Hekaton’sconfigurations lead to the worst performance, and the major reasonis that the use of append-only storage withO2Nordering severelyrestricts the scalability of the system. This experiment demonstratesthat both concurrency control protocol and version storage schemecan have a strong impact on the throughput.


    2. We also observed that the performance of a MVCC DBMS istightly coupled with its GC implementation. In particular, we foundthat a transaction-level GC provided the best performance with thesmallest memory footprint. This is because it reclaims expiredtuple versions with lower synchronization overhead than the otherapproaches. We note that the GC process can cause oscillations inthe system’s throughput and memory footprint.


    3. Lastly, we found that the index management scheme can alsoaffect the DBMS’s performance for databases with many secondaryindexes are constructed. The results inSect. 7.5show that logicalpointer scheme always achieve a higher throughput especially whenprocessing update-intensive workloads.


    4. We observed that the performanceof append-only and time-travel schemes are influenced by the effi-ciency of the underlying memory allocation schemes; aggressivelypartitioning memory spaces per core resolves this problem. Deltastorage scheme is able to sustain a comparatively high performanceregardless of the memory allocation, especially when only a subsetof the attributes stored in the table is modified. But this schemesuffers from low table scan performance, and may not be a good fitfor read-heavy analytical workloads.

      结论Version Storage

    5. Overall,we found thatMVTOworks well on a variety of workloads. Noneof the systems that we list in Table 1 adopt this protocol


    6. performance gap is enlarged to 40% with the number of secondaryindexes increased to 20.Fig. 23further shows the advantage of logi-cal pointers. The results show that for the high contention workload,the DBMS’s throughput when using logical pointers is 45% higherthan the throughput of physical pointers. This performance gapdecreases in both the low contention and high contention workloadswith the increase of number of threads

      索引管理:Logical pointer更优

    7. The results in Fig. 22b show that under high contention, logicalpointer achieves 25% higher performance compared to physicalpointer scheme

      索引管理:Logical pointer更优

    8. The results in Fig. 20a indicate that transaction-levelGCachievesslightly better performance than tuple-levelGCfor theread-intensive,but the gap increases to 20% inFig. 20bfor theupdate-intensiveworkload. Transaction-level GC removes expired versions in batches,thereby reducing the synchronization overhead. Both mechanismsimprove throughput by 20–30% compared to when GC is disabled.Fig. 21 shows that both mechanisms reduce the memory usage


    9. The results in Fig. 18 show thatCOOPachieves 45% higherthroughput compared toVACunder read-intensive workloads. InFig. 19, we see thatCOOPhas a 30–60% lower memory footprintper transaction thanVAC. Compared toVAC,COOP’s performanceis more stable, as it amortizes the GC overhead across multiplethreads and the memory is reclaimed more quickly. For both work-loads, we see that performance declines over time when GC isdisabled because the DBMS traverses longer version chains to re-trieve the versions. Furthermore, because the system never reclaimsmemory, it allocates new memory for every new version.


    10. the append-only and time-travel schemes’performance is stable regardless of the number of modified attributes.As expected, the delta scheme performs the best when the numberof modified attributes is small because it copies less data per version.But as the scope of the update operations increases, it is equivalentto the others because it copies the same amount of data per delta

      Version Storage比较modified attributes

    11. We use transaction-level background vacuuming GC andcompare the orderings using two YCSB workload mixtures. We setthe transaction length to 10. We fix the number of DBMS threads to40 and vary the workload’s contention level.

      Version Chain Ordering: N2O vs O2N

    12. We use theYCSBworkload mixtures inthis experiment, but the database is changed to contain a singletable with 10 million tuples, each with one 64-bit primary key and avariable number of 100-byte non-inlineVARCHARtype attributes. Weuse theread-intensiveandupdate-intensiveworkloads under lowcontention (θ=0.2) on 40 threads with each transaction executing 10operations. Each operation only accesses one attribute in a tuple.

      Version Storage: 标准setup

    13. MVOCCis more likely to abortNewOrdertransactions, whereas thePaymentabort rate inMV2PLis 6.8×higher thanNewOrdertransactions. These two transactionsaccess the same table, and again the optimistic protocols only de-tect read conflicts inNewOrdertransactions in the validation phase.SI+SSNachieves a low abort rate due to its anti-dependency track-ing, whereasMVTOavoids false aborts because the timestampassigned to each transaction directly determines their ordering


    14. MVOCCincurs wastedcomputation because it only detects conflicts in the validation phase.

      TPC-C: MVOOC still sucks

    15. there is not a great difference among the protocols exceptMV2PL; they handle write-write conflicts in a similar way and againmulti-versioning does not help reduce this type of conflicts

      MV2PL: WW Conflict

    16. Beyond this contention level, the performanceofMVOCCis reduced by∼50%. This is becauseMVOCCdoesnot discover that a transaction will abort due to a conflict until afterthe transaction has already executed its operations. There is nothingabout multi-versioning that helps this situation.

      MVOCC: reduced performance

    17. TPC-C:This benchmark is the current standard for measuring theperformance of OLTP systems [43]. It models a warehouse-centricorder processing application with nine tables and five transactiontypes. We modified the original TPC-C workload to include a newtable scan query, calledStockScan, that scans theStocktable andcounts the number of items in each warehouse. The amount of con-tention in the workload is controlled by the number of warehouses

      Benchmark: TPC-C -- nine tables & five transaction types

    18. YCSB:We modified the YCSB [14] benchmark to model differ-ent workload settings of OLTP applications. The database containsa single table with 10 million tuples, each with one 64-bit primarykey and 10 64-bit integer attributes. Each operation is independent;that is, the input of an operation does not depend on the output ofa previous operation. We use three workload mixtures to vary thenumber of reads/update operations per transaction: (1)read-only(100%reads), (2)read-intensive(80%reads, 20%updates), and(3)update-intensive(20%reads, 80%updates). We also vary thenumber of attributes that operations read or update in a tuple. Theoperations access tuples following a Zipfian distribution that is con-trolled by a parameter (θ) that affects the amount of contention (i.e.,skew), whereθ=1.0 is the highest skew setting.

      Benchmark:YCSB -- 独立操作,三个workloads

    19. For each trial, we execute the workload for 60 seconds to let theDBMS to warm up and measure the throughput after another 120seconds. We execute each trial five times and report the averageexecution time.


    20. We execute all transactions as storedprocedures under theSERIALIZABLEisolation level

      实验用的最高isolation level

  21. May 2020
  22. Apr 2020