111 Matching Annotations
  1. Sep 2018
    1. notes (para 73)

      73 However, having regard to the general structure of Directive 2002/58, the factors identified in the preceding paragraph of this judgment do not permit the conclusion that the legislative measures referred to in Article 15(1) of Directive 2002/58 are excluded from the scope of that directive, for otherwise that provision would be deprived of any purpose. Indeed, Article 15(1) necessarily presupposes that the national measures referred to therein, such as those relating to the retention of data for the purpose of combating crime, fall within the scope of that directive, since it expressly authorises the Member States to adopt them only if the conditions laid down in the directive are met.

    2. joined cases C-293/12 and C-594/12
    3. joined cases C-203/15 and C-698/15
    4. amends

      Article 5.1:

      Any interference with electronic communications data, such as by listening, tapping, storing, monitoring, scanning or other kinds of interception, surveillance or any processing of electronic communications data regardless of whether this data is in transit or stored, by persons other than the end-users, shall be prohibited, except when permitted by this Regulation.

      (the emphasised text is the addition by parliament)

    5. clarifies

      Article 7.1:

      ..., the provider of the electronic communications service shall erase electronic communications content or make that data anonymous after receipt of electronic communication content by the intended recipient or recipients. Such data may be recorded, or stored or otherwise processeed by the end-users after receipt, or by a third party entrusted by them to record, store or otherwise process such data. Such processing shall take place in accordance with Regulation (EU) 2016/679.

      (emphasis denotes the addition by the Council)

    6. General Data Protection Regulation (GDPR)
    7. its report on the ePrivacy Regulation

      Found here

    8. current ePrivacy Directive
    9. the most recent draft text from the Austrian Council Presidency 12336/18

      Published here (direct link to pdf). Version is from 20 September 2018.

  2. Apr 2018
    1. At the time ofinstallation or first usage

      Bothering the user directly at installation or first usage may not be the ideal user experience. For privacy settings that become relevant only later on, the questions could at least be postponed to that moment.

    2. every update

      Horrible idea. Software could update every day. This requirement would discourage updating frequently.

      The requirement would make sense for updates that introduce new features/settings with possible privacy implications. In any case, simple fixes such as security updates should be exempt from having to bother the user.

    3. by a third party

      Are Google Analytics and friends sneaking in again? Are their enough restrictions on their use of the user's data?

    4. social networks

      "social networks" not counted as interpersonal communications services?! I suppose a particular conception of "social networks" is meant here, e.g. involving only publicly visible messages. However, just putting the general term "social networks" here without further specification seems very confusing.

    5. [Confidentiality of electronic communications data shall apply to the transmission of machine-to-machine electronic communications where carried out via an electronic communications service.

      The related recital 12 remains mostly intact, but with the removal of this item there is no mention of machine-to-machine communications in any article. Is that intended?

  3. Oct 2017
    1. The danger is if regulators begin to mistakenly view the Internet as a breeding ground for “natural monopolies” then they might regulate it as such

      But perhaps the Internet is a breeding ground for unnatural monopolies, and should be regulated as such.

    2. Unfair and deceptive practices should be dealt with swiftly and consumer choice should be encouraged.

      Unlike implicitly suggested here, encouraging consumer choice may in fact require quite some regulation. Vendor lock-in and network effects are prevalent patterns in the industry.

    3. Certainly Facebook and Google are difficult to compete against, but that stems from the quality of their products and the consistency of their innovation.

      At least in the case of Facebook, their closed silo combined with the strong network effect seems a relevant factor that cannot be ignored. Nowadays, people are probably not on Facebook for its product quality, but simply because others are there too.

    4. When was the last time you heard of a public utility starting up in someone’s garage?

      Rethorically pleasant, but a misleading statement. Until some years ago, you had probably also never heard of a billion-dollar company starting up in someone's garage. The productive potential of garages has risen astonishingly.

  4. Aug 2017
  5. May 2017
  6. Feb 2017
    1. window.external.AddSearchProvider(engineURL);

      Not supported in Chrome/Chromium anymore. See a discussion about its deprecation here: https://github.com/whatwg/html/issues/713

  7. Nov 2016
  8. Oct 2016
  9. Jul 2016
    1. // Appending to the enter selection expands the update selection to include // entering elements; so, operations on the update selection after appending to // the enter selection will apply to both entering and updating nodes.

      Not anymore since v4: use merge instead.

    1. To reduce duplicate code, enter.append has a convenient side-effect: it replaces null elements in the update selection with the newly-created elements from the enter selection. Thus, after enter.append, the update selection is modified to contain both entering and updating elements

      Not anymore since v4: use merge instead.

    1. same
    2. The

      Note that

    3. his the

      this is the

    4. ptimally some f

      optimally, some

    5. e only thelocal objective for the training of each layer

      is to train each layer using only its local objective.

  10. May 2016
    1. If you can not know that your message is going to be accurately transmitted without someone else degrading its integrity, then you have lost the right to "send any message" in public

      This mistakes the right to speak with the right to be heard, which, according to this very same article, you do not possess. The receiver chooses to not listen to the original message, but to a modified version.

    2. The reality is that while you have the right to speak, and the right to listen, you do not have the right to be heard

      While the author applies this argument to the speech of annotators, exactly this point is being ignored in the remainder of the article when it would apply to the web page authors.

    3. there is no basis for complaint

      Yes there is: lack of consent.

    4. web page authors have the right to present their material without "interruption" from people, even if the reader desires that interruption

      This claim reveals the essential question: should authors have the right to dictate the reader's viewing experience, against their desires?

    5. Without consent from both the web page reader

      Activating third-voice is consenting: the reader requests to hear others' comments.

  11. Apr 2016
    1. follow the Docker team’s instructions on how to install Docker

      In Ubuntu this should work:

      sudo apt-get install docker.io
      
    2. $ docker pull hypothesis/hypothesis

      You may have to add yourself to the docker group first (or use sudo docker...). E.g.:

      sudo usermod -aG docker myusername
      

      (and log out & in to let it take effect)

    1. If you wish to install using Ubuntu-managed packages, consult your Ubuntu documentation

      This should do the trick:

      sudo apt-get install docker.io
      sudo usermod -aG docker myusername
      
    1. If your desired username is already taken, we will be happy to help you acquire it for your account or channel, provided that you have that same username on at least two of these services: Facebook, Twitter, Instagram.

      Strange policy - "acquiring it for you" implies "taking it from someone else". To make a Telegram account and be sure others will not take it from you, you thus have to also check and create accounts on those other services.

  12. Mar 2016
  13. Feb 2016
    1. This calculation does not make sense to me. Besides assuming purely linear relations between cigarettes sales and population deaths, as well as between PM2.5 particles and population deaths, it also seems to imply the absurd assumption that every member of the population smokes the same amount of cigarettes. If they would take into account that only a quarter of the population they used smokes at all (seems about right), their resulting number of 36 would become 8 instead (and this does not even take into account that not all smokers smoke the same amount). Either I missed something, or this is just a completely meaningless calculation covered up by a smart-looking explanation.

    1. Living in cities like Beijing and Delhi is comparable to smoking over 30 cigarettes a day

      This claim lacks citation or calculation. If, as is more often assumed, microparticles (PM2.5) are the main culprit in air pollution, the claim sounds hugely exaggerated. Even with beyond-index pollution records of almost \(1mg/m^3\) in Beijing (see e.g. Wikipedia), total microparticle (PM2.5) intake would be: $$15,000 liter/day * 1mg/1000liter = 15mg/day$$ This would be a bit more than one cigarette per day, not 30.

      See this post or this Q/A for more elaborate discussions on this question.

  14. Jan 2016
    1. Bij gebrek aan verkiezingen is de partij genoodzaakt prestaties te leveren, anders dreigt ze de steun van het volk en daarmee haar mandaat om te regeren kwijt te raken

      Juist met verkiezingen lijkt dit me al helemaal het geval. Hoe precies zou verminderde steun van het volk de partij beïnvloeden?

  15. Dec 2015
    1. Tounderstandthissubtlebutpowerfuleffect,considerahor-izontalsheetoftinwhichisresonatinginsuchawaythatsomepartshavestrongverticaloscillationsandotherpartsaremotionless.Sandscatteredonthetinwillaccumulateinthemotionlessareaseventhoughthetime-averagedgradientiszeroeverywhere.

      Nice metaphor, but I wonder how well it matches, because sand has inertia and would jump around the surface. Would the metaphor be more accurate if the sand has no inertia and would not jump? I am thinking of marbles being squeezed between two sheets, rolling downwards wherever the sheet is at an angle.

    1. the person asking will give them the interesting part of the problem to solve

      Perhaps this is worth a reason number 4, or addition to number 3: Colleagues are warned about a problem they might encounter themselves in the future, and will know to come to you when facing a similar issue.

  16. arxiv.org arxiv.org
    1. not clear how it can be applied toReLU autoencoders

      But other sparsity measures than KL divergence are still applicable, e.g. penalising the sum of the absolute activities (\(\sum_i \left|h_i\right|\))

    1. You can even chat with your team

      chat to your team would be more accurate, as you will not hear anything back while offline.. (or does it do something magic?)

    1. a probabilistic model QQ that in some sense approximates the natural distribution of data PP

      Note that many sources use the letters P and Q the other way around.

  17. Nov 2015
    1. Bron: EVO & TLN

      Specifieke bron? Zo te zien tabel 3 uit dit rapport. Dat rapport berekent een de schade gebaseerd op schatting van het aantal voertuigverliesuren van vrachtverkeer, maal €42,92 per uur als 'value-of-time' (genomen uit een ongespecificeerde bron van het KiM).

    1. The first zero-waste grocery store in the world

      For example Granel in Barcelona appeared around 2011 and seems to have the same concept. Besides, a century ago there may have been plenty too.

    1. De angstaanjagende effecten van toenemende mogelijkheden voor surveillance kunnen leiden tot een afname in de vrijheid van expressie en van deelname aan de maatschappij (…) Vergevorderde online-marketingtechnieken die werken met digitale persoonlijke profielen worden nu ingezet om de publieke opinie en politieke verkiezingen te manipuleren. Decentrale applicaties bieden niet alleen de mogelijkheid om deze problemen aan te pakken, maar ook om de democratie te versterken.

      De geciteerde paragraaf:

      The chilling effect of a growing surveillance potential could also decrease expression and participation. The loss of monitorability and control could increase alienantion and also decrease participation. Sophisticated marketing techniques that rely on profiles of individuals are now being used to manipulate public opinion and elections. The potential exists not only for reversing these problems, but for increasing democratization. For instance, with the kinds of systems presented here, multiparty secure election and polling could be conveniently conducted without centralized coordination and with those expressing their views able to show relevant credentials.

      Uit: Chaum, David. 1985. “Security without Identification: Transaction Systems to Make Big Brother Obsolete.” Communications of the ACM 28 (10): 1030–44. doi:10.1145/4372.4373. (pagina 15 van artikel, 1044 van uitgave)

    2. White Paper (2008)
    3. Forbes-artikel uit 1999
  18. Oct 2015
    1. Het aantal gevallen van schildklierkanker, een kankervariant die direct in verband kan worden gebracht met radioactieve straling, is ook na vier jaar nog niet toegenomen

      Uit de samenvatting van een krities tegenrapport:

      "Until December 31st , 2013, 33 children have been diagnosed with thyroid cancer and 41 more have tumor-suspect biopsies, indicating possible malignancies. Although it is not possible to determine whether a single case is caused by radiation, as cancers do not carry a “label of origin”, the number of cases found so far is unexpectedly high. Japanese cancer statistics suggest an incidence of less than 1 case of thyroid cancer in this population per year."

  19. Sep 2015
    1. gelukszoekers (“happiness-seekers”)

      "geluk" means both happiness and luck; the translation "happiness-seekers" misses part of the meaning.

  20. Jul 2015
  21. May 2015
    1. [","]

      Is this superfluous? It's not there after parameter_list in funcdef.

    2. (defparameter ",")* | "*" [parameter] ("," defparameter)* ["," "**" parameter] | "**" parameter | defparameter [","] )

      Something wrong here. At least, one unmatched closing parethesis at the end.

  22. Apr 2015
    1. If a molecule is too big, you give it to the chemists. The chemists, for them, if the molecule is too big or the system gets too big, you give it to the biologists. And if it gets too big for them, they give it to the psychologists, and finally it ends up in the hands of the literary critic, and so on.

      As depicted by xkcd

  23. Mar 2015
    1. Summary of policy gradients using backprop (last ten minutes):

      The 'policy' is defined as the probabilities of taking an action given a history of observations: \(p(a_t | h_t)\), with \(h_t = (o_0, o_1, ..., o_t)\).

      Reward comes from each action as \(r_t(a_t)\), expected return is: $$ J(\theta) = E[ \sum_{t=0}^{T} r_t(a_t)] $$

      The gradient of the expected reward with respect to the parameters \(\theta\) (= "which direction should I change \(\theta\)?") is taken as follows:

      Sample (& average over) many action sequences (= play many games), for each sequence \((a_0,a_1, ..., a_T)\) computing:

      ---- The sum of, for each action \(a_t\) in the sequence:

      ---- ---- [which direction to change theta to make my action \(a_t\) more (log) probable given history \(h_t\)] * [the total reward gotten from this action and subsequent actions]

      The whole thing can be read as: if an action in a game led to high rewards, try to do that action more often when in the same situation.

    2. Too long; didn't watch: Starts with introduction to reinforcement learning. From 20:30 he starts formalising the problem, to derive in the last ten minutes (from 45:00) how to compute policy gradients using backprop - supposedly the same method used by DeepMind to learn to play arcade games.

    1. this.dependency.changed();

      Forgot to check for attribute's existence?

      if (this.dependency) {
          this.dependency.changed();
      }
      
  24. Feb 2015
    1. arg min

      min

    2. everywhere

      Does "everywhere" mean "everywhere in the training data", or really everywhere in the input space?

    1. see my (and only my) commentary

      This user story has an extra requirement: to create a group that others cannot write to. Otherwise, "only my" would become "only my and my readers' "

  25. Jan 2015
    1. )
    2. };
    3. A reactive value could represent: * The currently selected item in a table in a user interface * The result of a database query (changing whenever the underlying information in the database changes) * In a web browser, the current size of the browser window, or (in a browser that supports the HTML5 pushState API) the current URL * A string such as "5 minutes ago" or "last Tuesday" that gives the difference between a particular time and the current time

      markup error

    4. tag: 0

      'tag' does not appear in the line just below the block

    5. is added or deleted

      is edited, added or deleted

    6. // Efficient code says: No pizza for you!

      I think this implies the previous value was "pizza". Perhaps there should be an extra line in the block above:

      var data = new ReactiveDict;
      data.set("favoriteFood", "pizza");
      
    7. Just made up the term "property selector" to differentiate from "field selector" (or "field specifier" in the docs); open to suggestion

      left-over comment?

    1. Any distribution of the form given by equation 9.1 is an example of a Boltzmann distribution

      Is the definition of a Boltzmann machine not more specific about how \(E(x)\) is calculated from the variables? (constraining it to something like \(E=\sum{w_{ij} x_i x_j}...\)).

  26. Dec 2014
    1. Wikipedia Wikia Common Crawl Bitly (bundles and stories via the Story API) Wayfinder.is

      Common Crawl makes most sense to me

    1. The Annotator Problem

      Test. An annotation on the edge of the posts content. Will it still attach to the original post?

    1. random B

      Would a random orthogonal matrix B perhaps improve the results further?

  27. Nov 2014
  28. Oct 2014
    1. We are

      We are a project with a mission. Should we not mention that first? Being non-profit with these mentioned funders is just how we go about, it's not the core of who we are.

    1. is cross-posted

      Should a cross-post like this provide a canonical URL, or another way to show the content equivalence?

  29. Aug 2014
  30. Jul 2014
    1. and \ request.accept_mimetypes[best] > \ request.accept_mimetypes['text/html']

      This check should not be necessary if the order of json and html were swapped in the list passed to best_match. (see the best_match description)

    1. a reply to a previous statement

      What if one writes a comment on a document in reply to another annotation? I suppose our reply will have two targets: the older annotation, and the document. How do we express that the annotation is a reply to one target and a comment to the other?

    2. oa:serializedAt Property The time at which the agent referenced by oa:serializedBy generated the first serialization of the Annotation, and any subsequent substantially different one. The annotation graph MUST have changed for this property to be updated, and as such represents the last modified datestamp for the Annotation. This might be used to determine if it should be re-imported into a triplestore when discovered. There MAY be exactly 1 oa:serializedAt property per Annotation, and MUST NOT be more than 1. The datetime MUST be expressed in the xsd:dateTime format, and SHOULD have a timezone specified.