Trolls!

Christian Robert points to this absurd patent of the Monte Carlo method (which, as Christian notes, was actually invented by Stanislaw Ulam and others in the 1940s).

The whole thing is pretty unreadable. I wonder if they first wrote it as a journal article and then it got rejected everywhere, so they decided to submit it as a patent instead.

What’s even worse is this bit:

This invention was made with government support under Grant Numbers 0612170 and 0347408 awarded by the National Science Foundation.

So our tax dollars are being given to IBM so they can try to bring statistics to a halt by patenting one of our most basic tools? I’d say this is just a waste of money, but given that our country is run by lawyers, there must be some outside chance that this patent could actually succeed?

Perhaps there’s room for an improvement in the patent that involves albedo in some way?

10 thoughts on “Trolls!

  1. Going through one or more iterations of Stigler’s law should it not be Galton’s Quincunx?

    A physical simulation directed at doing and understanding statistics.

    In particular for a better (more intuitive) concretization of Bayes Theorem than the Nearest Neighbours one I posted here a few years back see Fig 5 in

    Stigler, Stephen M. 2010. Darwin, Galton, and the statistical enlightenment. Journal of the Royal Statistical Society, Series A 173:469–482.

    K?

    • Should have guessed there are patents on the Quincunx!
      (Just type in the search on the link provided above.)

      Wonder if Jeff knows a lawyer that might be able to defend him re: his virtual online Quincunx?

      see http://probability.ca/jeff/java/uncunx.html (really nifty)

      And to head off my need for lawyers (the ? mark may not be enough).

      My _rudimentary_ R code for a 2 dimensional 2 stage virtual Quincunx (just an aerial view) is released to the public here.
      (It’s to demonstrate Bayesian Inference for the Log Odds Ratio when there are no successes in either group.)

      K?

      #R code
      library(animation)
      #set simulation granualrity (smaller more accurate)
      gran=.001
      #define transform function
      expit=function(x)
      exp(x)/(1 + exp(x))
      #generate random polar points in x and y
      x=cos(seq(0,2 * pi,by=gran))
      y=sin(seq(0,2 * pi,by=gran))
      #initialize matrix to hold results
      mat=matrix(NA,ncol=2,nrow=length(x))
      ### 1. How to setup a simple animation ###
      ## set some options first
      oopt = ani.options(interval = 0.005, nmax = 20)
      ii=sample(1:length(x),size=6)
      cone=function(k){
      clrs=2:6
      plot(x,y,xlim=c(0,1),ylim=c(0,1),type=”n”,xlab=”Pc”,ylab=”Pt”)
      text(hmat[,1],hmat[,2],”+”,col=hmat[,3],cex=2)
      if(k==1) lines(expit((x)),expit((y)),col=1)
      lines(expit(x + cumsum(x[ii])[k-1]),expit(y + cumsum(y[ii])[k-1]),col=clrs[k-1])

      text(expit(c(0,cumsum(x[ii]))[k]),expit(c(0,cumsum(y[ii]))[k]),”*”,col=clrs[k])
      xp=expit(c(0,cumsum(x[ii]))[k:(k+1)]);yp=expit(c(0,cumsum(y[ii]))[k:(k+1)])
      # text(xp[1],yp[1])
      slope=(yp[2]-yp[1])/(xp[2]-xp[1])
      difx=(xp[2]-xp[1])
      for(i in 0:10){
      if(k==1) text(xp[1] + i/10 * difx,yp[1] + (i/10 * difx) * slope,”*”,col=1)
      text(xp[1] + i/10 * difx,yp[1] + (i/10 * difx) * slope,”*”,col=clrs[k-1])
      }

      if(k==6) {
      return(c(xp[2],yp[2]))
      }
      }

      hmat=matrix(NA,ncol=3,nrow=300)
      for(j in 1:300){
      ii=sample(1:length(x),size=6)
      for(i in 1:6){
      ps=cone(i)

      if(i==6) {
      data=rbinom(n=c(1,1),prob=ps,size=c(10,2))
      text(ps[1],ps[2],”+”,col=(sum(data) < 1) + 1,cex=2)
      hmat[j,]=c(ps,(sum(data) < 1) + 1)
      }
      ani.pause()
      }
      }

      • If you (at least when I) replace the ” with the original ” the code runs.

        (Or email me at k_orourke(at)rogers.com)
        K?

  2. A huge problem is the failure to make the bar high enough in terms of not being obvious to someone skilled in the art.

    The usual process at labs like IBM is to jointly submit papers for review and for patent applications. Technology isn’t considered released to the public while under review, so it can’t count as prior art. If the lawyers like it (or your boss pushes hard enough), they file a provisional patent application as a placeholder. The patent lawyers then take your paper and convert it into a patent application by converting figures to look like they came from the 1950s and adding numbers to everything. The whole process costs upward of $40K and takes several years.

    So as not to pick on anyone else, here’s my one (joint) patent, and here’s the paper (pardon our relative ignorance of statistics).

    When I was at Bell Labs, they told us that they wanted us to “lay down a patent minefield in speech so that anyone stepping into the space can be sued”. One of my colleagues had a patent on “connecting a database to a speech recognizer” and another patent on “connecting more than one database to a speech recognizer”. So it’s not hard to see how IBM can generate thousands of patents per year.

    Patents are the main reason we have to pay lawyers to help with contracts at our boutique natural language processing company. Every customer wants to be indemnified against patent lawsuits in case we’ve inadvertently stepped on someone’s patents. If anyone were ever to sue one of our customers, we’re on line to defend them. Of course, we couldn’t, for instance, afford to defend Comcast against a lawsuit by IBM. Our little company would just go bankrupt. We like to think of it as the sword of Damocles model of keeping us honest.

    There are ridiculous non-software patents, too, like the “one click ordering” patent of Amazon’s.

  3. Actually, this is *not* a patent on Monte Carlo techniques.

    The embodiment and background in a patent do not determine what the invention is, except insofar that they have to describe enough so that somebody else could plausibly build the invention.

    The crux of what is actually being patented is in the claims. The claims in this case start with a description of hooking a database up to a sampler, maintaining many copies of records in the database and building a query planner that understands how to run queries against sampled data instead of single records. This is very far from patenting what Stan Ulam and Nick Metropolis were up to in the 50’s. To properly interpret these claims, btw, you also have to look back at the records of how the examiner and the inventor’s attorney haggled over how narrowly various terms are to be interpreted.

    I am not saying that software patents are a great thing or that the USPTO does a great job of examining patents, but this particular patent is not nearly as evil as it might appear. For instance, this patent: http://bit.ly/q0aFER really did issue and really does claim directing a coherent beam of light against a solid object and moving it to be out of reach of a cat. If you want to have an example of a bad patent, use that one instead.

    • @Ted- Thanks for clarifying. I have an an amateur interest in patent trolling (the phenomena, not doing it). Can you point to any short straight-forward guidelines on how to read and/or interpret patents?

  4. @Bob Thank you that is hilarious. I love how the word “apparatus” is added into the patent title to make it sound more “patent”-y

    I agree with the earlier poster that the recent episode of This American Life on patents should be required listening for anyone working in the realm of monetizing of ideas and algorithms…

  5. Pingback: Eleventh Linkfest

Comments are closed.