Posted on Saturday, 25th February 2012

Please read section 8.1 and post a comment.

Posted in Class | Comments (13)

  1. Eric VanEpps Says:
    February 26th, 2012 at 10:17 am

    I’m intrigued by the use of mean square error to represent risk that you touch on briefly in 8.1.4, but I don’t understand exactly what the parameters are in the risk function showed. Can you explain what each parameter is? Thanks.

  2. Jay Scott Says:
    February 27th, 2012 at 10:07 pm

    SEC. 8.1.3: What is the advantage of MSE over measures like Mean Absolute Error or Mean Absolute Scaled Error which do not put so much weight on outliers?

  3. David Zhou Says:
    February 27th, 2012 at 10:30 pm

    What does squaring the mean error do to the bias and variance mathematically? What is the primary advantage of calculating mean squared error?

    Can you talk a bit in depth about decision theory? It sounds very interesting.

  4. Sharlene Flesher Says:
    February 27th, 2012 at 11:47 pm

    Would there ever be a situation where it would be more useful to estimate lambda in a Poisson as the sample variance rather than the sample mean, as was discussed in the example in 8.1.1?
    Also, can you elaborate on the relationship between MSE and risk?

  5. Shubham Debnath Says:
    February 28th, 2012 at 1:05 am

    Regression splines are mentioned as a smoothing method for comparing pseudo-data. What other methods are there, and what the benefits/downfalls for them?

  6. Amanda Markey Says:
    February 28th, 2012 at 1:06 am

    I did not follow the theorem on page 221-222.

  7. Rex Tien Says:
    February 28th, 2012 at 1:23 am

    Is the weighted mean still a maximum likelihood estimator?

  8. Matt Panico Says:
    February 28th, 2012 at 1:41 am

    You mention that real spike trains almost always differ from the Poisson assumption. A member of my lab keeps pushing for me to use the Poisson distribution in my analysis. How does this difference from Poisson usually look?

  9. Ben Dichter Says:
    February 28th, 2012 at 2:39 am

    When it comes to bias vs. variance, how do you determine which is the problem and how do you correct the issue either way?

  10. Matt Bauman Says:
    February 28th, 2012 at 7:44 am

    I’ve heard about ‘change point analysis before. Is that simply the systematic identification of these change points? And is a change point simply a value at which _any_ statistic changes?

  11. Yijuan Du Says:
    February 28th, 2012 at 8:47 am

    Do we usually assume the estimator is unbiased in calculation and then check this assumption, similar as mentioned in last lecture?

  12. Scott Kennedy Says:
    February 28th, 2012 at 9:10 am

    Can we go over in detail the methods described in the SEF example? I’m confused about how the ‘true’ firing rate was calculated, as the spline fit and the PSTH are compared with this.

  13. Rob Rasmussen Says:
    February 28th, 2012 at 9:19 am

    I really liked the example of the Poisson distribution and using the sample mean versus sample variance to estimate lambda. Is there any correlation between the sample mean and variance (minus their expected values), or are they independent random variables? I guess this would relate to the estimated Fano factor having variance from a sample of the Poisson distribution.

Leave a Reply

You must be logged in to post a comment.