<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <p>Dear all,</p>
    <p>On Thursday (18 October) at 2pm in <span class="event-where">54
        / 7035 (7B), we have an S3RI seminar from </span><span
        class="event-where"><span class="event-description">Rebecca
          Killick (Lancaster University)</span> on </span><span
        class="event-where"><span class="event-description">"</span></span><span
        class="event-where"><span class="event-description"><span
            class="event-description"><span class="event-description">Computationally
              Efficient Multivariate Changepoint Detection with Subsets</span></span>".
          Details are given below.<br>
        </span></span></p>
    <p><span class="event-where"><span class="event-description"><span><span
              class="event-description">The seminar will also be
              available via a live web-cast at</span></span></span></span><br>
      <span class="event-description"><a class="moz-txt-link-freetext" href="https://coursec">https://coursec</a><wbr>ast.soton.ac.uk<wbr>/Panopto/Pages/<wbr>Viewer.aspx?id=<wbr>ec46c95b-eb59-4<wbr>da6-b85f-d81b07<wbr>0328c8</span></p>
    <p><span class="event-where"><span class="event-description"><span><span
              class="event-description">The talk will be followed by tea
              and cake in the staff reading room on level 4 of building
              54. </span></span></span></span></p>
    <p><span class="event-where"><span class="event-description"><span><span
              class="event-description">All are welcome!</span></span></span></span></p>
    <p><span class="event-where"><span class="event-description"><span>Best
            wishes,</span></span></span></p>
    <p><span class="event-where"><span class="event-description"><span>Helen</span></span></span></p>
    <span class="event-description">Computationally Efficient
      Multivariate Changepoint Detection with Subsets
      <br>
      <br>
      Rebecca Killick, Lancaster University
      <br>
      <br>
      Historically much of the research on changepoint analysis has
      focused on the univariate setting. Due to the growing number of
      high dimensional datasets there is an increasing need for methods
      that can detect changepoints in multivariate time series. In this
      talk we focus on the problem of detecting changepoints where only
      a subset of the variables under observation undergo a change, so
      called subset multivariate changepoints. One approach to locating
      changepoints is to choose the segmentation that minimises a
      penalised cost function via a dynamic program. The work in this
      presentation is the first to create a dynamic program specifically
      for detecting changes in subset-multivar<wbr>iate time series. The
      computational complexity of the dynamic program means it is
      infeasible even for medium datasets. Thus we propose a
      computationally efficient approximate dynamic program, SPOT. We
      demonstrate that SPOT always recovers a better segmentation, in
      terms of penalised cost, then other approaches which assume every
      variable changes. Furthermore under mild assumptions the
      computational cost of SPOT is linear in the number of data points.
      In small simulation studies we demonstrate that SPOT provides a
      good approximation to exact methods but is feasible for datasets
      that contain thousands of variables observed at millions of time
      points. Furthermore we demonstrate that our method compares
      favourably with other commonly used multivariate changepoint
      methods and achieves a substantial improvement in performance when
      compared with fully multivariate methods.</span><br>
    <br>
    <span class="event-where"><span class="event-description"><span><span
            class="event-description"><span> <span
                class="event-description"> </span><span
                class="event-description"><span
                  class="event-description">For the current schedule of
                  S3RI seminars, see <a class="moz-txt-link-freetext"
                    href="https://tinyurl.com/s3riseminar">https://tinyurl.com/s3riseminar</a>
                </span></span></span></span></span></span></span><br>
    <p><span class="event-where"><span class="event-description"><span><span
              class="event-description"></span></span></span></span></p>
  </body>
</html>