Hello all. This meeting has been cancelled. Thank you,<br>-- <br>Vicente Malave<br><br><div class="gmail_quote">On Fri, Nov 13, 2009 at 2:25 PM, Marta Kutas <span dir="ltr"><<a href="mailto:mkutas@ucsd.edu" target="_blank">mkutas@ucsd.edu</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
<br>
Monday, Nov 16, 3:30, CSB 280<br>
<br>
Vicente Malave<br>
<br>
Classification Caveats : Practical limits of Multi-Voxel Pattern Analysis in fMRI.<br>
<br>
Multivariate pattern classification techniques are attracting increasing attention as an alternative data analysis technique in cognitive neuroscience. The purpose of this talk is to provide a critical look at two inferences using these methods. The first inference is simply that two conditions of the experiment are different. We discuss the connection between pattern classification methods and classical multivariate hypothesis testing. We briefly discuss resampling-based statistics, statistical power, and the impact on experimental design. The second inference is in interpretation of the model learned by a linear classifier. In particular, many studies compute "importance maps" that purport to show where discriminating information between the two conditions is located in the brain. We provide a critical look at this inference by noting that many very different models can achieve similar classification accuracy, and many very similar models can perform at chance. We show some techniques for exploring the space of possible models, discuss proposed solutions from the literature, and urge caution when making inferences about the importance of particular feature in classification.<br>
<br>
<br>
all welcome<br>
</blockquote></div><br><br clear="all"><br>