1.1 --- a/grant.html Sat Apr 11 19:12:32 2009 -0700 1.2 +++ b/grant.html Sat Apr 11 19:50:21 2009 -0700 1.3 @@ -108,12 +108,12 @@ 1.4 to label more than a few genes. Therefore, we must select only a few genes as 1.5 features. 1.6 Principle 3: Use geometry in feature selection 1.7 - When doing feature selection with score-based methods, the simplest thing to 1.8 - do would be to score the performance of each voxel by itself and then combine 1.9 - these scores; this is pointwise scoring. A more powerful approach is to also use 1.10 - information about the geometric relations between each voxel and its neighbors; 1.11 - this requires non-pointwise, local scoring methods. See Preliminary Results for 1.12 - evidence of the complementary nature of pointwise and local scoring methods. 1.13 + When doing feature selection with score-based methods, the simplest thing to do 1.14 + would be to score the performance of each voxel by itself and then combine these 1.15 + scores (pointwise scoring). A more powerful approach is to also use information 1.16 + about the geometric relations between each voxel and its neighbors; this requires 1.17 + non-pointwise, local scoring methods. See Preliminary Results for evidence of 1.18 + the complementary nature of pointwise and local scoring methods. 1.19 Principle 4: Work in 2-D whenever possible 1.20 There are many anatomical structures which are commonly characterized in 1.21 terms of a two-dimensional manifold. When it is known that the structure that 1.22 @@ -121,6 +121,8 @@ 1.23 the analysis algorithm to take advantage of this prior knowledge. In addition, 1.24 it is easier for humans to visualize and work with 2-D data. 1.25 Therefore, when possible, the instances should represent pixels, not voxels. 1.26 + Aim 2 1.27 + todo 1.28 3 1.29 1.30 Aim 3 1.31 @@ -173,6 +175,7 @@ 1.32 conceivable that the methods we propose to develop could be used to suggest 1.33 modifications to the human cortical map as well. 1.34 Related work 1.35 + todo 1.36 Preliminary work 1.37 Justification of principles 1 thur 3 1.38 Principle 1: Combinatorial gene expression 1.39 @@ -198,8 +201,6 @@ 1.40 useful. 1.41 The requirement to find combinations of only a small number of genes limits 1.42 us from straightforwardly applying many of the most simple techniques from 1.43 - the field of supervised machine learning. In the parlance of machine learning, 1.44 - our task combines feature selection with supervised learning. 1.45 __________________________ 1.46 1“WW, C2 and coiled-coil domain containing 1”; EntrezGene ID 211652 1.47 2“mitochondrial translational initiation factor 2”; EntrezGene ID 76784 1.48 @@ -228,6 +229,8 @@ 1.49 genes which (individually) best match area AUD, according to gradient similar- 1.50 ity. From left to right and top to bottom, the genes are Ssr1, Efcbp1, Aph1a, 1.51 Ptk7, Aph1a again, and Lepr 1.52 + the field of supervised machine learning. In the parlance of machine learning, 1.53 + our task combines feature selection with supervised learning. 1.54 Principle 3: Use geometry 1.55 To show that local geometry can provide useful information that cannot be 1.56 detected via pointwise analyses, consider Fig. . The top row of Fig. displays