Linearized and Kernelized Sparse Multitask Learning for Predicting Cognitive Outcomes in Alzheimer’s Disease

Table 1

Performance comparison of various methods in terms of rMSE and nMSE on 10 cross validation cognitive prediction tasks.

Method	ADAS	MMSE	RAVLT
			TOTAL	TOT6	T30	RECOG
Ridge	7.556 0.294	2.656 0.134	11.41 0.498	3.907 0.236	4.052 0.224	4.331 0.294
Lasso	6.846 0.361	2.216 0.098	10.02 0.548	3.320 0.195	3.443 0.177	3.639 0.213
MKL	6.893 0.528	2.214 0.106	9.911 0.695	3.424 0.296	3.570 0.340	3.745 0.237
Robust MTL	7.651 0.442	3.326 0.266	11.02 0.590	3.574 0.235	3.704 0.171	3.858 0.310
CMTL	7.642 0.373	3.083 0.461	11.56 0.510	3.907 0.260	4.038 0.244	4.381 0.226
Trace	8.180 0.605	6.113 2.038	13.09 3.128	3.782 0.491	3.906 0.431	4.520 0.859
SRMTL	6.882 0.325	2.331 0.271	9.961 0.561	3.320 0.152	3.445 0.116	3.639 0.261
-MTL	6.772 0.312	2.206 0.081	9.606 0.448	3.344 0.154	3.440 0.151	3.644 0.247
-MKMTL	6.825 0.455	2.417 0.197	9.699 0.505	3.396 0.188	3.495 0.144	3.653 0.243
-MKMTL	6.806 0.447	2.185 0.106	9.628 0.510	3.331 0.196	3.467 0.172	3.627 0.199
Method	FLU		TRAILS		nMSE
	ANIM	VEG	A	B
Ridge	6.521 0.418	4.322 0.178	27.18 1.702	83.72 5.713	16.44 1.725
Lasso	5.352 0.447	3.701 0.093	23.75 1.398	71.23 2.812	12.05 0.758
MKL	5.342 0.510	3.761 0.137	24.71 1.781	78.09 6.916	13.56 1.133
Robust MTL	5.946 0.398	3.988 0.083	27.78 1.922	90.12 7.098	17.68 2.303
CMTL	6.608 0.561	4.398 0.284	27.46 1.980	83.66 5.418	16.67 1.912
Trace	6.743 1.425	4.672 0.778	28.82 3.278	89.68 7.838	20.23 5.215
SRMTL	5.327 0.334	3.713 0.088	25.09 1.421	80.00 4.637	14.01 1.169
-MTL	5.298 0.439	3.704 0.096	23.42 1.110	71.32 2.945	11.92 0.969
-MKMTL	5.304 0.350	3.676 0.094	23.09 1.438	70.28 0.898	11.72 0.222
-MKMTL	5.232 0.434	3.675 0.157	23.13 1.473	69.82 1.236	11.56 ± 0.602