Heterogeneous Multi-output Gaussian Process Prediction Pablo Moreno-Muñoz Antonio Artés-Rodríguez Mauricio A. Álvarez Poster #14 1
Motivation Output observations are a mix of continuous, binary, categorical or discrete variables Problem Multi-output Gaussian process models usually focus on all-regression or all-classification tasks Provide an extension of multi-output Gaussian processes Goal for prediction in arbitrary heterogeneous datasets 2
Motivation Medical Intensive Care Units (ICU) Electronic Health Records Spatio-temporal Applications Demographic, sociological or economic analysis of cities Bayesian Optimization Flat Other Functions with multiple outputs from different nature 2
<latexit sha1_base64="CJBuvCDKLyR1MrDv7b+NdJQsObM=">ACWXicbVFbS8MwGE3rbc7bdI+BIewvYxWBH0RH3wcYJTYZ0lTdMtLG1C8lUctX/SB0H8Kz6YziHePg5nPdchIpwQ143qvjLiwuLa/UVutr6xubW43tnRsjc01Zn0oh9V1EDBM8Y3gINid0oykWC30eS80m8fmDZcZtcwVWyYklHGE04JWCpsKNUOUgLjKCm5VMQSRGbaWqvIoAxA1J+yY9lp4NPcKC0jMiPvHL+Ki/F4exv81COMfLcJGy+t6s8B/gT8HLTSPXth4DmJ85RlQAUxZuB7CoYF0cCpYGU9yA1ThE7IiA0szEjKzLCYOVPifcvEOJHangzwjP1eUZDUVOvazGpH81uryP+0Q7J8bDgmcqBZfRzUJILDBJXNuOYa0ZBTC0gVHO7K6ZjogkF+xl1a4L/+8l/wc1B1/e6/tVh6/RsbkcN7aI91EY+OkKn6BL1UB9R9ILenSVn2XlzHbfm1j9TXWde0Q/wm1+AEN3tvg=</latexit> <latexit sha1_base64="CJBuvCDKLyR1MrDv7b+NdJQsObM=">ACWXicbVFbS8MwGE3rbc7bdI+BIewvYxWBH0RH3wcYJTYZ0lTdMtLG1C8lUctX/SB0H8Kz6YziHePg5nPdchIpwQ143qvjLiwuLa/UVutr6xubW43tnRsjc01Zn0oh9V1EDBM8Y3gINid0oykWC30eS80m8fmDZcZtcwVWyYklHGE04JWCpsKNUOUgLjKCm5VMQSRGbaWqvIoAxA1J+yY9lp4NPcKC0jMiPvHL+Ki/F4exv81COMfLcJGy+t6s8B/gT8HLTSPXth4DmJ85RlQAUxZuB7CoYF0cCpYGU9yA1ThE7IiA0szEjKzLCYOVPifcvEOJHangzwjP1eUZDUVOvazGpH81uryP+0Q7J8bDgmcqBZfRzUJILDBJXNuOYa0ZBTC0gVHO7K6ZjogkF+xl1a4L/+8l/wc1B1/e6/tVh6/RsbkcN7aI91EY+OkKn6BL1UB9R9ILenSVn2XlzHbfm1j9TXWde0Q/wm1+AEN3tvg=</latexit> <latexit sha1_base64="CJBuvCDKLyR1MrDv7b+NdJQsObM=">ACWXicbVFbS8MwGE3rbc7bdI+BIewvYxWBH0RH3wcYJTYZ0lTdMtLG1C8lUctX/SB0H8Kz6YziHePg5nPdchIpwQ143qvjLiwuLa/UVutr6xubW43tnRsjc01Zn0oh9V1EDBM8Y3gINid0oykWC30eS80m8fmDZcZtcwVWyYklHGE04JWCpsKNUOUgLjKCm5VMQSRGbaWqvIoAxA1J+yY9lp4NPcKC0jMiPvHL+Ki/F4exv81COMfLcJGy+t6s8B/gT8HLTSPXth4DmJ85RlQAUxZuB7CoYF0cCpYGU9yA1ThE7IiA0szEjKzLCYOVPifcvEOJHangzwjP1eUZDUVOvazGpH81uryP+0Q7J8bDgmcqBZfRzUJILDBJXNuOYa0ZBTC0gVHO7K6ZjogkF+xl1a4L/+8l/wc1B1/e6/tVh6/RsbkcN7aI91EY+OkKn6BL1UB9R9ILenSVn2XlzHbfm1j9TXWde0Q/wm1+AEN3tvg=</latexit> <latexit sha1_base64="CJBuvCDKLyR1MrDv7b+NdJQsObM=">ACWXicbVFbS8MwGE3rbc7bdI+BIewvYxWBH0RH3wcYJTYZ0lTdMtLG1C8lUctX/SB0H8Kz6YziHePg5nPdchIpwQ143qvjLiwuLa/UVutr6xubW43tnRsjc01Zn0oh9V1EDBM8Y3gINid0oykWC30eS80m8fmDZcZtcwVWyYklHGE04JWCpsKNUOUgLjKCm5VMQSRGbaWqvIoAxA1J+yY9lp4NPcKC0jMiPvHL+Ki/F4exv81COMfLcJGy+t6s8B/gT8HLTSPXth4DmJ85RlQAUxZuB7CoYF0cCpYGU9yA1ThE7IiA0szEjKzLCYOVPifcvEOJHangzwjP1eUZDUVOvazGpH81uryP+0Q7J8bDgmcqBZfRzUJILDBJXNuOYa0ZBTC0gVHO7K6ZjogkF+xl1a4L/+8l/wc1B1/e6/tVh6/RsbkcN7aI91EY+OkKn6BL1UB9R9ILenSVn2XlzHbfm1j9TXWde0Q/wm1+AEN3tvg=</latexit> Heterogeneous Likelihoods D Y p ( y | θ ( x )) = p ( y d | θ d ( x )) d =1 3
<latexit sha1_base64="CJBuvCDKLyR1MrDv7b+NdJQsObM=">ACWXicbVFbS8MwGE3rbc7bdI+BIewvYxWBH0RH3wcYJTYZ0lTdMtLG1C8lUctX/SB0H8Kz6YziHePg5nPdchIpwQ143qvjLiwuLa/UVutr6xubW43tnRsjc01Zn0oh9V1EDBM8Y3gINid0oykWC30eS80m8fmDZcZtcwVWyYklHGE04JWCpsKNUOUgLjKCm5VMQSRGbaWqvIoAxA1J+yY9lp4NPcKC0jMiPvHL+Ki/F4exv81COMfLcJGy+t6s8B/gT8HLTSPXth4DmJ85RlQAUxZuB7CoYF0cCpYGU9yA1ThE7IiA0szEjKzLCYOVPifcvEOJHangzwjP1eUZDUVOvazGpH81uryP+0Q7J8bDgmcqBZfRzUJILDBJXNuOYa0ZBTC0gVHO7K6ZjogkF+xl1a4L/+8l/wc1B1/e6/tVh6/RsbkcN7aI91EY+OkKn6BL1UB9R9ILenSVn2XlzHbfm1j9TXWde0Q/wm1+AEN3tvg=</latexit> <latexit sha1_base64="CJBuvCDKLyR1MrDv7b+NdJQsObM=">ACWXicbVFbS8MwGE3rbc7bdI+BIewvYxWBH0RH3wcYJTYZ0lTdMtLG1C8lUctX/SB0H8Kz6YziHePg5nPdchIpwQ143qvjLiwuLa/UVutr6xubW43tnRsjc01Zn0oh9V1EDBM8Y3gINid0oykWC30eS80m8fmDZcZtcwVWyYklHGE04JWCpsKNUOUgLjKCm5VMQSRGbaWqvIoAxA1J+yY9lp4NPcKC0jMiPvHL+Ki/F4exv81COMfLcJGy+t6s8B/gT8HLTSPXth4DmJ85RlQAUxZuB7CoYF0cCpYGU9yA1ThE7IiA0szEjKzLCYOVPifcvEOJHangzwjP1eUZDUVOvazGpH81uryP+0Q7J8bDgmcqBZfRzUJILDBJXNuOYa0ZBTC0gVHO7K6ZjogkF+xl1a4L/+8l/wc1B1/e6/tVh6/RsbkcN7aI91EY+OkKn6BL1UB9R9ILenSVn2XlzHbfm1j9TXWde0Q/wm1+AEN3tvg=</latexit> <latexit sha1_base64="CJBuvCDKLyR1MrDv7b+NdJQsObM=">ACWXicbVFbS8MwGE3rbc7bdI+BIewvYxWBH0RH3wcYJTYZ0lTdMtLG1C8lUctX/SB0H8Kz6YziHePg5nPdchIpwQ143qvjLiwuLa/UVutr6xubW43tnRsjc01Zn0oh9V1EDBM8Y3gINid0oykWC30eS80m8fmDZcZtcwVWyYklHGE04JWCpsKNUOUgLjKCm5VMQSRGbaWqvIoAxA1J+yY9lp4NPcKC0jMiPvHL+Ki/F4exv81COMfLcJGy+t6s8B/gT8HLTSPXth4DmJ85RlQAUxZuB7CoYF0cCpYGU9yA1ThE7IiA0szEjKzLCYOVPifcvEOJHangzwjP1eUZDUVOvazGpH81uryP+0Q7J8bDgmcqBZfRzUJILDBJXNuOYa0ZBTC0gVHO7K6ZjogkF+xl1a4L/+8l/wc1B1/e6/tVh6/RsbkcN7aI91EY+OkKn6BL1UB9R9ILenSVn2XlzHbfm1j9TXWde0Q/wm1+AEN3tvg=</latexit> <latexit sha1_base64="CJBuvCDKLyR1MrDv7b+NdJQsObM=">ACWXicbVFbS8MwGE3rbc7bdI+BIewvYxWBH0RH3wcYJTYZ0lTdMtLG1C8lUctX/SB0H8Kz6YziHePg5nPdchIpwQ143qvjLiwuLa/UVutr6xubW43tnRsjc01Zn0oh9V1EDBM8Y3gINid0oykWC30eS80m8fmDZcZtcwVWyYklHGE04JWCpsKNUOUgLjKCm5VMQSRGbaWqvIoAxA1J+yY9lp4NPcKC0jMiPvHL+Ki/F4exv81COMfLcJGy+t6s8B/gT8HLTSPXth4DmJ85RlQAUxZuB7CoYF0cCpYGU9yA1ThE7IiA0szEjKzLCYOVPifcvEOJHangzwjP1eUZDUVOvazGpH81uryP+0Q7J8bDgmcqBZfRzUJILDBJXNuOYa0ZBTC0gVHO7K6ZjogkF+xl1a4L/+8l/wc1B1/e6/tVh6/RsbkcN7aI91EY+OkKn6BL1UB9R9ILenSVn2XlzHbfm1j9TXWde0Q/wm1+AEN3tvg=</latexit> <latexit sha1_base64="NsLos3laW3WlkybC0/eUHSYSJS0=">AB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GNRDx5bsB/QhrLZTtq1m03Y3Qgl9Bd48aCIV3+SN/+N2zYHbX0w8Hhvhpl5QSK4Nq7RTW1jc2t4rbpZ3dvf2D8uFRS8epYthksYhVJ6AaBZfYNwI7CQKaRQIbAfj25nfkKleSwfzCRBP6JDyUPOqLFS465frhVdw6ySrycVCBHvV/+6g1ilkYoDRNU67nJsbPqDKcCZyWeqnGhLIxHWLXUkj1H42P3RKzqwyIGsbElD5urviYxGWk+iwHZG1Iz0sjcT/O6qQmv/YzLJDUo2WJRmApiYjL7mgy4QmbExBLKFLe3EjaijJjsynZELzl1dJ6LquVWvcVmp3eRxFOETuEcPLiCGtxDHZrAOEZXuHNeXRenHfnY9FacPKZY/gD5/MHl2mMyA=</latexit> <latexit sha1_base64="NsLos3laW3WlkybC0/eUHSYSJS0=">AB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GNRDx5bsB/QhrLZTtq1m03Y3Qgl9Bd48aCIV3+SN/+N2zYHbX0w8Hhvhpl5QSK4Nq7RTW1jc2t4rbpZ3dvf2D8uFRS8epYthksYhVJ6AaBZfYNwI7CQKaRQIbAfj25nfkKleSwfzCRBP6JDyUPOqLFS465frhVdw6ySrycVCBHvV/+6g1ilkYoDRNU67nJsbPqDKcCZyWeqnGhLIxHWLXUkj1H42P3RKzqwyIGsbElD5urviYxGWk+iwHZG1Iz0sjcT/O6qQmv/YzLJDUo2WJRmApiYjL7mgy4QmbExBLKFLe3EjaijJjsynZELzl1dJ6LquVWvcVmp3eRxFOETuEcPLiCGtxDHZrAOEZXuHNeXRenHfnY9FacPKZY/gD5/MHl2mMyA=</latexit> <latexit sha1_base64="NsLos3laW3WlkybC0/eUHSYSJS0=">AB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GNRDx5bsB/QhrLZTtq1m03Y3Qgl9Bd48aCIV3+SN/+N2zYHbX0w8Hhvhpl5QSK4Nq7RTW1jc2t4rbpZ3dvf2D8uFRS8epYthksYhVJ6AaBZfYNwI7CQKaRQIbAfj25nfkKleSwfzCRBP6JDyUPOqLFS465frhVdw6ySrycVCBHvV/+6g1ilkYoDRNU67nJsbPqDKcCZyWeqnGhLIxHWLXUkj1H42P3RKzqwyIGsbElD5urviYxGWk+iwHZG1Iz0sjcT/O6qQmv/YzLJDUo2WJRmApiYjL7mgy4QmbExBLKFLe3EjaijJjsynZELzl1dJ6LquVWvcVmp3eRxFOETuEcPLiCGtxDHZrAOEZXuHNeXRenHfnY9FacPKZY/gD5/MHl2mMyA=</latexit> <latexit sha1_base64="NsLos3laW3WlkybC0/eUHSYSJS0=">AB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GNRDx5bsB/QhrLZTtq1m03Y3Qgl9Bd48aCIV3+SN/+N2zYHbX0w8Hhvhpl5QSK4Nq7RTW1jc2t4rbpZ3dvf2D8uFRS8epYthksYhVJ6AaBZfYNwI7CQKaRQIbAfj25nfkKleSwfzCRBP6JDyUPOqLFS465frhVdw6ySrycVCBHvV/+6g1ilkYoDRNU67nJsbPqDKcCZyWeqnGhLIxHWLXUkj1H42P3RKzqwyIGsbElD5urviYxGWk+iwHZG1Iz0sjcT/O6qQmv/YzLJDUo2WJRmApiYjL7mgy4QmbExBLKFLe3EjaijJjsynZELzl1dJ6LquVWvcVmp3eRxFOETuEcPLiCGtxDHZrAOEZXuHNeXRenHfnY9FacPKZY/gD5/MHl2mMyA=</latexit> Heterogeneous Likelihoods D Y p ( y | θ ( x )) = p ( y d | θ d ( x )) d =1 different statistical data types D 3
Recommend
More recommend