- First version that can be conveniently installed using the pip install git+https://github.com/nicococo/niidbox.git Enjoy :)
- Transductive conditional random field regression (TCRFR) method with various inference schemes: (a) quadratic program approximation, (b) loopy belief propagation, (c) loopy approximation, and (d) brute force
This software combines the code for learning with latent dependency structures based on conditional random fields. Software code was written by Nico Goernitz in close collaboratoration with Luiz A. Lima, and Shinichi Nakaijma. TU Berlin, 2015.
Analyzing data with unknown spatial and/or temporal structure is a challenge for machine learning. We propose a novel nonlinear model for studying data with latent dependency structure. It successfully combines the concepts of Markov random fields, transductive learning and regression, heavily using the notion of joint feature maps. Our transductive conditional random field regression (TCRFR) model is able to infer latent structure by combining limited labeled data of high precision with unlabeled data containing measurement uncertainty. In this manner we can propagate certain information and greatly reduce uncertainty. We demonstrate the usefulness of our novel framework on toy data with known spatio-temporal structure and successfully validate on real world off-shore data from the oil industry. Here the generic challenge is to optimally integrate data resources such as oil well data containing highly accurate direct measurements of rock porosities, geologically hand-crafted data, and impedances with broad and imprecise data from spatial hydrophone array measurements.
NG was supported by BMBF ALICE II grant 01IB15001B. MK acknowledges support by the German Research Foundation through the grant KL 2698/1-1 and KL 2698/2-1. We also acknowledges the support by the German Research Foundation through the grant DFG MU 987/6-1 and RA 1894/1-1. KRM thanks for partial funding by the National Research Foundation of Korea funded by the Ministry of Education, Science, and Technology in the BK21 program. SN was supported by the German Ministry for Education and Research as Berlin Big Data Center BBDC, funding mark 01IS14013A. LAL acknowledges the support of Petrobras.
When using this software, or parts of it, in your own research, please cite to appear