Deep Multi-Objective Goal Modeling – We consider the problem of predicting the future in a single target environment through the use of the MDP. We assume that the target environment is a single image, and a prediction is performed to find the minimum distance to this image from its initial target. We derive a novel dimension-reduced task that maximizes the importance of these parameters, which are relevant to the solution of the MDP. This dimension-reduced task captures the latent state representation of the target object, which is a feature of the MDP. The task is evaluated using a challenging synthetic dataset, and an application to the Cityscaping dataset.

Propositional formula matching (PFFM) aims to extract a specific formula from the input data. For this purpose, we use one-to-one correspondence between a formula and the input set to learn the relationship between the formulas and the values of a metric function in the matrix space. In particular, we propose a method that learns the relationship between a formula and every value of a metric function in different matrices. We define a matrix factorization-based model which learns the matrix metric function for each set of formulas to provide a measure of similarity between the formulas and the values of metric functions. We also propose a novel feature selection method for PFFM, which we call Recurrent Matrix Factorization (RBMF) feature selection. Our method performs well on benchmark databases as well as benchmark data. Empirical results demonstrate that our approach significantly outperforms other existing feature selection methods on PFFM and other well-known database datasets, including the FITC database (1,2,3).

Feature Matching to Improve Large-Scale Image Recognition

Comparing Deep Neural Networks to Matching Networks for Age Estimation

Deep Multi-Objective Goal Modeling

End-to-End Action Detection with Dynamic Contextual Mapping

A Novel Feature Selection Method Using Backpropagation for Propositional Formula MatchingPropositional formula matching (PFFM) aims to extract a specific formula from the input data. For this purpose, we use one-to-one correspondence between a formula and the input set to learn the relationship between the formulas and the values of a metric function in the matrix space. In particular, we propose a method that learns the relationship between a formula and every value of a metric function in different matrices. We define a matrix factorization-based model which learns the matrix metric function for each set of formulas to provide a measure of similarity between the formulas and the values of metric functions. We also propose a novel feature selection method for PFFM, which we call Recurrent Matrix Factorization (RBMF) feature selection. Our method performs well on benchmark databases as well as benchmark data. Empirical results demonstrate that our approach significantly outperforms other existing feature selection methods on PFFM and other well-known database datasets, including the FITC database (1,2,3).