Showing papers by "Constantine Kotropoulos published in 2019"

Leveraging Image-to-image Translation Generative Adversarial Networks for Face Aging

Abstract: Here, face images of a specific age class are translated to images of different age classes in an unsupervised manner that enables training on independent sets of images for each age class. In order to learn pairwise translations between age classes, we adopt the UNsupervised Image-to-image Translation framework that employs Variational AutoEncoders and Generative Adversarial Networks. By mapping face images of different age classes to shared latent representations, the most personalized and abstract facial characteristics are preserved. To effectively diffuse age class information, a pyramid of local, neighbour, and global encoders is employed so that the latent representations progressively cover an increased age range. The proposed framework is applied to the FGNET aging database and compared to state-of-the-art techniques and the ground truth. Appealing experimental results demonstrate the ability of the proposed method to efficiently capture both intense and subtle aging effects.

A Simple Algorithm for Non-Negative Sparse Principal Component Analysis

Abstract: A novel adaptive method that computes sparse and non-negative eigenvectors is proposed. The proposed method achieves the sparsity in the derived eigenvectors implicitly through non-negative constraints and does not include any additional parameters except the learning rate. Although adding constraints to the standard Principal Component Analysis (PCA) leads to a reduction in the explained variance, the proposed method performs competitively with the standard PCA and other PCA variants. The assessment of the proposed method is conducted by performing a quantitative and qualitative evaluation.

Efficient Capon-Based Approach Exploiting Temporal Windowing for Electric Network Frequency Estimation

Abstract: Electric Network Frequency (ENF) fluctuations constitute a powerful tool in multimedia forensics. An efficient approach for ENF estimation is introduced with temporal windowing based on the filter-bank Capon spectral estimator. A type of Gohberg-Semencul factorization of the model covariance matrix is used due to the Toeplitz structure of the covariance matrix. Moreover, this approach uses, for the first time in the field of ENF, a temporal window, not necessarily the rectangular one, at the stage preceding spectral estimation. Krylov matrices are employed for fast implementation of matrix inversions. The proposed approach outperforms the state-of-the-art methods in ENF estimation, when a short time window of 1 second is employed in power recordings. In speech recordings, the proposed approach yields highly accurate results with respect to both time complexity and accuracy. Moreover, the impact of different temporal windows is studied. The results show that even the most trivial methods for ENF estimation, such as the Short-Time Fourier Transform, can provide better results than the most recent state-of-the-art methods, when a temporal window is employed. The correlation coefficient is used to measure the ENF estimation accuracy.

Block Randomized Optimization for Adaptive Hypergraph Learning

Abstract: The high-order relations between the content in social media sharing platforms are frequently modeled by a hypergraph. Either hypergraph Laplacian matrix or the adjacency matrix is a big matrix. Randomized algorithms are used for low-rank factorizations in order to approximately decompose and eventually invert such big matrices fast. Here, block randomized Singular Value Decomposition (SVD) via subspace iteration is integrated within adaptive hypergraph weight estimation for image tagging, as a first approach. Specifically, creating low-rank submatrices along the main diagonal by tessellation permits fast matrix inversions via randomized SVD. Moreover, a second approach is proposed for solving the linear system in the optimization problem of hypergraph learning by employing the conjugate gradient method. Both proposed approaches achieve high accuracy in image tagging measured by F 1 score and succeed to reduce the computational requirements of adaptive hypergraph weight estimation.

A Dyadic Particle Filter for Price Prediction

Abstract: The most difficult task in financial forecasting is the accurate price prediction based on previous values. Two cases are studied: stock price prediction and flight price prediction. A dyadic particle filter is proposed that is based on sequential importance resampling. This dyadic particle filter captures the dynamic evolution of a pair of latent vectors. In stock price prediction, one latent vector is defined for each stock. This latent vector is paired with a market segment latent vector introduced for each group of companies of the same category. Both latent vectors capture the hidden information of the stock market and reinforce the state estimation procedure. This hidden information influences strongly the performance of the particle filter, yielding more accurate prediction of stock prices than the state-of-the-art techniques. For flight price prediction, the pair of latent vectors corresponds to route and destination, respectively. Given the price range of each flight, promising results are disclosed.

A Weighted Ordered Probit Collaborative Kalman Filter For Hotel Rating Prediction

Abstract: A successful recommender system interacts with users and learns their preferences. This is crucial in order to provide accurate recommendations. In this paper, a Weighted Ordered Probit Collaborative Kalman filter is proposed for hotel rating prediction. Since potential changes may occur in hotel services or accommodation conditions, a hotel popularity may be volatile through time. A weighted ordered probit model is introduced to capture this latent trend about each hotel popularity through time. It is demonstrated by experiments that such model of hotel popularity trends reinforces the performance of Collaborative Kalman filter, yielding more accurate potential recommendations.

Efficient Capon-Based Approach Exploiting Temporal Windowing For Electric Network Frequency Estimation

Abstract: Electric Network Frequency (ENF) fluctuations constitute a powerful tool in multimedia forensics. An efficient approach for ENF estimation is introduced with temporal windowing based on the filter-bank Capon spectral estimator. A type of Gohberg-Semencul factorization of the model covariance matrix is used due to the Toeplitz structure of the covariance matrix. Moreover, this approach uses, for the first time in the field of ENF, a temporal window, not necessarily the rectangular one, at the stage preceding spectral estimation. Krylov matrices are employed for fast implementation of matrix inversions. The proposed approach outperforms the state-of-the-art methods in ENF estimation, when a short time window of $1$ second is employed in power recordings. In speech recordings, the proposed approach yields highly accurate results with respect to both time complexity and accuracy. Moreover, the impact of different temporal windows is studied. The results show that even the most trivial methods for ENF estimation, such as the Short-Time Fourier Transform, can provide better results than the most recent state-of-the-art methods, when a temporal window is employed. The correlation coefficient is used to measure the ENF estimation accuracy.

Block Randomized Optimization for Adaptive Hypergraph Learning

Abstract: The high-order relations between the content in social media sharing platforms are frequently modeled by a hypergraph. Either hypergraph Laplacian matrix or the adjacency matrix is a big matrix. Randomized algorithms are used for low-rank factorizations in order to approximately decompose and eventually invert such big matrices fast. Here, block randomized Singular Value Decomposition (SVD) via subspace iteration is integrated within adaptive hypergraph weight estimation for image tagging, as a first approach. Specifically, creating low-rank submatrices along the main diagonal by tessellation permits fast matrix inversions via randomized SVD. Moreover, a second approach is proposed for solving the linear system in the optimization problem of hypergraph learning by employing the conjugate gradient method. Both proposed approaches achieve high accuracy in image tagging measured by F1 score and succeed to reduce the computational requirements of adaptive hypergraph weight estimation.