Santosh Chapaneri

As an essential task for recommender systems, the rating prediction problem over several contexts has attracted more attention over the recent years. The traditional approaches ignore the contexts and thus fail to predict the ratings for the unseen data in the rating tensor for varying contextual scenarios. Matrix factorization is preferred over decomposing the rating tensor for avoiding the burden of very high computational complexity while learning the interaction of users’ and items’ latent features. In this work, we propose a novel kernel loss function for optimizing the objective function of matrix factorization in a non-linearly projected rating space under multiple contexts in an optimum manner and also incorporate the implicit feedback of items in the learning process. Further, the optimization is regularized by applying different weights for each regularization term depending on the users’ and items’ participation. Extensive experimental evaluation on five benchmark context-aware datasets indicates the superiority of the proposed work for capturing the non-linearity and predicting the ratings of unseen items for users under varying contexts over the existing and baseline methods. The proposed kernel loss function is also shown to be resistant against shilling attacks in the recommender system. A detailed ablation study demonstrates the validity of the proposed work and the results are shown to be statistically significantly better with an improvement in the range of 3% to 11% over the baseline methods.

This study concerns structured regression problems wherein the issue of covariate shift is addressed, which aims at reducing the discrepancy in training and test data distributions, using computationally efficient and sparse optimization principles. In particular, the projection-free Frank-Wolfe optimization algorithms are used to learn the importance weights and re-weight the training data in the context of covariate shift. To determine the unbiased estimates of the weights, Kullback-Leibler importance estimation procedure is used, but its computational cost can be high since it is based on projected gradient optimization. Instead of using the standard Frank-Wolfe algorithm, we adapt its variants and propose Away-steps Frank-Wolfe and Pairwise Frank-Wolfe Covariate Shift algorithms to correct the covariate shift. The results highlight the improved computational efficiency and sparsity achieved while learning the importance weights on synthetic as well as benchmark datasets. An importance weighted Sharma-Mittal Twin Gaussian Process structured regression framework is further proposed to incorporate the learned weights from covariate shift algorithms and its equations are derived for importance weighted derivatives and uncertainties. The performance of proposed algorithms is evaluated on two applications of structured regression, namely human pose estimation and music mood estimation, where the benefit of handling covariate shift is demonstrated with improved performance relative to the state-of-the-art techniques.

Due to the subjective nature of music mood, it is challenging to computationally model the affective content of music. In this work for estimating music mood, we present a deep Gaussian process regression model that is inspired from deep learning architectures. A variational Bayesian approach is used to learn the Gaussian mixture model representation for acoustic features. Since the exact inference on deep Gaussian processes (DGP) is intractable, the pseudo-data approximation is used to reduce the training complexity and Monte Carlo sampling technique is used to solve the intractability problem during training. A detailed derivation of a 3-layer DGP is presented that can be easily generalized to a L-layer DGP. The proposed work is evaluated on PMEmo dataset containing valence and arousal annotations of Western popular music and achieves an improvement in the coefficient of determination of 16.9% for arousal and 28.8% for valence estimation relative to the baseline single-layer Gaussian process.

Saliency is the quality by which any object or a pixel in an image stands out relative to its neighbours. Detecting such regions from an image is a crucial problem of research, since it has wide applications in advertising, automatic image compression, image thumbnailing, etc. In this paper, a salient region detection approach is proposed by using machine learning. In order to train the saliency model, low level features such as color channels and their probabilities, also probabilities using 3D color histograms, subband features along with statistical priors such as frequency prior, color prior, chance of happening (CoH) and center bias prior (CBP) are used. The proposed model is compared with existing state of art algorithms. Human eye fixation points are used to compare the models by estimating area under ROC curves. Other parameters such as precision, recall, F-measure are also used for comparison. This comparison shows that the proposed saliency model gives better performance than the existing salient region detection approaches.

Security is a major concern in digital image transmission applications. In this paper, a novel color image encryption scheme is proposed to enhance security and efficiency. The proposed scheme is a single round based hyper chaotic system due to bi-directional pixel diffusion which contributes towards increased security and improved efficiency. Security analysis such as key sensitivity, histogram, information entropy, correlation coefficient and diffusion is conducted.

In this paper, a No-Reference Image Quality Assessment (NR-IQA) algorithm is implemented with the help of Extreme Learning Machine (ELM) using spatial and spectral features. ELMs are single hidden layer feed-forward neural networks that provides optimum solution in a single iteration, hence ELMs can be used for performing classification and regression at high speeds. Proposed NR-IQA algorithm can quantify the amount of distortion for images caused by JPEG compression, JPEG2000 compression, Additive White Gaussian Noise, Gaussian Blurring effect and Rayleigh's Fast Fading effects. The proposed algorithm is evaluated using LIVE IQA database via Spearman's Ranked Ordered Correlation Coefficient (SROCC) and Root Mean Square Error (RMSE). The proposed algorithm outperforms existing NR-IQA methods.

In this paper, the performance of multi-taper spectral estimate is investigated relative to conventional single taper estimate for the application of emotion recognition from speech signals. Typically, a single taper/window helps in reducing bias of the estimate, but due to its high variance, the resulting spectral features tend to give poor recognition performance. The weighted averages of the multi-tapered uncorrelated eigen-spectra results in more discriminative spectral features, thus increasing the overall performance. We demonstrate that the application of six Multi-peak multi-tapers with support vector machine results in 81% classification accuracy on seven emotions from Berlin emotion database considering only spectral features, compared to 72% using conventional Hamming window method.