Sharif Digital Repository

In this paper, we introduce a document clustering method based on Sparse Topical Coding, called Cluster-based Sparse Topical Coding. Topic modeling is capable of improving textual document clustering by describing documents via bag-of-words models and projecting them into a topic space. The latent semantic descriptions derived by the topic model can be utilized as features in a clustering process. In our proposed method, document clustering and topic modeling are integrated in a unified framework in order to achieve the highest performance. This framework includes Sparse Topical Coding, which is responsible for topic mining, and K-means that discovers the latent clusters in documents... 

A measure of correlation is said to have the tensorization property if it does not change when computed for i.i.d. copies. More precisely, a measure of correlation between two random variables X, Y denoted by rho (X, Y), has the tensorization property if ρ(Xn, Yn)=ρ (X, Y) where (Xn, Yn) denotes n i.i.d. copies of (X, Y). Two well-known examples of such measures are the maximal correlation and the hypercontractivity ribbon (HC ribbon). We show that the maximal correlation and the HC ribbon are special cases of the new notion of Φ-ribbons, defined in this paper for a class of convex functions Φ. Φ-ribbon reduces to the HC ribbon and the maximal correlation for special choices of Φ, and is a... 

We present our work on improving the numerator graph for discriminative training using the lattice-free maximum mutual information (MMI) criterion. Specifically, we propose a scheme for creating unconstrained numerator graphs by removing time constraints from the baseline numerator graphs. This leads to much smaller graphs and therefore faster preparation of training supervisions. By testing the proposed un-constrained supervisions using factorized time-delay neural network (TDNN) models, we observe 0.5% to 2.6% relative improvement over the state-of-the-art word error rates on various large-vocabulary speech recognition databases. © 2018 IEEE  

Semantic segmentation is one of the basic, yet essential scene understanding tasks for an autonomous agent. The recent developments in supervised machine learning and neural networks have enjoyed great success in enhancing the performance of the state-of-the-art techniques for this task. However, their superior performance is highly reliant on the availability of a large-scale annotated dataset. In this letter, we propose a novel fully unsupervised semantic segmentation method, the so-called Information Maximization and Adversarial Regularization Segmentation (InMARS). Inspired by human perception which parses a scene into perceptual groups, rather than analyzing each pixel individually, our... 

A controversial issue in artificial intelligence is human emotion recognition. This paper presents a fuzzy parallel cascades (FPC) model for predicting the continuous subjective emotional appraisal of music by time-varying spectral content of electroencephalogram (EEG) signals. The EEG, along with an emotional appraisal of 15 subjects, was recorded during listening to seven musical excerpts. The emotional appraisement was recorded along the valence and arousal emotional axes as a continuous signal. The FPC model was composed of parallel cascades with each cascade containing a fuzzy logic-based system. The FPC model performance was evaluated using linear regression (LR), support vector... 

Due to unstoppable growth of social networks and the large number of users, the detection of communities have become one of the most popular and successful domain of research areas. Detecting communities is a significant aspect in analyzing networks because of its various applications such as sampling, link prediction and communications among members of social networks. There have been proposed many different algorithms for solving community detection problem containing optimization methods. In this paper we propose a novel algorithm based on extended distributed learning automata for solving this problem. Our proposed algorithm benefits from cooperation between learning automata to detect... 

We present our work on end-to-end training of acoustic models using the lattice-free maximum mutual information (LF-MMI) objective function in the context of hidden Markov models. By end-to-end training, we mean flat-start training of a single DNN in one stage without using any previously trained models, forced alignments, or building state-tying decision trees. We use full biphones to enable context-dependent modeling without trees, and show that our end-to-end LF-MMI approach can achieve comparable results to regular LF-MMI on well-known large vocabulary tasks. We also compare with other end-to-end methods such as CTC in character-based and lexicon-free settings and show 5 to 25 percent... 

Similarity measure is a main core of image registration algorithms. Spatially varying intensity distortion is an important challenge, which affects the performance of similarity measures. Correlation among the pixels is the main characteristic of this distortion. Similarity measures such as sum-of-squared-differences (SSD) and mutual information ignore this correlation; hence, perfect registration cannot be achieved in the presence of this distortion. In this paper, we model this correlation with the aid of the low rank matrix theory. Based on this model, we compensate this distortion analytically and introduce rank-regularized SSD (RRSSD). This new similarity measure is a modified SSD based... 

We propose two nonlinear regression methods, namely, Adversarial Orthogonal Regression (AdOR) for additive noise models and Adversarial Orthogonal Structural Equation Model (AdOSE) for the general case of structural equation models. Both methods try to make the residual of regression independent from regressors, while putting no assumption on noise distribution. In both methods, two adversarial networks are trained simultaneously where a regression network outputs predictions and a loss network that estimates mutual information (in AdOR) and KL-divergence (in AdOSE). These methods can be formulated as a minimax two-player game; at equilibrium, AdOR finds a deterministic map between inputs...