Restricted Boltzmann Machines for Collaborative Filtering

0. Gentle Description of RBMs Restricted Boltzmann Machines (RBM) are neural networks consisting of stochastic binary nodes that are utilized to learn the ‘connection’ between a layer of observed nodes and a layer of latent nodes. The ‘restricted’ property refers to the fact that every node is unconnected to every other node within the same… Continue reading Restricted Boltzmann Machines for Collaborative Filtering

Towards theoretical understanding of Domain Adaptation Learning

1. Introduction Machine Learning (ML) is used in many real world applications thanks to its powerful theory that has led to a wide variety of highly successful practical tools. However, the main problem with the theory is that it makes simple unrealistic assumptions that clearly fail in many real world applications. More specifically, a basic… Continue reading Towards theoretical understanding of Domain Adaptation Learning

On The Expressive Power Of Deep Neural Networks

1. Introduction In recent years, deep neural networks have achieved unreasonable effectiveness on a variety of tasks, such as image classification, object detection, speech recognition, natural language processing and so on. But we haven’t understood thoroughly why and how they can work. There are three factors that affect the effectiveness of neural networks: trainability, which… Continue reading On The Expressive Power Of Deep Neural Networks

Online Learning for Matrix Factorization and Sparse Coding

Online Learning for Matrix Factorization and Sparse Coding Problem Statement For each find a latent representation such that: it is sparse: the vector has zeros we can reconstruct the original input as well as possible Consider a finite training set of signals and optimize the empirical cost function More formally, where is the reconstruction and… Continue reading Online Learning for Matrix Factorization and Sparse Coding

A Case-Study of Sparse Subspace Clustering

1. Introduction Clustering is an important task which has usage over, several domains. For instance, in computer networks there are a lot of scenarios that we want to cluster users/nodes based on their bandwidths, locations or censors. Another example is in field of security where users can be divided based on their malicious/malevolent behaviors. The… Continue reading A Case-Study of Sparse Subspace Clustering