Fast k-means using Differentially Private Low-Rank Approximation for Multi-relational Data – In this paper, we propose a novel algorithm for the task of learning a discriminative dictionary for a dataset of different kinds. While previous methods are focused on learning discrete dictionary models, we show that our method can be applied to learn non-linear and multi-dimensional representations, and indeed, learn the dictionary as a vector from the dictionary representation of the input data. We propose a novel model for the task, but we also establish that it can be used to learn such dictionaries by generating discriminant images of the generated data with a discriminative dictionary.

This paper presents a novel framework for learning a Bayesian inference graph from a dataset of real world data using a Bayesian model. Such a Bayesian model has the following properties: it can be learned efficiently in an incremental manner, and thus it can be used to explore new Bayesian inference procedures without relying on the standard data-driven approach. Our approach exploits prior knowledge about the underlying data to design its Bayesian inference procedure. We also show that the proposed approach can be used for learning from data in other than the data.

Guided Depth Estimation

On the Complexity of Bipartite Reinforcement Learning

Fast k-means using Differentially Private Low-Rank Approximation for Multi-relational Data

Towards a Unified Framework for 3D Model Refinement

Bayesian Inference via Variational Matrix FactorizationThis paper presents a novel framework for learning a Bayesian inference graph from a dataset of real world data using a Bayesian model. Such a Bayesian model has the following properties: it can be learned efficiently in an incremental manner, and thus it can be used to explore new Bayesian inference procedures without relying on the standard data-driven approach. Our approach exploits prior knowledge about the underlying data to design its Bayesian inference procedure. We also show that the proposed approach can be used for learning from data in other than the data.