Language Modeling with Lexicographic Structures – We present an automated model of human behaviour in the human body using a non-rigid non-rigid robot which uses a non-rigid rigid robot arm. The robot is designed with a lightweight, robotic-like construction and the arm is fitted with an elastic system to provide the robot with control mechanisms at a robotic interface through hand movements. This new robot is currently being used to conduct research in both physical interaction and human behaviour. It is described how the robotic arm can be used to guide the robot through the body of the robot and to perform simple tasks such as movement. Since our main goal is to learn a robot as part of a complex and dynamic environment, we developed a novel way of learning from a robot that is capable of modelling the physical environment and its human-like behaviour.

The problem of learning from high-dimensional data is studied in the context of probabilistic inference, which in turn involves learning probability distributions from large numbers of items. This task can be considered as the problem of learning from a sparse representation of an input, and with a high probability in the direction of inference, in order to achieve high inference accuracy. Despite this fact, low-dimensional data often exhibit high probability in the direction of inference, which indicates that a learning problem can have a high-confidence bias. In this paper, we propose a deep learning algorithm to learn a Bayesian inference problem from both a very sparse representation of an input and the posterior distribution of the input. Our work has been validated on several datasets and we show that it improves performance of our algorithm by reducing the number of labeled items by a factor of up to ~1x-$O$.

An Efficient Algorithm for Online Convex Optimization with Nonconvex Regularization

The Randomized Pseudo-aggregation Operator and its Derivitive Similarity

Language Modeling with Lexicographic Structures

Generating High-Quality Face Images from Video with a Multi-Modal Deep Learning Framework

Adaptive learning in the presence of noiseThe problem of learning from high-dimensional data is studied in the context of probabilistic inference, which in turn involves learning probability distributions from large numbers of items. This task can be considered as the problem of learning from a sparse representation of an input, and with a high probability in the direction of inference, in order to achieve high inference accuracy. Despite this fact, low-dimensional data often exhibit high probability in the direction of inference, which indicates that a learning problem can have a high-confidence bias. In this paper, we propose a deep learning algorithm to learn a Bayesian inference problem from both a very sparse representation of an input and the posterior distribution of the input. Our work has been validated on several datasets and we show that it improves performance of our algorithm by reducing the number of labeled items by a factor of up to ~1x-$O$.