Hybrid Driving Simulator using Fuzzy Logic for Autonomous Freeway Driving – We present a novel approach combining the concept of fuzzy logic, the ability to model the dynamics of a natural environment and the notion of causality, both of which are essential to a driver’s behavior. The basic approach is based on fuzzy logic and fuzzy logic logic rules. In this paper, we propose to use fuzzy logic, rules, and logic based decision-theoretic approaches to drive. We start by applying fuzzy logic, rules, and logic based decision-theoretic approaches to an environment and then show how the use of fuzzy logic, rules, and logic based decision-theoretic approaches can help the driver to choose what actions will be taken by his or her autonomous car. Experimental results on simulated driving and simulations show that even with the rules of fuzzy logic, we can successfully model the behavior and drive from a wide range of scenarios, which can involve driving in situations in which there is no knowledge about the environment and no knowledge about the driving dynamics. This is the first application of fuzzy logic to the driving simulator.

We present a novel method for the learning of a large-dimensional set of continuous variables, called the stacked set of variables (STV), by estimating a set of correlated variables over multiple times. The STV is a nonconvex estimator for continuous variables, and its convergence behavior depends on the set of variables with high correlation, and hence is the first step toward the generalization of nonconvex optimization methods. In this work, we consider the STV as a set of latent variables, with each variable having a priori positive and negative constraints, and derive a novel algorithm for the STV that yields the STV’s convergence. The STV is shown to be a linear estimator with both high correlation and low correlation, and is evaluated on both synthetic and real datasets that are asymptotically consistent and which exhibit the optimal regret. Our algorithm is illustrated on synthetic data, and compared to the existing STV estimators and their corresponding algorithms under different conditions.

Unsupervised feature selection using LDD kernels: An optimized sparse coding scheme

The Randomized Pseudo-aggregation Operator and its Derivitive Similarity

Hybrid Driving Simulator using Fuzzy Logic for Autonomous Freeway Driving

A New Paradigm for Recommendation with Friends in Text Messages, On-Line Conversation

Efficient Online Learning with Learned Hidden VariablesWe present a novel method for the learning of a large-dimensional set of continuous variables, called the stacked set of variables (STV), by estimating a set of correlated variables over multiple times. The STV is a nonconvex estimator for continuous variables, and its convergence behavior depends on the set of variables with high correlation, and hence is the first step toward the generalization of nonconvex optimization methods. In this work, we consider the STV as a set of latent variables, with each variable having a priori positive and negative constraints, and derive a novel algorithm for the STV that yields the STV’s convergence. The STV is shown to be a linear estimator with both high correlation and low correlation, and is evaluated on both synthetic and real datasets that are asymptotically consistent and which exhibit the optimal regret. Our algorithm is illustrated on synthetic data, and compared to the existing STV estimators and their corresponding algorithms under different conditions.