Efficient Learning with Label-Dependent Weight Functions – We present the first ever dataset of the full word labels in the context of machine learning (ML) classification. By modeling the label distribution under the full word label distribution, we propose a novel and practical learning algorithm that combines Bayesian and Bayesian methods. We show the advantage of our algorithm, using the new training dataset as our dataset, and also show that the new dataset will provide a valuable framework for analyzing and designing a Bayesian ML methodology. The dataset is composed of 5,700 annotated sentences from 2,000+ annotated datasets. For each annotation, we propose a label-dependent weight function, and test it on various datasets, while incorporating a data-driven approach for learning. The experimental results show that the proposed approach achieves state-of-the-art performance when evaluated by using the same label distribution (without using any label labels). We also present experiments showing that the proposed method generalizes well to a variety of ML tasks, including learning to classify trees and the estimation of word embeddings.

We propose a new deep learning based technique aimed at solving traffic prediction problems with limited knowledge about the current traffic. The existing work on deep networks and deep learning in general uses the maximum available information from the underlying networks, and thus does not properly model the network structures and dynamics. Besides, traffic prediction in this setting is not limited either. This paper is an application of deep learning to a multi-modal learning problem, where the data consists of multi-modal traffic signals. First, we propose a method to learn a prediction network which is capable of predicting a user’s preferences (i.e., road trip time, speed and lane width). Then we also consider the possibility of using different traffic signals to learn a neural network, and the network of such a network is shown to be well-defined. Finally, we use a multi-modal machine learning model called ResNet to predict the road trip time, which is described as a time-dependent function and can be used as a basis for a prediction network.

A Survey of Recent Developments in Human Action Recognition

Visual concept learning from concept maps via low-rank matching

Efficient Learning with Label-Dependent Weight Functions

Reinforcement Learning with External Knowledge

Deep Learning for Real-Time Traffic Prediction and ClusteringWe propose a new deep learning based technique aimed at solving traffic prediction problems with limited knowledge about the current traffic. The existing work on deep networks and deep learning in general uses the maximum available information from the underlying networks, and thus does not properly model the network structures and dynamics. Besides, traffic prediction in this setting is not limited either. This paper is an application of deep learning to a multi-modal learning problem, where the data consists of multi-modal traffic signals. First, we propose a method to learn a prediction network which is capable of predicting a user’s preferences (i.e., road trip time, speed and lane width). Then we also consider the possibility of using different traffic signals to learn a neural network, and the network of such a network is shown to be well-defined. Finally, we use a multi-modal machine learning model called ResNet to predict the road trip time, which is described as a time-dependent function and can be used as a basis for a prediction network.