Deep Learning of Spatio-temporal Event Knowledge with Recurrent Neural Networks – We propose a novel algorithm for the automatic retrieval of spatio-temporal temporal dependencies in a real-time manner. We present efficient and interpretable algorithms for different domain-specific spatio-temporal dynamics. We test our algorithms on both synthetic and real world data sets. Finally, we show how to use our algorithms to build a neural network that models and predicts future spatio-temporal temporally dependent behaviors.

We present a novel method for generating sentence-level sentences by applying the recently-developed word embeddings to the sentence embedding network which combines word embeddings with a deep recurrent neural network. We train these deep recurrent neural network models on an image corpus where we learn to model the sentence structure over a short period of time. Our approach successfully generates sentences which are consistent with a given corpus with at most a few tens of thousands phrases. Our method has been applied to different tasks using various datasets including video, image and image-based tasks. We show that our approach is particularly robust when dealing with long term dependencies in a noisy environment such as a video or a sentence. We show that the model outperforms a baseline CNN model by an average of 4.5-7.2 TFLOPs per sentence. The task-specific results are also presented and compared to the CNNs that produce short duration sentences.

Convex Penalized Bayesian Learning of Markov Equivalence Classes

A novel approach to natural language generation

Deep Learning of Spatio-temporal Event Knowledge with Recurrent Neural Networks

Machine Learning Methods for Multi-Step Traffic Acquisition

Multi-dimensional representation learning for word retrievalWe present a novel method for generating sentence-level sentences by applying the recently-developed word embeddings to the sentence embedding network which combines word embeddings with a deep recurrent neural network. We train these deep recurrent neural network models on an image corpus where we learn to model the sentence structure over a short period of time. Our approach successfully generates sentences which are consistent with a given corpus with at most a few tens of thousands phrases. Our method has been applied to different tasks using various datasets including video, image and image-based tasks. We show that our approach is particularly robust when dealing with long term dependencies in a noisy environment such as a video or a sentence. We show that the model outperforms a baseline CNN model by an average of 4.5-7.2 TFLOPs per sentence. The task-specific results are also presented and compared to the CNNs that produce short duration sentences.