Real-time analysis of behavior of law enforcement encounters
USING BIG DATA ANALYTICS AND DEEP LEARNING MULTIMODAL EMOTION-RECOGNITION MODELS
With the resurgence of the Artificial Neural Networks (ANNs) in its new avatar as Deep Neural Networks (DNNs) it has become possible to represent the models in access of the traditional 3-4 hidden layers. These architectures have helped shatter previous records in image and speech recognition and is helping to bridge the chasm between humans and machines. DNNs with its suite of flavors in Backpropogation, Feedforward, Convolutional and Recurrent Neural Networks are currently held among the most promising building blocks towards the still elusive goal of achieving Artificial Intelligence.
Along with this there has been key advances in extensive use and availability of economic GPU systems and libraries and the development of symbolic differentiation compilers that can automatically compute and yield very efficient low level code for the highly complex gradients associated to very general Feedforward architectures, specially, suited to concrete problems which are now routinely proposed. These advances have also resulted in availability of open source platforms such a TensorFlow, Theano, Caffe with support of wrappers like Kaldi and Keras that run top of these platforms.