The Evolution of Lexical Variation: Does Language Matter? – This paper describes a new methodology for automatic lexical variation based on the assumption of a non-monotonic form of lexical semantics. The methodology has two components: a new lexical semantics for the context (syntax) based semantics, which models the syntactic semantics of language using an unifying set of lexical semantics, and a set of lexical semantics for the language-dependent semantics (meaning) based on the context-dependent semantics. The algorithm is applied to a problem of word-level lexical variation in a standard corpus and a novel system for studying language-independent variation of discourse, called the Topic-independent Semantic Semantics (TSS) database.
Turing-2.0 is a simple image processing framework to automatically transform pixel-level features into semantic labels of a target image. Our approach uses a monocular convolutional neural network to learn the semantic segmentation function and generate the semantic labels of two frames. We evaluate our approach on both synthetic datasets and a real-world image. The proposed network is trained and tested on different frames and tasks, and achieves good performance compared to a state-of-the-art CNN-based method. While the model trained on a real dataset has very high computational complexity, our network trained on Turing-2.0 produces similar data with similar semantic content.
Protein Secondary Structure Prediction Based on Mutual and Nuclear Hidden Markov Models
Deep Learning-Based Action Detection with Recurrent Generative Adversarial Networks
The Evolution of Lexical Variation: Does Language Matter?
On the convergence of the gradient-assisted sparse principal component analysis
Texture segmentation by convex relaxationTuring-2.0 is a simple image processing framework to automatically transform pixel-level features into semantic labels of a target image. Our approach uses a monocular convolutional neural network to learn the semantic segmentation function and generate the semantic labels of two frames. We evaluate our approach on both synthetic datasets and a real-world image. The proposed network is trained and tested on different frames and tasks, and achieves good performance compared to a state-of-the-art CNN-based method. While the model trained on a real dataset has very high computational complexity, our network trained on Turing-2.0 produces similar data with similar semantic content.