Translating information between text and image is a fundamental problem in artificial intelligence that connects natural language processing and computer vision. In the past few years, performance in image caption generation has seen significant improvement through the adoption of recurrent neural networks (RNN). Meanwhile, text-to-image generation begun to generate plausible images using datasets of specific categories like birds and flowers. We’ve even seen image generation from multi-category datasets such as the Microsoft Common Objects in Context (MSCOCO) through the use of generative adversarial networks (GANs). Synthesizing objects with a complex shape, however, is still challenging. For example, animals and humans have many degrees of freedom, which means that they can take on many complex shapes. We propose a new training method called Image-Text-Image (I2T2I) which integrates text-to-image and image-to-text (image captioning) synthesis to improve the performance of text-to-image synthesis. We demonstrate that %the capability of our method to understand the sentence descriptions, so as to I2T2I can generate better multi-categories images using MSCOCO than the state-of-the-art. We also demonstrate that I2T2I can achieve transfer learning by using a pre-trained image captioning module to generate human images on the MPII Human Pose …
Vertical Hoeffding Tree (VHT)
The Vertical Hoeffding Tree (VHT) is a distributed extension of the VFDT (Domingos and Hulten, 2000). The VHT uses vertical parallelism to split the workload across several machines. Vertical parallelism leverages the parallelism across attributes in the same example, rather than across different examples in the stream. In practice, each training example is routed through the tree model to a leaf. There, the example is split into its constituting attributes, and each attribute is sent to a different Processor instance that keeps track of sufficient statistics. This architecture has two main advantages over one based on horizontal parallelism. First, attribute counters are not replicated across several machines, thus reducing the memory footprint. Second, the computation of the fitness of an attribute for a split decision (via, e.g., entropy or information gain) can be performed in parallel. The drawback is that in order to get good performance, there must be sufficient inherent parallelism in the data. That is, the VHT works best for sparse data (e.g, bag-of-words models).
Vertical Hoeffding Tree (VHT) classifier is a distributed classifier that utilizes vertical parallelism on top of the Very Fast Decision Tree (VFDT) or Hoeffding Tree classifier. Hoeffding Tree or VFDT is the standard decision tree algorithm for data stream classification. VFDT uses the Hoeffding bound to decide the minimum number of arriving instances to achieve certain level of confidence in splitting the node. This confidence level determines how close the statistics between the attribute chosen by VFDT and the attribute chosen by decision tree for batch learning. For a more comprehensive summary of VFDT, read chapter 3 of ‘Data Stream Mining: A Practical Approach’.
Modularity is one measure of the structure of networks or graphs. It was designed to measure the strength of division of a network into modules (also called groups, clusters or communities). Networks with high modularity have dense connections between the nodes within modules but sparse connections between nodes in different modules. Modularity is often used in optimization methods for detecting community structure in networks. However, it has been shown that modularity suffers a resolution limit and, therefore, it is unable to detect small communities. Biological networks, including animal brains, exhibit a high degree of modularity. …