Action recognition with skeleton data has recently attracted much attention in computer vision. Previous studies are mostly based on fixed skeleton graphs, only capturing local physical dependencies among joints, which may miss implicit joint correlations. To capture richer dependencies, we introduce an encoder-decoder structure, called A-link inference module, to capture action-specific latent dependencies, i.e. actional links, directly from actions. We also extend the existing skeleton graphs to represent higherorder dependencies, i.e. structural links. Combing the two types of links into a generalized skeleton graph, we further propose the actional-structural graph convolution network (AS-GCN), which stacks actional-structural graph convolution and temporal convolution as a basic building block, to learn both spatial and temporal features for action recognition. A future pose prediction head is added in parallel to the recognition head to help capture more detailed action patterns through self-supervision. We validate AS-GCN in action recognition using two skeleton data sets, NTU-RGB+D and Kinetics. The proposed AS-GCN achieves consistently large improvement compared to the state-of-the-art methods. As a side product, AS-GCN also shows promising results for future pose prediction. Our code is available at https://github.com/limaosen0/AS-GCN . 1

No abstract

No abstract

No abstract

Glioblastoma multiforme (GBM) is a highly aggressive type of brain cancer with very low median survival. In order to predict the patient's prognosis, researchers have proposed rules to classify different glioma cancer cell subtypes. However, survival time of different subtypes of GBM is often various due to different individual basis. Recent development in gene testing has evolved classic subtype rules to more specific classification rules based on single biomolecular features. These classification methods are proven to perform better than traditional simple rules in GBM prognosis prediction. However, the real power behind the massive data is still under covered. We believe a combined prediction model based on more than one data type could perform better, which will contribute further to clinical treatment of GBM. The Cancer Genome Atlas (TCGA) database provides huge dataset with various data types of many cancers that enables us to inspect this aggressive cancer in a new way. In this research, we have improved GBM prognosis prediction accuracy further by taking advantage of the minimum redundancy feature selection method (mRMR) and Multiple Kernel Machine (MKL) learning method. Our goal is to establish an integrated model which could predict GBM prognosis with high accuracy.

No abstract

Interleaving experiments are an attractive methodology for evaluating retrieval functions through implicit feedback. Designed as a blind and unbiased test for eliciting a preference between two retrieval functions, an interleaved ranking of the results of two retrieval functions is presented to the users. It is then observed whether the users click more on results from one retrieval function or the other. While it was shown that such interleaving experiments reliably identify the better of the two retrieval functions, the naive approach of counting all clicks equally leads to a suboptimal test. We present new methods for learning how to score different types of clicks so that the resulting test statistic optimizes the statistical power of the experiment. This can lead to substantial savings in the amount of data required for reaching a target confidence level. Our methods are evaluated on an operational search engine over a collection of scientific articles.

No abstract