Sign language bridges the communication between the deaf-mute and the rest of the world. With the development of computer technology, the research on sign language recognition is prospering. Currently, there are two kinds of research methods. One is based on machine vision, the other is based on wearable input devices. Comparing with machine vision, wearable devices have the advantage of being able to get real-time information of the bend of the fingers and the movement of the hand. The essay applies recent technology in the field of voice recognition, and aims at finding out the best algorithm for medium vocabulary continuous sign language recognition. By building a hybrid Deep Neural Network-Hidden Markov Model, and designing and making a low-cost digital glove, the essay successfully compares the features between Dynamic Time Warping (DTW), Gaussian Mixture Model-Hidden Markov Model (GMM-HMM) and Deep Neural Network-Hidden Markov Model (DNN-HMM) as of solving the problem of sign language recognition. Tests show that in terms of calculating the observation probability, DNN has a much better performance than HMM, especially when syntax is not provided, meaning that DNN better suits the developing trend of sign language recognition. Real-time recognition is achieved on the intelligent terminal with an accuracy of over 97%, using the trained model and a decoding program.

Our paper begins with a simple but beautiful subject given by Martin Gardner (Dividing ten gold coins and ten silver coins into two containers which look the same outside, then randomly choose one of the containers and take out one coin from the chosen container, Is the probability of getting golden coin becomes higher when the distributive method changes). And we change and consider the problem from some other points of view. 1. Researching the original subject’s probability. 2. Discussing the minimal unit of distributing golden coins. 3. Researching the number of containers. 4. Researching the minimal and maximal probability of getting golden coin which is the function expression of containers’ number. As a result, we will understand the subject more deeply. The essence of the subject tells us: the reason of probability’s changing can not only be interpreted by “Sample Space difference and Sample Point difference”, but also “Classical Probability’s infection to Geometry Probability”. According to this theory, we can model this subject: Gold→A and Silver→not A. Then we use the model in our example, successfully interpreting what J·BERTRAND had found. And we also get a conjecture and prove it. Further, we did some transformation of the original formulation according to the actual situation. This is our paper’s main content.

The Durfee conjecture, proposed in 1978, relates two important invariants of isolated hypersurface singularities by a famous inequality; however, the inequality in this conjecture is not sharp. In 1995, Yau announced his conjecture which proposed a sharp inequality. The Yau conjecture characterizes the conditions under which an affine hypersurface with an isoalted singularity at the origin is a cone over a nonsingular projective hypersurface; in other words, the conjecture gives a coordinate-free characterization of when a convergent power series is a homogeneous polynomial after a biholomorphic change of variables. In this project, we prove that the Yau conjecture holds for n = 5. As a consequence, we have proved that &5!pg≤μ-p(v)&, where &p(v) = (v-1)^5-v(v-1) ……(v-4)& and pg,μ,and v are, respectively, the geometric genus, the Milnor number, and the multiplicity of the isolated singularity at the origin of a weighted homogeneous polynomial. In the process, we have also defined yet another sharp upper bound for the number of positive integral points within a 5-dimensional simplex.

The development of modern technology is closely related to the use of metals, whose demand has been shifting from common industrial metals to a variety of minor metals, such as cobalt or indium. The industrial importance and limited geological availability of some minor metals have led to them being considered more “critical.” In this research, we develop a novel approach to investigate metal criticality. Instead of quantifying the supply risk of critical metals with static indicators, as used in conventional criticality measures, we propose a dynamic method by looking at the stock trend of the major critical metal producers, who would, in turn, influence the future supply. This allows us to utilize high-frequency time series data in creating machine learning models that predict the stock prices using metal-related features. Specifically, we use regularized linear regression, logistic regression, support vector machine and gradient boosted trees as base models. An ensemble of them is then used for the final prediction, with methods including majority vote, logistic regression stacking, and gradient boosted trees stacking. Based on a misclassification error of 34% in the validation set, we further develop a stock trading strategy, which leads to a back-tested return of 313%, or an excess return of 147%. More importantly, a set of significant features selected by our models leads to insightful suggestions that certain unseeming features, such as markets of other metal cycles and currency exchange rates, may play important roles in influencing the supply of critical metals.

No abstract uploaded!