درخت حوزه‌های تخصصی

Due to the rapid advancements in computer technology, researchers are attracted to solving challenging problems in many different fields. The price of rainbow options is an interesting problem in financial fields and risk management. When there is no closed-form solution to some options, numerical methods must be used. Choosing a suitable numerical method involves the most appropriate combination of criteria for speed, accuracy, simplicity and generality. Monte Carlo simulation methods and traditional numerical methods have expensive repetitive computations and unrealistic assumptions on the model. Deep learning provides an effective and efficient method for options pricing. In this paper, the closed-form formula or Monte-Carlo simulation are used to generate data in European and Asian rainbow option prices for the deep learning model. The results confirm that the deep learning model can price the rainbow options more accurately with less computation time than Monte-Carlo simulation.

Human Identification has been widely implemented to enhance the efficiency of surveillance systems, however, systems based on common CCTV (closed-circuit television) cameras are mostly incompatible with the advanced identification algorithms which aim to extract the facial features or speech of an individual for identification. Gait (i.e., an individual’s unique walking pattern/style) is a leading exponent when compared to first-generation biometric modalities as it is unobtrusive (i.e., it requires no contact with the individual), hence proving gait to be an optimal solution to human identification at a distance. This paper proposes an automatic identification system that analyzes gait to identify humans at a distance and predicts the strength of the match (i.e., probability of the match being positive) between two gait profiles. This is achieved by incorporating computer vision, digital image processing, vectorization, artificial intelligence, and multi-threading. The proposed model extracts gait profiles (from low-resolution camera feeds) by breaking down the complete gait cycle into four quarter-cycles using the variations in the width of the region-of-interest and then saves the gait profile in the form of four distinct projections (i.e., vectors) of length 20 units each, thus, summing up to 80 features for each individual’s gait profile. The focus of this study revolved around the speed-accuracy tradeoff of the proposed model where, with a limited dataset and training, the model runs at a speed of 30Hz and yields 85% accurate results on average. A Receiver Operating Characteristic Curve (ROC) is obtained for comparison of the proposed model with other machine learning models to better understand the efficiency of the system