Artificial Intelligence

2211 Submissions

[5] viXra:2211.0124 [pdf] submitted on 2022-11-21 01:15:22

Representing a Neural Network as a Ggraphed Set

Authors: Ho Yeol Choi
Comments: 5 Pages. (Note by viXra Admin: Please avoid hand-drawing and write compete article with scientific references!)

I studied how to implement general neural network weights. The overlapping intersection between sets has a high signal ratio. What I'm trying to say is that weight gain in conventional neural networks is what happens in the part of the intersection between sets.
Category: Artificial Intelligence

[4] viXra:2211.0106 [pdf] submitted on 2022-11-19 04:49:26

Generic Natural Language Distance Via Online Semantic Volumetric Inference

Authors: Alex-Pauline Poudade, Pascal Rabier, Neau-Monier Sarah, Olivier Poudade, Grimault Valérie, Emmanuel Martins, Ludwig De Sousa
Comments: 9 Pages. Data at https://doi.org/10.7910/DVN/WKLWF8

This paper discusses the approach of creating semantic meaning ad hoc through direct explicit volumetric adherence or relative intersection, from online databases, such as Wikipedia or Google. We demonstrate this approach through use of correlation, between a dictionary index — a lexicon - and an import/export industry ISO A129 standard used by the Ministry of Finances, in the French language. We conclude, this approach by giving the most and least meaningful industrial results, for the French language. This questions whereas online apparent generic Natural language processing (NLP) pivot Chomsky Universal grammar (UG) representation, could inherit implicit initial national culture. https://doi.org/10.7910/DVN/WKLWF8 (2022-11-18)
Category: Artificial Intelligence

[3] viXra:2211.0054 [pdf] submitted on 2022-11-10 01:32:55

Anomalous Payload Detection System Using MUXConv Neural Network with Parameter Optimization

Authors: CholRyong Pak, HakMyong O, HyokChol U, Hun Nam
Comments: 7 Pages.

This paper proposes how to detect malicious network data in effective and accurate way using MUXConv neural network(MUXCNN) with parameter optimization. First of all, in order to increase detection speed, packets are directly entered into the input of MUXCNN without decoding. Next of all, after training MUXCNN with learning data, we judge that its traffic is normal or abnormal. Simulations and experiments show that the proposed abnormal network-detecting system is more efficient in detection and higher in accuracy than the other multi-layer neural networks.
Category: Artificial Intelligence

[2] viXra:2211.0015 [pdf] submitted on 2022-11-03 01:50:04

The Acceleration of Multi-Factor Merton Model on FPGA

Authors: Pengyu Guo
Comments: 66 Pages.

Credit risk stands for the risk of losses caused by unwanted events, such as the default of an obligor. The managing of portfolio credit risks is crucial for financial institutions. The multi-factor Merton model is one of the most widely used tools that modelling the credit risks for financial institutions. Typically, the implementation of the multi-factor Merton model involves Monte Carlo simulations which are time-consuming. This would significantly restrict its usability in daily credit risk measurement. In this report, we propose an FPGA architecture for credit-risk measurements in the multi-factor Merton models. The presented architecture uses a variety of optimization techniques such as kernel vectorization and loop unrolling, to optimize the performance of the FPGA implementation. The evaluation results show that compare to a basic C++ implementation running on a single-core Intel i5-4210 CPU, our proposed FPGA implementation can achieve an acceleration of up to 22 times, with a precision loss of less than 10−8.
Category: Artificial Intelligence

[1] viXra:2211.0014 [pdf] submitted on 2022-11-03 01:50:31

Parallel Parameter Estimation for Gilli-Winker Model Using Multi-Core CPUs

Authors: Pengyu Guo
Comments: 36 Pages.

Agent-based modeling is a powerful tool that is widely used to model global financial systems. When the parameters of the model are appropriate, the price time series generated by the model exhibit marked similarities with actual financial time series and even reproduces some of their statistical characteristics.By using Kirman’s Ant model as a prototype, this report systematically explored Gilli and Winker’s parameter optimization method. In view of some limitations of this method, this report promoted some improvements, including a local-restart strategy to enhance the convergence ability of the original optimization method, as well as incorporate Simulated Annealing into the original method to help the algorithm escape from local optimums. Furthermore, since the parameter optimization of agent-based modeling tends to be very time-consuming, an acceleration method was also proposed to speed up this procedure. In the end, the presented methods have been validated with the EUR/USD exchange rate.
Category: Artificial Intelligence