Artificial Intelligence

2207 Submissions

[4] viXra:2207.0146 [pdf] submitted on 2022-07-26 01:08:50

Generalized Attention Mechanism and Relative Position for Transformer

Authors: R. V. R. Pandya
Comments: 6 Pages.

In this paper, we propose generalized attention mechanism (GAM) by first suggesting a new interpretation for self-attention mechanism of Vaswani et al. . Following the interpretation, we provide description for different variants of attention mechanism which together form GAM. Further, we propose a new relative position representation within the framework of GAM. This representation can be easily utilized for cases in which elements next to each other in input sequence can be at random locations in actual dataset/corpus.
Category: Artificial Intelligence

[3] viXra:2207.0064 [pdf] replaced on 2022-07-22 00:19:25

Lyrics-Based Music Band and Genre Topic Similarity Analysis

Authors: Dimitrios Geromichalos
Comments: 10 Pages. Updated version

Based on hundreds of thousands of song lyrics from thousands of bands, Word2Vec models have been trained to quantitatively identify similarities between band texts and terms. Using prominent examples, this demonstrates for the cases studied, that music bands can be assigned to a similarity network solely on the basis of their song lyrics, which also corresponds to their musical style. Furthermore, using exemplary words, it is demonstrated that semantic term networks vary strongly from genre to genre. In addition, the semantic similarity matrices were studied using network analysis methods. As it turned out, term and band text networks differ significantly. While the former resemble random networks, the latter partly exhibit powerlaw behavior. Both also exhibit threshold-dependent regimes.
Category: Artificial Intelligence

[2] viXra:2207.0062 [pdf] submitted on 2022-07-08 16:38:56

Wave Function Collapse Visualization

Authors: Vishal Pandey, Ishanvi Pandey
Comments: 7 Pages.

Wave Function Collapse initializes output bitmapin a completely unobserved state, where each pixel value is in a superposition of colors of the input bitmap (so if the input was black-white then the unobserved states are shown in different shades of grey). The coefficients in these superpositions are real numbers, not complex numbers, so it doesn’t do the actual quantum mechanics, but it was inspired by QM. In this, we have been matching each tile to tile value by pixel to pixel by namingas it as "socket". We know that in code when we match the tile it would be in a random order so we had rotated them into a specific order to match each socket to socket which indicates the overlapping of tiles as the superposition of several Eigen states. It was first introduced in 2016 by Maxim Gumin which can generate procedural patterns from a sample image or from a collection of tiles. So we are just visualizing it in a mathematical way
Category: Artificial Intelligence

[1] viXra:2207.0056 [pdf] submitted on 2022-07-07 23:38:20

Designing Potential Drugs That Can Target Sars-COV-2’s Main Protease: A Proactive Deep Transfer Learning Approach Using LSTM Architecture

Authors: Omar Dasser, Moad Tahri, Louay Kila, Abderrahim Sekkaki
Comments: 23 Pages.

Drug discovery is a crucial step in the process of delivering a new drug to the market that can take up to 2-3 years which can be more penalizing given the current global pandemic caused by the outbreak of the novel coronavirus SARS-CoV 2. Artificial Intelligence methodologies have shown great potential in resolving tasks in various domains such as image classification, sound recognition, also in the range of the previous years, Artificial Intelligence proved to be the go-to for generative tasks for use cases such as music sequences, text generation and solving also problems in biology. The goal of this work is to harvest the power of these architectures using generative recurrent neural network with long short-term memory (LSTM) gating techniques in order to generate new and non-existing molecules that can bind to the main COVID-19 protease, which is a key agent in the transcription and replication of the virus, and thus can act as a potential drug that can neutralize the virus inside of an infected host. As of today, there are no specific targeted therapeutic agents to treat the disease and all existing treatments are all very limited. Known drugs that are passing clinical trials such as Hydroxychloroquine and Remdesivir showed respectively a binding energy with SARS-CoV-2’s main protease of -5.3 and -6.5, the results of the newly generated molecules exhibited scores ranging till -13.2.
Category: Artificial Intelligence