| viXra.org > Artificial Intelligence > 2201 |
 |
 |
| viXra.org > Artificial Intelligence > 2201 |
|
|
[4] viXra:2201.0188 [pdf] submitted on 2022-01-26 03:38:42
Authors: Chengkai Guo, Kai Yang
Comments: 9 Pages.
Preliminary concept of AGI for brain-like intelligence is presented in this paper. The solution is mainly in two aspects: firstly, we combine information entropy and generative network (GAN like) model to propose a paradigm of General Intelligent Network (GIN). In the GIN network, the original multimodal information can be encoded as low information entropy hidden state representations (HPPs), which can be reverse parsed by the contextually relevant generative network into observable information. Secondly,we propose a generalized machine learning operating system (GML system), which includes an observable processor (AOP), an HPP storage system, and a multimodal implicit sensing/execution network. Our code will be released at https://github.com/ggsonic/GIN
Category: Artificial Intelligence
[3] viXra:2201.0177 [pdf] submitted on 2022-01-25 19:40:24
Authors: Manish Bhargav, Satish Kumar Alaria, Manish Kumar Mukhija
Comments: 10 Pages.
Twitter has turned into a tiny source of dynamic data for blogging places. People post on a wide range of topics and constantly communicate their assumptions, discuss current concerns, and positively review what they use in their daily lives on Twitter wall. The main goal is to assess the emotions expressed in tweets using various machine learning algorithms that identify tweets as positive or negative. If the tweet contains both negative and positive elements, the most dominant component should be chosen as the final component. In tweets, emojis, usernames, and hashtags must be managed and translated into a standard structure. Bigrams and unigrams, for example, must be removed as well. In any case, just relying on a single model, which did not give high accuracy, is taken into account when selecting a model with high precision. To be honest, organizers for these items have begun to investigate these modest internet journals (blogs) in order to get a general sense of their item. They frequently monitor and reply to client comments on smaller websites. One issue is coming up with new ways to recognize and abbreviate a broad sentiment. Several persons, such as Facebook, Twitter, and Instagram, were brought into interpersonal connection stages as recently as last year. Most people use social media to convey their feelings, ideas, or assumptions about objects, places, or people. Strategies Twitter, a micro-blogging platform, is a massive repository of public opinion for a variety of people, offers, businesses, and products, among other things. The public analysis system evaluations are known as sentiment assessment. Combination of sentiment analysis on Twitter give valuable context to what's being said on Twitter. The wide availability of internet exams and social media postings the media provides critical criticism to organizations in order to improve expert options and steer their marketing tactics to leisure and user selections. As a result, social media plays a key role in influencing the public's perception of the services or products chosen. The numerous tactics utilized for product classification critiques are highlighted in this study (which may be in the form of tweets) Tweet complaints to see if mass behaviour is positive, negative, or neutral. Analysis of the Product Market. The information used here comes from our Twitter product reviews, which were used to categorize opinions as satisfying.
Category: Artificial Intelligence
[2] viXra:2201.0144 [pdf] replaced on 2023-02-09 18:52:37
Authors: Dimiter Dobrev
Comments: 92 Pages.
Artificial Intelligence — What is it, how can we do it and what shall we do once we do it? This is a PhD thesis.
Category: Artificial Intelligence
[1] viXra:2201.0094 [pdf] submitted on 2022-01-16 15:17:12
Authors: Jai Sharma, Milind Maiti, Christopher Sun
Comments: 17 Pages.
Cardiovascular disease causes 25% of deaths in America (Heart Disease Facts). Specifically, misdiagnosis of cardiovascular disease results in 11,000 American deaths annually, emphasizing the increasing need for Artificial Intelligence to improve diagnosis. The goal of our research was to determine the probability that a given patient has Cardiovascular Disease using 11 easily-accessible objective, examination, and subjective features from a data set of 70,000 people. To do this, we compared various Machine Learning and Deep Learning models. Exploratory Data Analysis (EDA) identified that blood pressure, cholesterol, and age were most correlated with an elevated risk of contracting heart disease. Principal Component Analysis (PCA) was employed to visualize the 11-D data onto a 2-D plane, and distinct aggregations in the data motivated the inference of specific cardiovascular conditions beyond the binary labels in the data set. To diagnose patients, several Machine Learning and Deep Learning models were trained using the data and compared using the metrics Binary Accuracy and F1 Score. The initial Deep Learning model was a Shallow Neural Network with 1 hidden layer consisting of 8 hidden units. Further improvements, such as adding 5 hidden layers with 8 hidden units each and employing Mini-Batch Gradient Descent, Adam Optimization, and He’s Initialization, were successful in decreasing train times. These models were coded without the utilization of Deep Learning Frameworks such as TensorFlow. The final model, which achieved a Binary Accuracy of 74.2% and an F1 Score of 0.73, consisted of 6 hidden layers, each with 128 hidden units, and was built using the highly optimized Keras library. While current industrial models require hundreds of comprehensive features, this final model requires only basic inputs, allowing versatile applications in rural locations and third-world countries. Furthermore, the model can forecast demand for medical equipment, improve diagnosis procedures, and provide detailed personalized health statistics.
Category: Artificial Intelligence
| Third-Party Links: |
|