Chapter 5 Discussion Questions: Artificial intelligence
Chapter 5 Discussion Questions
- An artificial neural network is a part of Artificial intelligence which is made of neurons as the basic processing unit and make computers behave intelligently like a human brain. It is programmed in such a way that it has millions of artificial neurons arranged in series of layers. It can be used across a large range of problem domains such as handwriting recognition, in solving travelling salesman problems to find the shortest path to travel in cities in a given area, in image compression, in stock exchange prediction by investors, in pattern and sequence recognition systems, robotics, data processing and modelling in ventilating and air-conditioning systems, refrigeration, load-forecasting, control of power-generation systems and solar radiation and in complex mapping and system identification.
- Both biological and artificial neural networks are deep learning technologies which are applied to solve a wide range of problems. Besides they are both made of neurons and several layers although at different variations. However, there are some key aspects of biological networks that cannot be mimicked by the artificial ones. Size is one key aspect. Biological neural networks contain about 86bilion neurons and above 100trillion synapses, while artificial networks have between 10-1000milion neurons(Chung et al.,2021).Topology: all artificial neural networks Layers aren’t connected to non-neighboring layers thus their neurons cant fire inparallel,however biological networks on the other hand have, neurons that can fire asynchronously in parallel.Speed:biological neurons can fire around 200 times a second on average as opposed to the artificial neural networks which fire at a lower rate per second. Fault-tolerance: biological neuron networks due to their topology are also fault-tolerant. Power consumption: the brain consumes about 20% of all the human body’s energy — despite its large cut, an adult brain operates on about 20 watts. While, an artificial neural network runs on 250 watts. Unlike the biological neural network, artificial neural networks don’t learn by recalling information — they only learn during training, but will always “recall” the same, learned answers afterwards, without making a mistake.
- There are various artificial neural network architectures for different types of problems. According to Zhang, (2020) the most common artificial neural networks include; The Multilayer Perceptron which is designed for estimating continuous functions and can also be used in solving problems that are nonlinear separable. It can be applied in several cases, such as pattern classification, prediction, recognition, and approximation (Menzies et al., 2015). Convolutional Neural Networks architectures, designed to solve computer vision problems such as image detection or image classification. It can be applied in, object detection in a self-driving car, social media face recognition and image analysis in the medical field (Géron, 2019). Recurrent Neural Networks, used in handling sequential data such as time series and natural language processing. It is applied in, machine translation, sentiment analysis, text summarization, document generation and chatbots. Generative Adversarial Networks, used to generate new data with similar characteristics as the data provided during training (Abirami & Chitra, 2020).
- Artificial neural networks learn in the supervised mode by the help of well-labeled attributes used to train the model. The model is allowed to learn to discover information within data using ground facts; in other words, using previous knowledge collected on the output values as a sample (Le, 2018). In contrast, in unsupervised learning, the model does not need to be fixed and thus it learns by discovering all complex tasks within the data as opposed to supervised learning.
- Support vector machines is a machine learning technique implemented through the soft margin because of its attractive characteristics. It is mostly used because of its superior predictive power which is as a result of structural risk minimization. It can be used in dealing with nonlinear problems by kernel methods. It’s applied in many fields of science and engineering such as fault diagnosis and condition monitoring. Support vector machines work by mapping data to a high-dimensional feature space for data points to be categorized even when the data are not linearly separable by the use of kernel functions for the transformation.
- Maximum margin hyperplanes are the building block that is used in directing to build optimally single output classes from each other by making use of training data. Each of the classes are separated by a maximum of two-dimensional hyperplanes(n-1) where “n”is the number of labels in a class. Maximum margin hyperplanes are very paramount in support vector machines as they help optimize the accuracy and reliability of the support vector machine predictive power.
- The optimal kernel type and kennel parameters can be determined by; Structural risk minimization, optimizing hyperplane in a kernel space where training instances are linearly separable, setting clear margin separation, providing the support vector machine with limited data, and finding a properly adjusted parameter in kernel functions by the use of grid search, such as s in the Gaussian radial
- Using Artificial Neural Network or support vector machines is a very promising approach in the stock market because of several capabilities this technology possess. The algorithms are designed to understand complicated issues that cannot be undertaken by basic machine learning algorithms or accessible neural networks. They are made by utilizing many novel metaheuristic algorithms like social spider optimization as well as the bat algorithm. Also, have stock prediction accuracy of about 60%–70% because of its use of integrating techniques such as Random Forest and Genetic Algorithms. However, they are faced by a few caveats which can make their predictions fail, they are not able to predict stock prices across time zones, they are faced by cold start problems and at times suffer from multiple local minima,
Questions For the Opening Vignette in Chapter 5
- It is important to study medical procedures so as to improve operational effectiveness and efficiency. This in return supports accurate and timely decision making when predicting outcomes.
- The most important factors to consider in better understanding and managing health care is Understanding how to use predictive and explanatory analysis of large and feature-rich data sets which provides invaluable information to make more efficient and effective decisions managerial and clinical decisions.
- The impact of predictive modelling on healthcare and medicine is increasing the effectiveness and efficiency of the managerial and medical personnel to make the accurate decisions in treatment of diseases and management but cannot replace either medical or managerial personnel.
- According to Delen et al. (2012) the results indicated that data mining was a superior tool in predicting the outcome and in analyzing the prognostic factors of complex medical procedures such as CABG surgery. Besides, it was also evident that using more than one prediction methods produced more reliable results than just using one prediction method. With SVM producing the best prediction accuracy at 88% on the data sample.
- In a 2018 study conducted by KP and the Mental Health Research Network, on the risk of suicide because of brain problems and found a positive relationship between brain problems and suicide (Woelbert et al., 2018). At Wake Forest Baptist Health in North Carolina, analytics tools helped the oncology infusion center anticipate peak utilization timesand adjust its scheduling practices accordingly, said Karen Craver, Clinical Practice Administrator.
References
Chung, S., & Abbott, L. F. (2021). Neural population geometry: An approach for understanding biological and artificial neural networks. Current opinion in neurobiology, 70, 137-144.
Zhang, Q., Deng, D., Dai, W., Li, J., & Jin, X. (2020). Optimization of culture conditions for differentiation of melon based on artificial neural network and genetic algorithm. Scientific Reports, 10(1), 1-8.
Woelbert, E., Lundell-Smith, K., White, R., & Kemmer, D. (2021). Accounting for mental health research funding: developing a quantitative baseline of global investments. The Lancet Psychiatry, 8(3), 250-258.