20.1 Neural Networks and Human Brain

The inspiration to build artificial neural networks (ANNs) has come from the deep desire to simulate the working of the human brain. As our understanding of the human brain is getting enriched and improved daily due to the ongoing research on this topic, researchers are also improving the ANNs accordingly. One such recent addition is the recurrent neural networks (RNNs) and the long short-term memory (LSTM). In this chapter, we will discuss these developments in a simplified manner so that our readers can understand this amazing technology and use it for their development projects.

Sections 20.1 and 20.2 draw their inspiration from the blog titled “The Ultimate Guide to Recurrent Neural Networks (RNN)” published by SuperDataScience Team. Most of the figures used in these sections are adapted from this blog.

For detailed information on the blog, please refer to the “Additional Resources” section of this chapter.

Research on the human brain gives us the idea that the human brain has three main parts: cerebrum, cerebellum, and brainstem, as shown in Figure 20.1. Let us briefly discuss the main functions of these parts and their components to understand the association between neural networks and the human brain.

Cerebrum

From Figure 20.1, we can understand that the cerebrum has four lobes. These are as follows.

Welcome to the World of Machine Learning!

In this ever-evolving era of technology, the field of machine learning (ML) stands at the forefront of innovation, promising unprecedented insights and solutions to complex problems. This textbook provides a comprehensive understanding of the theoretical foundations of ML algorithms, coupled with practical implementation using Python, designed to be a companion for both beginners venturing into the field and seasoned practitioners seeking to deepen their understanding.

Why Machine Learning?

ML is not just a technical endeavor; it is a transformational force shaping the future of how we interact with data. From enhancing search engines and revolutionizing social media to powering self-driving cars and advancing artificial intelligence (AI), the applications of ML are boundless. ML has paved the way for developing sophisticated chatbots, generative AI models, and AI copilots. This book aims to demystify the intricate concepts of ML, making them accessible to learners of all backgrounds.

What This Book Offers

• - Foundations: We begin with the fundamentals, laying a solid groundwork to help you grasp the core concepts of ML.

• - “Learning by Doing” Approach: This book adopts a “learning by doing” approach, offering step-by-step coding instructions for various ML techniques. The aim is to empower you with the knowledge and skills to implement these principles effectively.

• - Real-World Applications: Understanding theory is essential, but applying it to real-world scenarios is where the true power of ML unfolds. This book bridges theory and application, ensuring a holistic learning experience.

For Whom This Book Is Intended

Whether you are a student exploring the realms of ML for the first time, a data professional aiming to expand your skill set, or a business leader seeking insights into how ML can elevate your organization, this book is crafted with you in mind.

How to Use This Book

Feel free to navigate the chapters based on your familiarity with the subject. If you’re a beginner, start from the beginning, and if you’re seeking advanced knowledge, dive into specific sections of interest. Explore GitHub resources that provide access to datasets, sample code, and examples used in this book for hands-on learning. For enhanced visual understanding, this book is complemented by an online video course available at learncompscience.com, allowing you to reinforce your knowledge through engaging video sessions.

In this chapter, the first law of thermodynamics is developed using a series of experimental setups. In doing so, some new terminology and concepts are introduced. The idea of specific heat is revisited, and it is discovered that there are two forms of specific heat: a specific heat at constant volume and a specific heat at constant pressure. The important concepts of pressure work and thermodynamic cycles are also introduced.

This chapter is dedicated to the process of semantic analysis, the goals and strategies involved, and the pitfalls to avoid. This process involves three problems: identifying semantic and pragmatic effects on the observable (or at rate, observed) extension of an expression – the usage extension (Section 6.1); identifying and classifying the semantic and pragmatic sources of these effects (Sections 6.2–6.5); and accounting for inter-speaker and intra-speaker variation (Section 6.6).

Any empirical research presupposes at least a minimal set of theoretical assumptions concerning the nature of the object of study that must be made at the outset, but can be revised subsequently on the basis of empirical findings. This chapter introduces the metalanguages and ontologies most widely used in semantic research. Semantic metalanguages are codes designed to represent the meanings of linguistic expressions and utterances in semantic analyses. Ontologies are assumptions about the types of elements that “populate” a given domain; cf. Section 2.2. The organization distinguishes between metalanguages used to represent the referential properties of utterances (Sections 3.1–3.3) and metalanguages used to represent the cognitive properties of utterances (Section 3.4).

All languages seem to have provisions for the representation of time in their lexicons and in the practices that speakers customarily rely on when structuring discourses. But evidence has been mounting in recent years that the ways in which the representation of time is inscribed into the grammars of different languages varies substantially. The aim of this chapter is to review this evidence and introduce some empirical and analytical tools that facilitate the study of tense-mood-aspect systems in the field. The chapter begins by laying out some basics of temporal semantics in Section 10.2. Section 10.3 explores crosslinguistic variation with respect to this framework. Then Section 10.4 zeroes in on the topic of tenselessness, and specifically on future-time reference (FTR) in tenseless languages in Section 10.5, drawing on data from a variety of languages. Section 10.6 is a methodological guide to the study of temporality in the field and Section 10.7 concludes.

19.1 Building Image Classifier with CNN

The convolutional neural network (CNN) is the perfect choice for building an image classifier system. In this chapter, we will implement various CNN classifiers in Python by considering various case studies.

19.2 Dog–Cat Classifier

Consider an image classification problem that involves identifying photos as either containing a cat or dog. Our task is to develop a CNN model that can identify the object in the image as a dog or cat. The CNN model will be trained on a dataset of images containing a dog or cat, and then this trained model will be used for classification. Once the model is built, it can be used as a template to build other models for predicting the class of any image on the trained dataset. It means we can use the same model to classify the tumors by training them on medical images.

Preparing the Dataset

To build a dog–cat classifier, we will use 11000 manually annotated photos of dogs and cats, where 5500 photos are of dogs, and the remaining ones are of cats. This is the partial dataset that has been created from the dogs-vs-cats dataset available at Kaggle. The link for this dataset is given below.

https://www.kaggle.com/competitions/dogs-vs-cats/data

This full dataset contains 37500 images (24000 in the train and 13500 in the test folders), which must be manually segregated into dog and cat folders. Since the dataset is huge and segregation is time-consuming, we took 10000 images (5000 images of dogs and 5000 images of cats) in the train folder and 1000 images in the test folder (500 images of dogs and 500 images of cats).

12.1 Introduction to Clustering

In machine learning (ML), labeling the data is one of the crucial tasks. But sometimes, we do not have the labeled data. Even though the data is not labeled, we can still analyze it using clustering techniques. As you know, algorithms in ML are broadly classified as supervised and unsupervised techniques. In the case of supervised learning, the input data points/examples are labeled, while in unsupervised learning, the input data points/examples are not labeled. Cluster analysis, also called clustering, comes under unsupervised learning. Here, the input data points are not labeled. Clustering is the most popular technique in unsupervised learning.

Clustering is defined as grouping the input data points into various clusters/groups based on their similarity.

A cluster contains objects that are more similar to each other. In other words, during cluster analysis, the data is grouped into classes or clusters, so that records within a cluster (intra-cluster) have high similarity with one another but have high dissimilarities in comparison to objects in other clusters (inter-cluster).

The clustering algorithm aims to minimize the intra-cluster distance and maximize the inter-cluster distance, as shown in Figure 12.1.

An example of clustering is shown in Figure 12.2. Here, records in the input have different shapes. Here, we only have the input data without any label of shape. After applying the clustering algorithm, they were classified into three types of clusters. Here, the clustering algorithm considers the dimensions of the object and its color as the input features. Records whose features are highly similar are gathered to form a single cluster. In this case, we get three clusters representing three types of records, i.e., it clusters rhombus, circle, and triangle separately, as shown in Figure 12.2.

The central questions of an empirical epistemology and methodology for semantics are the following: (i) Which observable properties of communicative behavior can serve as valid data in semantic research? (ii) How and under what constraints can evidence of these observable properties be gathered in a fashion that permits valid analyses? (iii) Which analytical techniques are appropriate and optimal for bringing a particular dataset to bear on a particular research question? The present chapter addresses (ii). I discuss the tools – the methods – the researcher has at their disposal for collecting data from one or more of the sources of evidence discussed in Chapter 4.

In this chapter, the equations that were defined in Chapters 1–3 are used to solve some problems. In addition, the existing equations are modified to make them more useful for finding various property values. Also, two new thermodynamic properties are very briefly introduced, Gibbs energy (or Gibbs free energy) and Helmholtz energy (or Helmholtz free energy), in order to provide a full set of equations often associated with the laws of thermodynamics.

18.1 Image Recognition

Using a convolutional neural network (CNN), technological development in image recognition has revolutionized far beyond our imagination. Let us consider the comic scene shown in Figure 18.1, as it provides interesting insights into the development of image recognition and depicts a decade-back possible scenario. Here, a manager asks his computer programmer to “Develop an app which can check whether the user is in a national park or not, when he clicks some photo!” Being an easy and feasible task, the computer programmer responds that the task is merely of few hours. But, the manager's curiosity goes up, and he asks the programmer further to check whether the image is of a bird or not? Surprisingly, the programmer responds, “I need a research team and five years for this task.”

This surprised the manager as he expected it to be an easy task. But the programmer who has a knack in the field knows that it is one of the complex problems to be addressed in computer science.

In the last decade, we have provided solutions to many complex problems in the field of computers. But, for the last 50 years, we have been struggling to solve the problems in image recognition. However, thanks to the efforts of researchers and computer scientists across the globe, we can solve these problems now. Even a three-year-old child can identify a bird's photo, but identifying a way by which computers can do the same task was not a cake-walk; hence, it took almost 50 years!

We have finally found a promising approach for object recognition using deep CNN in recent years. In this chapter, we will discuss the working principle and concepts of CNN, a deep neural network approach to solving the problem of image recognition.

7.1 Introduction to Multiple Linear Regression

In simple linear regression, we have one dependent variable and only one independent variable. As we discussed in the previous chapter, the stipend of a research scholar is dependent on his years of research experience.

But most of the time, the dependent variable is influenced by more than one independent variable. When the dependent variable depends on more than one independent variable, it is known as multiple linear regression.

Figure 7.1 indicates the difference between simple and multiple regressions mathematically.

In Figure 7.1(b), b0 is the constant while x1, x2, and xn are independent variables on which the dependent variable y depends. You can point out that mathematically multiple linear regression is derived similarly to simple linear regression. The major difference is that in multiple linear regression, we have multiple independent variables, as it is from x1 to xn instead of only one independent variable, x1, in the case of simple linear regression. It is also important to note that we have b1 to bn as the coefficients of these independent variables, respectively.

The price prediction of a house can be viewed as a multiple linear regression problem, where factors such as plot size, number of bedrooms, and location play a significant role in determining the price. Unlike simple linear regression, which relies solely on plot size, multiple linear regression considers various features to accurately estimate the house price.

Let us understand this concept further with a real-life example. Consider a case where a venture capitalist has hired a data scientist to analyze different companies’ data to predict their profit. Identifying the company having maximum profit will help the venture capitalist to select the company he could invest in the near future to earn maximum profit.

21.1 Introduction

In recent years, a recurrent neural network (RNN) has become one of the most prominent neural networks for predicting values based on time series data. The time series is a collection of data points ordered over even intervals in time. Hence, time series analysis is useful when we want to study some parameter changes over time. In RNNs, the output from the previous steps is fed into the input of the current state. Thus, RNN could predict the next letter of any word, the next word of the sentence, and the diesel prices, stock prices, or weather. In all these tasks, there is a need to remember the previous values, and consequently, the output at t+1 will be dependent on the output at the time t interval. In this chapter, we will implement an RNN to predict the trend of diesel prices.

We will implement an RNN through long short-term memory (LSTM) units. We will use stacked layers of LSTM, as RNNs suffer from the issue of vanishing gradient and exploding gradient. Whereas LSTM encompasses gates through which they can regulate the flow of information. Because of this added advantage, LSTMs are commonly used for implementing RNNs.

21.2 Implementation of RNN in Python

Problem Statement: In this experiment, our primary objective is to predict the trend of diesel prices in Delhi, i.e., we wish to predict that the diesel prices in Delhi will witness an upward or downward trend for the near future. It is difficult to predict the exact future price of diesel on a particular day, as our dataset for this problem is quite small. Hence, using an RNN model (implemented through stacked layers of LSTM), we will predict an approximate value of future diesel prices in Delhi, which helps to predict the trend accurately (upward or downward).