Explore topic-wise InterviewSolutions in Current Affairs.

Game theory is a branch of AI that attempts to define a strategic game with predefined rules and OUTCOMES between two players of equal rationality. Every player is selfish and tries to maximize the reward to be obtained using a particular strategy. All the players abide by certain rules in order to receive a playoff- which is a reward. Therefore, a game can be defined as a set of players, actions, strategies, and a final reward.

Game theory and AI are related to each other and complement each other. Game theory is used in AI situations where multiple agents are in an environment trying to achieve a goal. Various games are logical and have a set of pre-decided rules like chess, poker, etc., which can be MADE AVAILABLE digitally with the help of Artificial Intelligence and Game Theory.

A rational agent is a person or entity, based on realistic MODELS, that has preferences for advantageous outcomes and will try to achieve them in all scenarios. A rational agent is one who has defined preferences, models UNCERTAINTY and acts in such a way that its performance measure is maximized using all available actions. The proper things are stated to be done by a reasonable agent. AI is concerned with the development of rational agents for application in game theory and decision theory in a variety of real-world contexts. The rational ACTION is the most crucial for an AI agent because, in an AI reinforcement learning algorithm, an agent receives a positive reward for each best FEASIBLE action and a negative reward for each incorrect action. 

Rationality is the ability to remain reasonable and just in all POSSIBLE scenarios. The performance metric of an agent is used to determine its rationality. The following criteria can be used to assess rationality:

• The success criterion is defined by a performance metric.
• The agent has prior knowledge of its surroundings.
• The most effective activities that an agent can take.
• The order in which precepts appear.

Artificial Neural networks, simply called Neural networks, are computer systems based on units called nodes or artificial neurons, which resemble the neurons in human brains. Each node can transmit a signal from one node to another. 

An Artificial Neural Network

A neural network includes SEVERAL layers, each of which performs a specialized job. As the model's complexity grows, the NUMBER of layers grows as well, which is why it's called a multi-layer perceptron.

A neural network in its purest form includes three layers: an INPUT layer, a hidden layer, and an output layer. The input layer receives the input SIGNALS and passes them on to the next layer, with the output layer delivering the final prediction. These neural networks, like machine learning methods, must be taught with some training data before being used to solve a specific problem. Let's learn more about the perceptron now.

An ANN CONSISTS of multiple layers including an input layer, an output layer, and hidden layers. Some commonly used ANN are:

• Feed Forward Neural Network
• Convolution Neural Network
• Recurrent Neural Networks
• Auto-encoders.

Regression and classification are both supervised learning algorithms. Both works on LABELED data and are USED to predict in machine learning. The difference, however, arises from the manner in which they are used.

An expert system is a computer PROGRAM that simulates the JUDGEMENT and behavior of a human or an organization with expert knowledge and expertise in a particular field using artificial intelligence (AI) technologies.

The expert systems belong to an important domain of Artificial Intelligence, which is USED to solve complex problems using extraordinary human intelligence and expertise. 

The different components used to build an expert system are:

• Knowledge base- It is a storage area that CONTAINS domain-specific, high-quality knowledge.
• Inference ENGINE- The Inference engine uses and manipulates the knowledge from the knowledge base.
• User Interface- It provides interaction between the expert system and the user.

• Eigenvalues are the COEFFICIENTS given to EIGENVECTORS that determine the length or magnitude of the vectors. Eigenvalues are unit vectors having magnitude 1. A NEGATIVE eigenvalue, for example, MAY scale the eigenvector in the opposite way.
• Eigenvectors are unit vectors, meaning their length or magnitude is the same as 1.0. They're also known as right vectors, which simply means "column vectors" (as opposed to a row VECTOR or a left vector). A right-vector is a vector in the traditional sense.

Fuzzy logic (FL) is a method of REASONING in Artificial INTELLIGENCE that resembles human reasoning. According to this logic, the OUTCOME can take any values between TRUE and FALSE (digitally, 0, or 1). For example, the outcome can be certainly yes, possibly yes, not SURE, possibly no, or certainly no. 

Conventional logic states that a computer can take input and produce definite output which is True or False, which is equivalent to human YES or NO.

Fuzzy logic: the outcome can take any value between 0 and 1

Natural Language Processing (NLP) is a field of Artificial Intelligence, concerned with GIVING computers the ability to understand and interact in human languages in a way humans can.

NLP combines rule-based modeling of human language with statistical, machine learning, and deep learning models. This makes a computer fully understand and comprehend human language in the form of voice or text. Voice-operated GPS systems, speech-to-text systems, customer service chat boxes, etc. USE NLP.

Natural language processing encompasses a wide range of methods for analyzing human language, including statistical and machine learning techniques, as well as rules-based and algorithmic approaches. Because text- and voice-based data, as well as practical applications, require a diverse set of methodologies. Tokenization and parsing, lemmatization/stemming, part-of-speech tagging, language detection, and semantic link identification are all basic NLP activities. You've probably DONE these chores manually PREVIOUSLY if you ever diagrammed sentences in elementary school. NLP tasks, in general, BREAK down language into smaller, more basic bits and attempt to comprehend the relationships between them. These tasks are used in higher NLP capabilities like- content categorization, topic discovery and modelling, sentiment analysis, speech-to-text, and text-to-speech conversion.

If we can detect overfitting at an early stage, it will be very useful for our training model. There are several methods up our sleeves that can be used to avoid overfitting-

• Cross-validation: Cross-validation is a resampling technique for evaluating machine learning models on a small sample of data.
• Remove features: We can remove the unnecessary features of the models to encompass the outliers.
• Early stopping: Early stopping is a type of REGULARIZATION used in machine learning to MINIMIZE overfitting when using an iterative METHOD like gradient descent to train a learner. Early stopping CRITERIA specify how many iterations can be completed before the learner becomes over-fit.
• Training with more data: We can train our model with more data to accommodate outliers.
• Regularization: In machine learning, regularization is a method to solve the over-fitting problem by adding a penalty term with the cost function.
• Ensembling: Ensemble learning refers to combining the predictions from two or more models.

Overfitting is a CONCEPT in data science when a data POINT does not FIT against its TRAINING model. When the raining model is fed with data, there is a possibility that it might encounter some noise that cannot fit into the statistical model. This happens when the algorithm cannot perform accurately against unseen data.

Overfitting data points- In the above linear model, some points do not fit in. These points are outliers.

Hyperparameters are the parameters that control the entire training PROCESS. These variables are adjustable and have a direct impact on how SUCCESSFULLY a model trains. They are DECLARED beforehand. Model hyperparameters, which cannot be inferred while fitting the machine to the training set because they refer to the model selection task, and algorithm hyperparameters, which have no effect on the model's performance but affect the speed and quality of the learning process, are TWO types of hyperparameters.

The selection of good hyperparameters is crucial for the training process.  ACTIVATION function, alpha learning rate, hidden layers, number of epochs, number of branches in a decision tree, etc. are some of the examples of hyperparameters.

The HIDDEN Markov model is a probabilistic model which is used to identify the probabilistic character of any event. It SAYS that an observed event is related to a set of probability distributions. If a system is being modeled into a Markov’s chain, then the main goal of HMM is to identify the hidden LAYERS of the Markov’s chain. Hidden means that the particular state is not observable to the observer. It is generally used for temporal DATA. HMM finds its application in reinforcement learning, temporal PATTERN recognition, etc.

A Hidden Markov Model. Z₁…..Zₜ₊₁ are hidden states

REWARD maximization is a technique used in Reinforcement LEARNING. Reinforcement learning is a subset of AI algorithms made up of three main components: an ENVIRONMENT, agents, and incentives. The agent alters its own and the environment's state by completing actions. The agent is awarded or penalized based on how much their activities affect the goal the agent must attain. Many reinforcement learning challenges begin with the agent having no prior knowledge of the environment and conducting random behaviors. The agent learns to optimize its actions and adopt policies that maximize its reward based on the feedback it receives.

The goal is to maximize the reward and the action of the agent by using optimal policies. This is called reward maximization. Any abilities that are repeatedly requested by the agent's environment must eventually be created in the agent's behavior if it can alter it to improve its cumulative reward. A good reinforcement learning agent could eventually learn perception, language, social INTELLIGENCE, and other skills while maximizing its reward.

Markov’s decision process (MDP) is a mathematical approach for reinforcement learning. Markov's decision process (MDP) is a mathematical framework used to solve problems where outcomes are partially random and partly controlled. To solve a complex problem using Markov’s decision process, the following basic things are needed-

• Agent- The agent is an entity that we are going to train. For example, a robot that is going to be trained to assist in cooking is an agent.
• Environment- The surroundings around the agent are called Environment. The kitchen is an environment in the case of the above-mentioned robot.
• State (S)- The current situation of the agent is called the state. So, in the case of the robot, the position where the robot is, the temperature of the robot, its posture, etc. collectively define the state of the robot.
• Action (A)- The robot can move left or right, or it can pass an ONION to the chef, these are some of the actions that the agent (robot) can take.
• Policy (𝜋)- The policy is the reasoning behind taking an action.
• Reward (R) -A reward is RECEIVED by the agent for taking a desirable action.
• Value (V)- The value is the POTENTIAL future reward that the agent can receive.

The working of Markov’s model can be understood from the following diagram.

In simple words, the agent has to do some action to start from its initial state. While doing so, it RECEIVES rewards based on the actions it takes. The policy defines the action it takes, and the reward collected defines the value (V).

• Simple Reflex Agents: Simple reflex agents ignore the HISTORY of the ENVIRONMENT and its interaction with the environment and act entirely on the current situation.
• Model-Based Reflex Agents: These models perceive the world according to the predefined models. This model also keeps track of internal conditions, which can be adjusted according to the changes made in the environment.
• Goal-Based Agents: These kinds of agents react according to the GOALS GIVEN to them. Their ultimate aim is to REACH that goal. If the agent is provided with multiple-choice, it will select the choice that will make it closer to the goal.
• Utility-based Agents: Sometimes, reaching the desired goal is not enough. You have to make the safest, easiest, and cheapest trip to the goal. Utility-based agents chose actions based on utilities (preferences set) of the choices.
• Learning Agents: These kinds of agents can learn from their experiences.

Reinforcement learning is an area of machine learning which works on reward-based models for prediction and decision making. It deploys a feedback-based mechanism to reward a machine when it MAKES good decisions. Negative feedback is provided to the machine when it does not perform well. This encourages the machine to find the best possible behavior in a particular situation. Unlike supervised learning, the agent learns autonomously utilizing feedback and no labelled data in Reinforcement Learning. Because there is no labeled data, the agent MUST rely only on its own experience to learn.  RL is used to tackle a certain sort of problem in which sequential decision-making is required and the aim is long-term, such as game-playing, robotics, and so on. The agent interacts with and explores the world on its own. In reinforcement learning, an agent's primary goal is to increase PERFORMANCE by obtaining the most positive rewards. The agent learns through trial and error, and as a result of its experience, it improves its ability to complete the task.

Reinforcement learning can be best understood by taking an example of a dog. When the owner of the dog wants to cultivate a good habit in his dog, he will train his dog to do that thing with the help of a treat. The dog will be rewarded with the treat if he obeys his owner. If he disobeys the owner, the owner will use a negative reinforcement technique by not giving his dog his favorite treat. This way, the dog will associate the habit with the treat. This is precisely how reinforcement learning works in a machine.

Applying Reinforcement learning Principles to dogs. Repeatedly awarding the dog with treats (positive reinforcement) can MAKE the dog adapt to the good habits (walking in this case) quickly. 

A Bayesian network is a probabilistic graphical model based on a set of variables and their dependencies, represented in the form of an acyclic graph. Bayesian networks are based on PROBABILITY distribution, and they predict outcomes and detect ANOMALIES using probability theory. The Bayesian networks are used to perform tasks such as prediction, detecting anomalies, REASONING, gaining insights, diagnostics, and decision-making. A Bayesian network, for example, COULD be used to illustrate the probability correlations between DISEASES and symptoms. The network may be used to calculate the chances of certain diseases being present based on symptoms.

Computer Vision in the field of AI enables computer systems to deduce meaningful interpretations from images or any other visual STIMULATION and take ACTION based on that input. AI gives the system the ABILITY to think, then computer vision gives the system the ability to observe. Computer vision is very much similar to human vision.

Pattern recognition is the foundation of today's computer vision algorithms. We train computers on a vast amount of visual data—images are processed, things are labeled, and patterns are found in those items. For example, if we submit a MILLION photographs of flowers to the computer, it will analyze them, find patterns that are common to all flowers, and produce a model "flower" at the end of the process. As a result, every time we submit them photographs, the computer will be able to precisely recognize whether a certain image is a flower. Computer vision is can be witnessed in many areas of our lives. 

For example- Content organization in Apple Photos, Facial Recognition systems, self-driving CARS, augmented reality, etc. use computer vision.

The Turing test is used to determine whether or not a machine is capable of thinking like a human. IT was developed by Alan Turing in 1950.

The Turing Test is like an interrogation game between three players. There is an interrogator who is a human. He has to interrogate two other players- one computer and one human. The interrogator has to figure out which one is a computer by asking questions. The computer has to do its best to make itself HARD to be distinguished from the human. The machine will be considered intelligent if it makes it hard to be distinguished from the human.

Consider the following SCENARIO: Player A is a computer, Player B is a human, and Player C is a questioner. The interrogator is aware that one of them is a robot, but he needs to figure out which one. Since all players communicate via keyboard and screen, the outcome is UNAFFECTED by the machine's capacity to transform words into SPEECH. The exam result is determined not by the number of correct answers, but by how closely the responses resemble those of a human. The computer is permitted to do whatever it can to force the interrogator to make a false identification.

The question-answer session can be like this-

Interrogator: Are you a computer?

Player A (computer)- No.

Interrogator: Multiply two huge numbers, such as (256896489*456725896).

Player A- After a long delay, he gives the incorrect answer.

If an interrogator is unable to distinguish between a machine and a human in this game, the computer passes the test, and the machine is said to be intelligent and capable of thinking like a human. This game is POPULARLY called ‘imitation game’.

Q Learning is a model-free learning policy that chooses the best COURSE of action in an environment, depending on where in the environment the AGENT is (an agent is an entity that makes a decision and enables AI to be put into action). Model-free learning policy means that the nature and predictions of the environment to learn and move forward. It does not reward a system to learn, it uses the trial and error method instead.

The model's goal is to determine the optimum course of action given the CURRENT situation. To accomplish this, it may devise its own set of rules or act outside of the policy that has been established for it to obey. This means there isn't a real need for a policy, which is why it's called off-policy. The agent's experience is SAVED in the Q table in Q-learning, and the value in the table INDICATES the long-term reward value of executing a certain action in a specific condition. The Q learning algorithm, according to the Q table, can instruct the Q agent the action to take in a given situation to maximize the predicted reward.