Artificial Intelligence
Tutorial Study Image

Artificial Intelligence Interview Questions



  • Aims to develop a system that mimics a human being and simulates natural intelligence to solve tough issues.
  • More concentrate in decision making or AI is a decision-maker.
  • Extend probability of success
  • Aims to develop self-learning algorithms. It makes machines able to learn from the data on the task and thereby maximize the performance of the machine.
  • ML enables the machine to learn new things from the data or is concentrate on learning.
  • Extend accuracy.

Artificial Intelligence is the technique used to make a machine mimic a human whereas Machine Learning can be described as a subset of AI and is a technique used to implement AI. ML makes the system able to learn new tricks on its own from the given data without being explicitly programmed to do so.

DeepFace( deep learning facial recognition system) algorithm is used by Facebook for face verification. It is based on the face verification algorithm with the support of AI techniques using neural network models. 

Input: A wild form of photos with large complex data which involves blurry images, high-intensity images, images with high contrast are scanned to make the input.
Process: 4 step process is employed to complete the  modern face recognition
1.    Detect and analyze facial features
2.    The features are compared and aligned.
3.    3D graphs are used to represent the key patterns. 
4.    Images are classified based on similarities. 
Output: A face representation is derived from a 9-layer deep neural network.
Training Data: More than 4 million facial images uploaded by more than 4000 users.
Result: Facebook can detect whether the same person is represented by the two images. 

Market basket analysis is a data mining technique used by large retailers which explains the combinations of items that frequently occur in transactions and identify relationships between the items that people buy.

For example, if a person buys Toothpaste, there is a 40% chance that he might also buy Toothbrushes. Companies can provide relevant offers and discounts on items by understanding the correlations between the items and thereby they can grow their business.

Machine Learning algorithms like Association Rule Mining and Apriori algorithm are the basic logic behind Market basket analysis:
•    Association rule mining is a technique used to determine the correlation between items.
•    Association rules are generated by the Apriori algorithm from frequent itemsets. It follows the concept that a subset of a frequent itemset must also be a frequent itemset.

For example, if a person buys item X then he will also buy item Y. The retailer then can announce a discount offer which states that on purchasing Item X and Y, there will be a 40% off on item Z. Such rules are generated using Machine Learning and are then applied on items to increase sales and grow a business.

Artificial Intelligence is the stimulation of human Intelligence on man-made machines. The machines are well programmed with a set of algorithms. The machine can work with its Intelligence and act like a human.

Google search engine is one of the most popular examples of AI applications. We get recommendations to choose from immediately as we open up our chrome browser and start typing something. Artificial Intelligence is the logic behind the search engine.

Predictive analytics, NLP, and Machine Learning are used by AI to recommend relevant searches to us. Google uses the data collected about us to make these recommendations, such as our search history, location, age, etc. Thus, What a user might be looking for is predicted by Google with the help of AI

An agent is an entity that analyzes and recognize its environment with the help of sensors and acts upon their environment by effectors. The agent can be Robots, Programs, and Humans, etc.

Neural networks/Artificial Neural Network/Simulated Neural Network is a structure that resembles and mimics the functionalities of the human brain. It receives the data, processes the data, and gives the output based on the algorithm and empirical data.

Frames are data structures used to divide the substructures by representing  “stereotyped situations”.It is an extensive part of knowledge representation and reasoning scheme. In Scripts, the stereotyped sequence is described in a structured representation with a particular order and is similar to frames.  Scripts are used to organize knowledge base in natural language understanding systems. The organized knowledge base is in a way that the system should understand 

•    Python
•    Java
•    Julia
•    Haskell
•    Lisp

Strong AI Weak AI
  • Human-level intelligence
  • Can be applied widely
  • Extensive scope association and clustering is used
  • Examples: Robotics
  • Characteristics that resemble human intelligence.
  • Application is limited to some specific tasks.
  • The scope can be minimal.
  • unsupervised and supervised learning is used
  • Examples: Apple’s Siri, Amazon’s Alexa

As with its name it is all about learning and pattern recognition. ML algorithms observe the patterns and try to learn from them. The ML programs will keep learning and try to improve with each attempt. A great example is our Facebook newsfeed.

•    Chatbots
•    Facial recognition
•    Image tagging
•    Self-driving cars
•    Natural Language Processing

•    Reinforcement learning
•    Semi-supervised learning
•    Supervised learning
•    Transduction
•    Unsupervised learning

Based on capabilities:
•    Weak AI/Narrow AI/Artificial Narrow Intelligence
     Capable of performing some specific tasks with intelligence.
•    General AI/Strong AI/Artificial General Intelligence
      Capable of performing any intellectual tasks with human-like efficiency.
•    Super Intelligence/Artificial Superhuman Intelligence
Capable of performing anything or everything that a human can do and more.
Based on functionalities:
•    Reactive Machines
     The basic types of AI. It is only based on the present actions.cannot use the experience to form current decisions.
•    Limited Memory
     Past experiences are stored for a limited duration.
•    Theory of mind
      An advanced type of AI. It can understand human emotions and people in the real world.
•    Self-Awareness
      The future of Artificial Intelligence. It will possess human-like consciousness, emotions, and reactions.

•    Deep Learning
•    Machine Learning
•    Expert System
•    Fuzzy Logic System
•    Robotics
•    Natural Language Processing
•    Neural Networks.

•    Scikit Learn
•    PyTorch
•    TensorFlow
•    MxNet
•    Auto ML

Reinforcement Learning is a type of machine learning algorithm based on a feedback loop between agent and their environment.
By performing certain actions and analyzing the rewards and results of those actions, the agent learns how to behave in an environment. This is a behavior-driven technique based on the reinforcements learned via trial-and-error.

An AI program with expert-level knowledge about a specific area of data and utilizes the information to react appropriately. These systems can be used to substitute a human expert and solve real-life problems.

A method of inquiry in AI for check whether or not a computer can think like a human being. If the computer can mimic human responses under specific conditions then it can pass the test and can consider as intelligent.

Fuzzy logic is a subset of AI in which human learning is encoded for artificial processing. It comes with mathematical concepts of set theory. Representation of IF-THEN rules and the digital values of YES and NO is used. It is based on the degrees of truth.

•    Stock trading
•    Weather forecasting systems
•    Facial pattern recognition
•    Air conditioners, washing machines, and vacuum cleaners
•    Medical diagnosis and treatment plans

1.    Recurrent neural networks
2.    Feedforward neural networks
3.    Convolution neural networks

Game theory is a branch of mathematics. It is more relevant in multi-agent situations. Here we deal with deciding on a given set of options. As the environment is multi-agent, the choice will affect the choices of the opponent or other agents. Here every agent is equally rational and tries to get maximum rewards. Game theory is not restricted to games, it is also relevant to applications like ML Algorithms, Generative Adversarial Networks.                                                                                                                                                                                                      

•    Random forest
•    K-means clustering
•    Logistics regression
•    Naïve Bayes
•    Decision making

•    Bio-informatics
•    Market segmentation
•    Image, face, and speech recognition

Python is the most leading programming language used in AI.
It is an open-source modular programming language. Its clearness and dependable coding method lead it to direct the AI industry. It is popular with its open-source libraries like Matplotlib and NumPy, frameworks such as Scikit-learn, logical version libraries like TensorFlow and VTK.

Neural networks in AI can be defined as the mathematical model of the human brain functions. This technique facilitates the machine to think and understand like humans. This is how technology recognizes speech, objects, and more today. 

It is an open-source software library used in machine learning and neural network research. It is initially developed by the Google Brain Team for data-flow programming. Tensorflow makes it much easier to make  Certain AI  features into applications like natural language processing and speech recognition.

It is a subset of Machine Learning, which in turn makes the system to imitates the way human gains certain types of knowledge. Deep learning algorithms are arranged in a hierarchy of increasing complexity. Multi-layered neural networks are used to process data. Layers of neural networks arranged on top of each for use in deep learning are called deep neural networks.

Ai is designed to imitate humans and emulate the human brain. Humans can be said as visual. Therefore, training machines to analyze and categorize images is a crucial part of AI. Image recognition also makes the machine learn from repeated processing to recognize the images better and processing those images.  

Q-Learning is an algorithm used in reinforcement learning. Here Q means quality. In this case, Quality represents the degree of usefulness of a given action to gain some future reward. Bellman equation is the principle behind this algorithm. Here the agent learns some optimal policies from experience that can provide the best actions to perform to maximize the rewards under certain conditions. The agent tries to maximize the value of Q.

a.    User Interface:  used to interact or communicate with the expert system to find the solution for a problem.
b.    Inference Engine:  The main processing unit or brain of the expert system. It is used to extract information from the knowledge base. The inference engine applies different inference rules to the knowledge base to get a conclusion from it. 
c.    Knowledge Base: A storage area that stores domain-specific and high-quality knowledge.

It is an AI software or agent which uses NLP to simulate a conversation with the user. The interaction is performed through any applications, websites, or massaging apps. This can interact with the human in the form of text or through voice and are also called digital assistants. AI chatbots are broadly used in business applications.

•    Semantic Network Representation
•    Production Rules
•    Logical Representation
•    Frame Representation

a.    Rainbird
b.    TensorFlow
c.    Microsoft Azure AI platform
d.    Infosys Nia
e.    IBM Watson

An ideal rational agent is an agent that can perform in the best possible action and maximize the performance measure. The actions from the alternatives are selected based on:
•    Percept sequence
•    Built-in knowledge base
The actions of the rational agent make the agent most successful in the percept sequence given. Rational agents are the highest performing agents.  

A graphical model is used to show the probabilistic relationship between a set of variables. It can be described as a cycle graph with multiple edges, where each edge represents a  conditional dependency.

Since these networks are built from a probability distribution and use probability theory for prediction and anomaly detection it can be said as probabilistic. And it is important in AI because it can be used to answer probabilistic questions and based on the Bayes theorem.

Machine Learning Deep Learning.
The amount of data is small Training time is less No need for high–performance hardware Less accuracy Output is numerical values, like classification or score. A large amount of data is required Need a lot of time to train. Requires high-performance hardware. High accuracy Output is anything from numerical values to free-form elements, like text and sounds.

•    Supervised Learning- Learns by using labeled data.
•    Unsupervised Learning- Learns from unlabeled data without any guidance.
•    Reinforcement Learning- Interaction of agent with the environment by actions and analyzing errors and rewards.

•    Input layer- Inputs are received and forwards to the hidden layer for analysis.
•    Hidden layer- Computations are carried out here and the result is forward to the output layer. Depending on the problem to be solved the number of hidden layers may vary.
•    Output layer- information is transferred from the neural network to the outside world through this layer.

•    It is based on the basic unit of the brain called a neuron. An artificial neuron  (perceptron) was developed.
•     Like dendrites, which are used to receive inputs, the perceptron receives multiple inputs and applies various functions, and provides an output.
•    Just like a human brain, artificial neurons (perceptrons) are connected to form a network called a deep neural network.

•    The perceptrons have a set of inputs, each of them is assigned with some specific weight. Perceptron then computes some functions on these weighted inputs and gives the outputs.

Feedforward Neural Network is the simplest form of ANN where the data travels in uni-direction. The data passes through the input nodes and exits through the output nodes. Hidden layers are conditional

•    Feedforward Neural Network
•    Convolutional Neural Network
•    Recurrent Neural Network(RNN) 
•    Autoencoders

The RL agents work on the principle of reward maximization. For that, the RL agent must be trained in a way that takes the best among the alternatives to get a maximum reward. After each action, the agent changes its state and that of the environment. The agent gets its maximum reward based on how much those actions by the agent affect the goal to be achieved.

The parameters used to determine the entire training process are called Hyperparameters. Their values cannot be estimated from data and are external to the model. For example, Learning Rate, Hidden units.

•    Grid Search
•    Random Search
•    Bayesian Optimization

Text Mining is a technique used to extract useful insights from structured and unstructured text. It can be done using text processing like Perl, statistical models, etc. The output is the frequency of words, patterns, correlations.

Supervised classification is better for image classification in terms of accuracy.
The images are manually fed and interpreted by the ML expert to create future classes in supervised classification. Whereas, in unsupervised classification, the feature classes are created by the ML software based on image pixel values.

Parametric model Non-parametric model
  • Faster computations
  • Less data requirement
  • Strong assumptions on data
  • A fixed number of parameters are used to build the model
  • Slower computations
  • Data requirement is more
  • A fewer assumption about data
  • A flexible number of parameters are used to build the model.

•    Acquisition efficiency
•    Inferential adequacy
•    Inferential efficiency
•    Representation adequacy

•    Alternate keys
•    Artificial keys
•    Compound keys
•    Foreign keys
•    Natural keys 
•    Primary keys
•    Super keys

Random forest is a data construct used by ML projects.  The random forest method is used to develop a large number of random decision trees for variables. These algorithms will improve the way that the technologies are used to analyze complex data sets. A basic concept employed here is that multiple weak learners can be combined to form a strong learner. It can work with large labeled and unlabeled data sets with large attributes. And hence it is a perfect tool for AI projects. With some missing data, it can hold its accuracy. It can also be used for dimensionality reduction.

When a case arises on the stationarity, or when the data source has been changed we have to think about an update on the algorithm

Statistical AI is more related to inductive thought (induce the trend) 
Classical AI is more related to deductive thought(deduce a conclusion)

State-space search is considered the most straightforward approach for planning algorithms. It takes care of everything to find a solution. 

Unification is a method of substitution. It makes two different logical atomic expressions identical by applying a substitution. Two literals are taken as input and turn them into identical by using substitution.

•    The predicate symbols have to be the same
•    Both the expressions must be with an identical number of arguments.
•    No two similar variables within the same expression.

By using the process of unification we can make different logical expression looks identical. A substitute is applied to make a different expression look identical.

•    Validity
•    Logical Equivalence
•    Ability satisfaction

For extracting the meaning from a group of sentences, AI uses semantic analysis.

The meaning of P*Q is determined from  P, Q, and *. This process of deriving meaning is called compositional semantics.

Currently, almost all speech recognition systems use this model. It is an ever-present tool for model sequence behavioral data. It is used to solve temporal probabilistic reasoning.

A single Discrete Random Variable is used to describe the state of the process in HMM.

The additional variables can be added to a temporal model.

•    Arithmetic literals
•    Predicates
•    Equality and inequality

We can use the Inverse Resolution algorithm to inverts a complete resolution. It is the best algorithm for learning first-order theories.

Acoustic signals are used to identify a sequence of words in speech recognition.

Inductive logic programming is used to combines inductive methods with the power of first-order representations.

The independence of a node under a certain condition from its predecessors is the consequence between a node and its predecessors in Bayesian Network. 

A computer algorithm is extensively used to find the path or traverse a graph to find out the most optimal path between various points called nodes. It will search for the shortest path between the initial and final state or it can be used to calculate the shortest distance between the initial and final state. 

It is a search algorithm used to reduce the number of nodes searched by a minimax algorithm in the search tree. It can be defined as an optimization technique for the minimax algorithm. We can compute the correct minimax decision without checking each node of the game tree and this technique is called pruning. Alpha-Beta pruning involves two threshold parameters for the future expansion, Alpha and Beta.

•    Medical diagnosis and treatment plans
•    Weather forecasting systems
•    Facial pattern recognition
•    Stock trading
•    Control of subway systems

Breadth-First Search Depth First Search
Not suitable for decision making trees for games and puzzles Suitable for searching vertices that are closer to the source. The queue data structure is used Traverse through a minimum number of edges to reach the destination from the source. More suitable for games or puzzle problems Suitable for finding solutions away from the source. The stack data structure is used Traverse through more edges to reach the destination from the source.

A uniform cost search algorithm can be said to be identical to that of BFS if each iteration is with the same cost. It performs sorting in increasing the cost of the path to a particular node.

It is a powerful machine learning algorithm used for predictive modeling. It consists of a set of algorithms with a common principle based on the Bayes Theorem. There will be an independent and equal contribution from each feature to the outcome is the fundamental assumption of Naïve Bayes.

It is an unsupervised learning algorithm. The unlabeled dataset is grouped into different clusters where each dataset belongs to only one group with similar properties. Here K is used to define the number of pre-defined clusters that have to be created in the process.

Intelligence Knowledge
It is built-in Unique to each one Cannot be gained through practice Helps to decide how to deal with situations or solve the problem Gained through the learning process Versatile and can be same for different people Acquired through persistent practice Helps to know about the situation and what is going to happen.

Markov’s Decision process can be described as a mathematical approach to formalize the Reinforcement Learning problem. It aims to gain maximum positive rewards by choosing the optimum policy. In this process, action A is performed by the agent to take a transition from the initial state to the final state. The agent will get some rewards while doing this action A. The series of actions done by the agent is called the policy.

•    Removing unnecessary Features
•    Regularization
•    Cross-Validation
•    Early Stopping the training.
•    Ensembling
•    Training With more data

Overfitting is one of the main issues sounds in machine learning. When the ML algorithms try to capture all the data points, they also capture noises along with data accidentally and there occur overfitting in the models. Because of this overfitting issue, the algorithm shows a low bias and a high variation in the output.

Dropout Technique
It is the regularization technique used to avoid overfitting issues in neural network models. It selects neurons randomly and dropped during training.   complex co-adaptations on training data are avoided.

1.    Natural Language Understanding
2.    Natural Language Generation

Computer vision is a field of AI. It is used to train the computer to obtain information's from the visual world. Computer vision uses AI to solve problems like image processing, object detections 

•    Logical Representation
•    Semantic Network Representation
•    Frame Representation
•    Production Rules

•    Q-Learning
•    Deep Q Neural Network
•    SARSA

•    Autonomous Transportation
• Education system powered by AI.
•    Healthcare
•    Predictive Policing
•    Space Exploration
•    Entertainment, etc

•    F1 score
•    Confusion Matrix
•    AUC-ROC curve
•    Gain and lift charts
•    Gini coefficient
•    Cross-Validation
•    Root mean squared error

Integrating multiple elements to create a unique identifier for the construct in the absence of a single data element that defines the construct uniquely.

The heuristic approach is the best way to deal with a game-playing problem. It uses the intelligent guesswork technique.

The easiness of the method adapted to different domains of application is represented with the term “Generality”

Three terms are required; one conditional probability and two unconditional probability.

It is a data structure for implementing an associative array that can map key values. It can also compute an index into an array where the desired value can be found. It consists of two parts. an array where data is stored and a mapping function known as a hash function.

First, a problem statement based on the business problem has to be created. This will help to ensure that we are fully aware of the type of the problem and the input and output of the problem. The statement must be simple and within a single sentence. Then we have to choose an appropriate algorithm concerning the problem we are trying to solve from the following groups:
•    Classification algorithms
•    Clustering algorithms
•    Regression algorithm
•    Recommendation algorithm

•    Data collection
•    Data preparation
•    Choosing an appropriate model
•    Training the dataset
•    Evaluation
•    Parameter tuning
•    Predictions

E-commerce websites use ML to recommend products to customers. A process of comparing users with similar shopping behaviors and recommending products to a new user with similar shopping behavior is called collaborative filtering and is the basic idea of recommendation engines.

Popular Programs