Artificial Intelligence Interview Questions

Basically, machine learning is pattern recognition. Like Youtube’s video recommendations, Facebook’s News Feeds, etc. are a perfect example of pattern recognition. Machines observe patterns and learn from the examples. The type of videos you see on YouTube, you get video recommendations of similar type. The outcome of the machine learning program keeps improving with every attempt and trials.

Breath-First search involves traversing a binary search tree one level at a time. Starting with the root node, proceeding through neighboring nodes moving towards the next level of nodes. As this process can be performed utilizing FIFO (First in First Out) data structure. This strategy gives the shortest path to the solution. BFS assigns two values to each node: distance and predecessor.

• A distance is calculated by giving the minimum number of edges in any path from the source node to node “v”.
• The predecessor node of “v” along with some shortest path from the source node. The source node's predecessor is some special value, such as null, indicating that it has no predecessor.

If there is no path from the source node to node “v”, then v's distance is infinite, and it is assumed that predecessor has the same special value as the source's predecessor.

Few differences between inductive machine learning and deductive machine learning are

Classification is used when the output variable is a category such as “red” or “blue”, “spam” or “not spam”. It is used to draw a conclusion from observed values. Differently from, regression which is used when the output variable is a real or continuous value like “age”, “salary”, etc. When we must identify the class, the data belongs to we use classification over regression. Like when you must identify whether a name is male or female instead of finding out how they are correlated with the person.

The Depth-First Search (DFS) algorithm is a recursive algorithm that uses the method of backtracking. It involves extensive traversing through all the nodes by going ahead, if possible, else by backtracking.

When you are moving forward and there are no more nodes left along the current path, you go back on the same path to find new nodes to traverse. The next path will only be selected if all the nodes on the current path will be visited. DFS is performed using the stack and it goes as follows:

Pick a source/starting node and push all the adjacent nodes into a stack. Now pick a node from the stack to select the next node to visit and push all its adjacent nodes into a stack.

Repeat the process until the stack is empty. To ensure that all the visited nodes are marked, as this will prevent you from re-visiting the same node. If you do not mark the visited nodes and you may visit the same node more than once and end up in an infinite loop.

A game is the most visible area of progress in the AI system. AI systems could be improved using game theory, which requires more than one participant to narrows the field quite a bit. It serves two fundamental roles:

 
• Participant Design: Game theory is used to improve the decision of the participant to get maximum utility.
 
• Mechanism Design: Inverse game theory, works at designing a game for a group of intelligent participants. Ex. Auctions.

Model accuracy is a subset of model performance. Model performance operates on the datasets feed as input to the algorithm and Model accuracy is based on the algorithm of model performance.

A hash table is a data structure that is used to create an associative array of arbitrary size which is mainly used for database indexing.

TensorFlow is an open source machine learning framework. It is fast, flexible and a low-level toolkit meant for doing a complex algorithm and offers customizability to build experimental learning architectures. AlphaGo and Google Cloud Vision are built on Tensorflow framework.

• Supervised Learning
• Unsupervised Learning
• Semi-supervised Learning
• Reinforcement Learning
• Learning to Learn
• Transduction

Regularization is a technique used to solve the overfitting or underfitting issues in any statistical model. It is basically used to reduce fitting error in the dataset. A new piece of information is fit in the data set to minimize or reduce fitting issues.

Gradient checking is done to detect hidden bugs in complicated software that manifest themselves under very specific conditions. It is a debugging process done on the back prop algorithm, it also performs a derivative checking procedure, and ensure that the implementation is correct.

• Principal Component Analysis
• Independent Component Analysis
• Kernel-Based Principal Component Analysis
• Linear Discriminant Analysis
• Generalized Discriminant Analysis