Explain the terms: Artificial Intelligence, Machine Learning and Deep Learning

Artificial Intelligence is the simulation of how how human brain works and solves the complex problems. These algorithms are used to solve problems which can not be solved using classical programming approach or problem takes unreasonable amount of time to be solved. It makes machine to think like human being and solves the problem. Algorithms under AI are inspired by the nature. Speech recognition, natural language processing, expert systems etc. are the applications where AI has special role to play.

Machine Learning is subset or the branch of AI, which creates mathematical model from the historical data. Machine learning algorithms are centered around probability, linear algebra, optimization techniques and statistical theories. They are widely used in industries for prediction and automation. Classification, regression, clustering, anomaly detection etc. problems can be solved with the help of machine learning.

Deep learning or popularly known as deep neural network are special class of machine learning algorithms. Philosophy of deep learning algorithms is derived from the way human brain works. Deep learning algorithms are quite useful in solving computer vision problems. Deep learning algorithms got popularity after the ImageNet challenge organized in 2012. AlexNet proposed to solve the ImageNet – 2012 was a big break through in dep learning.

Relation between AI, ML and DL algorithms is depicted in following figure.

Differentiate Machine Learning and Deep Learning

Machine Learning technology gives machines the ability to make business decisions without any external help, using the knowledge gained from past data. Machine Learning systems require relatively small amounts of data to train themselves, and most of the features need to be manually coded and understood in advance. Here, the given business problem is dissected into two, and they are solved individually. Once the solutions of both parts have been acquired, they are then combined.

Deep Learning allows machines to make various business-related decisions using artificial neural networks, which is one of the reasons why it needs a vast amount of data for training. Since there is a lot of computing power required, it requires high-end systems as well. The systems acquire various properties and features with the help of the given data, and the problem is solved using an end-to-end method.

Deep learning algorithms are in fact a subset of machine learning algorithms. Machine learning algorithm requires hand crafted features as an input where as deep learning model does not require manual feature extraction.

Machine learning employs a set of algorithms to analyze and interpret data, learn from it, and make the best judgments possible based on those learnings. Deep learning, on the other hand, divides algorithms into numerous layers to build a “artificial neural network.” This neural network is capable of self-learning and making intelligent decisions.

Differentiate supervised and unsupervised learning

Supervised learning algorithms are those that are learned with labelled data. The models use direct input to confirm whether the expected output is right. Furthermore, both the input and output data are presented to the model, with the primary goal being to train the model to predict the output when fresh data is received.

For training, unsupervised learning algorithms employ unlabeled data. The models in this situation do not take any feedback and, unlike supervised learning, they uncover hidden data trends. The input data is the sole thing given to the unsupervised learning model, and its major goal is to find hidden patterns to extract knowledge from unknown sets of data.

Classification and regression are supervised machine learning tasks. Clustering and anomaly detection are unsupervised machine learning tasks. Due to availability of label at the training time, supervised algorithms are more accurate then their counter part

What is cross validation? State different validation methods.

In Machine Learning, the cross-validation approach allows a system to improve the performance of a given Machine Learning algorithm to which multiple sample data from the dataset is fed. This sampling procedure is used to divide the dataset into smaller sections with the same amount of rows, from which a random part is chosen as a test set and the remainder is kept as train sets.

Following are the some of the most popular cross validation techniques:

• Hold out method
• K-fold cross validation method
• Leave One Out Cross Validation (LOOCV) method

What do you mean by entropy in Machine Learning?

In Machine Learning, entropy is a metric that assesses the randomness of the data to be processed. The more the entropy in the data, the more difficult it is to derive any relevant conclusions from it. Take, for example, the event of flipping a coin. The outcome is unpredictable because it does not favour heads or tails. Because there is no specific relationship between the action of flipping and the different outcomes, the outcome for any number of tosses cannot be anticipated simply.

What is Bias and Variance? What is their impact?

The bias is the difference between our model’s average prediction and the correct value. If the bias value is high, the model’s prediction is inaccurate. As a result, in order to make the desired predictions, the bias value should be as low as possible.

Variance is the number that represents the difference between a training set’s prediction and the expected value of other training sets. High variance may result in large fluctuations in output. As a result, the model’s output should have a low variance.

Additional Reading: Entropy and Information Gain. Click to read