Artificial Intelligence & Machine Learning's Most Important Concepts for Interviews

Let’s learn the most important concepts of Artificial Intelligence and Machine Learning through alphabets!

ANN
Artificial Neural Network (ANN) mimics the human brain’s neural network. Hidden units of the hidden layer in the artificial neural network architecture can be thought of as Neurons and they introduce non-linearity with inputs and few biases.

Bagging

Bagging means making multiple training sets from a training set and training separate models on those separate training sets and finally averaging them out.

Correlation
Correlation measures the relationship between two features or variables.

Deep Learning
The cornerstone of deep learning is artificial neural networks. Whenever you have artificial neural network with many hidden layers, its most likely a deep learning problem which is being solved.

Error
In a supervised machine learning setup, you always have an actual number on which you are supervising the model and a predicted and an estimated number and the difference between these two numbers is error. You tend to minimise the error in these kinds of machine learning problems which are a kind of supervised machine learning problems.

Feature
Feature or variable or input, all it means is what you feed into your data science models.

 

Gradient Descent
Gradient descent is a way to approximate global minima using local minima following an Iterative approach.

Hypothesis Testing
Using test statistic, you either reject the null hypothesis in the favour of alternative or fail to reject the null.

Intercept
Intercept in linear regression is that parameter which has the same impact for all inputs. For example, if your intercept is 3, the impact of intercept on y will always be 3, irrespective of x.

Julia
Julia is the next breakthrough in data science programming, as it is specially designed for quantitative researchers or data scientists.

KNN
K-Nearest Neighbours or KNN is the method which can be adopted to both regression and classification. The key idea here is that here you find K nearest points of any data point, and using this create a new model. And most importantly, this is a non-parametric model, here K is the hyper-parameter.

Linear Regression
Linear Regression is the simplest, yet the most powerful predictive ML Model. Its components are the intercept and coefficients of all the features or predictors. 

Model
Model is an abstraction or simplification of reality using mathematics and statistics.

Normal Distribution
People generally talk about normal distribution when they are new to data science. It is not the only statistical distribution of importance. There are others as well. To completely identify a normal distribution, you need its mean and variance. And the heuristics to understand whether a distribution is normal or not, is that it will be symmetric about its mean with bell shape. 

Overfitting
Overfitting is prone to happen when the model is too complex, fits the training data well but fails miserably when presented with new data.

P- Value
P-value measures the strength of rejection or failure to rejection. In statistical terms, it is the observed level of significance.

Q-Q Plot
Q-Q Plot compares two probability distributions. It is a Test for Normality.

Random Forest
Random forest combines the number of decision trees similar to bagging with a twist on the choice of features while making the decision trees.

SVD
SVD along with PCA is a dimension reduction technique and if you feed into it p features, when p is very large, it gives you a smaller set of modified features. In PCA terms, most important principal components which helps you in reducing the dimensionality of the problem.

T Test
T-test in statistical terms is a statistical test to compare means of two different distributions.

Underfitting
Underfitting is the opposite of overfitting. It is a deficient model which has missed out on important patterns or features while training a model.

Variance
Variance measures how spread the data is about the mean. So, if a data set is 2, 2, 2, 2…., it will definitely have a very low variance, which in this case is zero. And, if let’s say, data is 2, 5, 10, 15, 30, definitely its variance is going to be very high and not zero.

Web Scraping
Web Scraping is a risky way to get data from websites. Hence, read all the possible violations while doing web scraping. 

X (Inputs)
X is what you now.

Y (Outputs)
Y is what you need to know.

Z-Score
Z-Score measures how far data point is from the mean. In terms of standard deviation, just subtract the mean from the data point and divide it by the standard deviation.