What is machine learning?

Derived from Artificial Intelligence, machine learning is a method of data analysis that automates analytical activities, providing computers the ability to learn, identify patterns, and make decisions autonomously.

With the growing amounts of data being generated on a daily basis, the smarter, more substantial computational processing skills of modern systems, and with more affordable data storage options available, not only is it now more practical than ever before to utilize machine learning, it is more powerful.   

But, why do we need it? The lure of machine learning is in the promise of automating processes by using a computer without the need to code the logic for every single task. Instead, we need machines, on their own, to learn and figure out the logic by analyzing data and events.

To put this into perspective, Thomas H. Davenport, author and senior advisor at Deloitte Analytics, in an excerpt from the Harvard Business Review,  said it best: “Cognitive insights provided by machine learning differ from those available from traditional analytics in three ways: They are usually much more data-intensive and detailed, the models typically are trained on some part of the data set, and the models get better—that is, their ability to use new data to make predictions or put things into categories improves over time.”

The strength of machine learning comes from using a variety of algorithms to build models that help uncover connections, patterns, trends, and insights that help organizations make smarter, better decisions that require little-to-no human intervention. According to Study.com, an algorithm in computer science is “a well-defined procedure that allows a computer to solve a problem. A particular problem can typically be solved by more than one algorithm. Optimization is the process of finding the most efficient algorithm for a given task.”

Machine learning and the use of an optimized selection of algorithms is being rapidly embraced by a number of industries given that it provides valuable insights, oftentimes in real-time, which provide organizations with the opportunity to act on them in a timely manner and gain an advantage over the competition. Some of the industries that have welcomed machine learning into their array of processes include the financial services industry, the government, the healthcare industry, the retail industry, and transportation, to name a few.

Machine learning is still in its early stages of adoption, which means that the role associated with it, the machine learning engineer, is relatively new as well. To get the most out of machine learning, machine learning engineers must be are professionals who:  

• Are experts at implementing machine learning methods and techniques such as deep learning, robotics, natural language processing, and more.
• Have strong backgrounds in software engineering skills, algorithm skills, data structure skills, and probability and statistics skills.
• Are proficient in programming languages such as Python, C, C++, R.

Possess a demonstrated ability to stay up to date with the latest technologies, collaborate effectively with other teams, and think holistically to solve problems to their full extent.

Mistakenly, the role is typically confused with that of the data scientist. Unlike the machine learning engineer, the data scientist uses scientific methods to determine the best machine learning approach to use to solve a problem and conveys it to the machine learning engineer who is tasked with executing it. Machine learning engineers take the models designed by the data scientists and feed the necessary data into them to build programs that control computers.

Technology stack and list of machine learning algorithms

What technologies and algorithms do machine learning engineers need to master? Let’s take a look.

Machine Learning Technology Stack

The following are some of the most popular tools and technologies used to implement and leverage machine learning.

• Numpy: Python library that adds support for large, multidimensional arrays and matrices alongside a comprehensive collection of high-level mathematical functions to manipulate and operate said arrays.
• Scikit-learn: Free Python library solution used in machine learning applications that features classification, regression, and clustering algorithms, along with vector machines support.
• Matplotlib: Plotting library for Python that provides an object-oriented API to embed plots into applications via general-purpose GUI toolkits such as Tkinter, Qt, or GTK+.
• Pytorch & Torch: Pytorch is a machine learning library for the Python programming language that is based on the Torch library.
• CometML: Open-source and free tool that tracks code, experiments, and results of most machine learning libraries to graph outcomes.
• TensorFlow: TensorFlow is a free and open-source software library that helps machine learning algorithms construct and train frameworks and neural systems to mimic human perception, thought processes, and learning.
• Pandas: Considered one of the top solutions for Python libraries, Pandas has established itself as one of the key tools for machine learning. It supports reading and writing Excel spreadsheets, CVSs, and a great deal of manipulation.
• Apache Spark: Open-source distributed cluster-computing framework that can access data in a variety of sources to provide a unified analytics engine for Big Data. Its in-memory processing works with machine learning projects to deliver real-time analytics.
• Apache Hadoop: Software library and framework that facilitate the use of a network of multiple computers to solve problems that involve massive amounts of data and computation. The Hadoop framework is made up of the following models: Hadoop Distributed File System (HDFS), Hadoop Common, Hadoop YARN, and Hadoop MapReduce.

Along with this list, there are a number of general-purpose technologies and tools that are good to have in the arsenal of a machine learning technology stack. These include:

• AWS Deep Learning AMI: Infrastructure and tools that accelerate machine learning and deep learning in the cloud. Users can easily and quickly launch Amazon EC2 instances along with popular frameworks such as TensorFlow, PyTorch, Keras, etc.
• Google Cloud ML Engine: It’s a managed service that allows developers and machine learning engineers to build and run superior machine learning models in production. It allows the scaling up of machine learning algorithms to train models on large datasets in a short amount of time.
• GitHub: Web-based hosting service for version control using Git and development platform designed to host and manage projects, review code, and develop software.
• Keras: High-level neural network written in Python that is designed to enable fast experimentation through its modular and extensible framework.
• Docker: Software-as-a-service and Platform-as-a-service product that uses virtualization to develop and deliver software in packages called containers. For machine learning, Docker is of great assistance when dealing with the installation process of software.

Machine Learning Algorithms

The following algorithms pertain to the supervised, unsupervised, and reinforcement machine learning types. For more information, take a look at the Deep learning vs machine learning article.

Supervised machine learning

Supervised learning builds a model that makes predictions based on evidence. To develop these predictive models, it uses regression and classification techniques.

• Linear Regression: Linear regression expresses a linear relationship between the input and the output.
• Polynomial Regression: It expresses a polynomial regression between input and the output. It gives a more precise result compared to linear regression.
• Decision Trees: A decision tree is a series of nodes, a directional graph. This is for the predictions that result from a series of feature-based splits.
• Random Forests: Random forests use many decision trees. They are an ensemble of decision trees. Each decision tree is created by using a subset of the attributes. Random forest is an effective supervised classification algorithm.
• Classification: This set of algorithms includes a K-Nearest Neighbors Algorithm (KNN), Logistic Regression, Native-Bayes, Support Vector Machine (SVM), etc.

Unsupervised machine learning

Unsupervised learning discovers patterns in data by drawing conclusions from unlabeled input data. 

• Clustering: Grouping objects in clusters. Algorithm clustering is k-means, which identifies the best k-cluster centers in an iterative form.
• Dimensionality reduction: Singular-Value Decomposition (SVD) is a matrix decomposition method for reducing a matrix to its constituent parts. There are two approaches to dimensionality reduction: Feature selection and feature extraction. The main algorithm used is the Principle Component Analysis (PCA).
• Association analysis: It helps to extract the most frequent and largest item sets within big data sets. It is used for discovering relations between variables in large databases.

Reinforcement Learning

The algorithms comprised in this type of machine learning use software agents to take actions in an environment to maximize values of cumulative reward. It typically includes the Markov decision process as well as the following algorithms: 

• Criterion of optimality.
• State-value function.
• Brute force.
• Monte Carlo methods.
• Temporal difference methods.
• Direct policy search.

Machine learning engineers are strongly advised to be proficient in the aforementioned technologies, tools, and algorithms to succeed in their role. Next, let’s review the basic machine learning interview questions that engineers are typically asked and the best way to approach them.

Basic machine learning interview questions

During an interview process, machine learning engineers are usually tested for a variety of skills which include: technical and programming skills, the ability to structure solutions for open-ended problems, knowledge of  how to apply machine learning effectively, expertise in data analysis with several methods, communication skills, cultural fit, and an overall mastery of  fundamental concepts of machine learning.

Here’s a list of the basic machine learning interview questions.

What is your interpretation of machine learning?

This question is intended to let applicants showcase their viewpoint and approach to machine learning. In essence, machine learning is a method that automates analytical model building. It is a branch of Artificial Intelligence and as such is based on the idea that machines can learn from data, identify patterns, and make decisions with minimal human supervision/intervention.

What is the relevance of machine learning in today’s technology landscape?

Machine learning engineers can highlight their knowledge about cutting-edge technologies in machine learning as well as the latest trends, all of this in the context of highlighting how machine learning can benefit numerous industries and be applied to a multitude of applications to simplify routine tasks by having machines learn. Candidates should talk about the power that machine learning has in uncovering connections and patterns in data that can create opportunities for businesses and take their technology strategy further with the use of a carefully assembled list of algorithms that improve the decision-making process.

What is the difference between artificial intelligence, machine learning, and deep learning?

Candidates should clearly pinpoint that deep learning is a subfield of machine learning which in turn is a subfield of artificial intelligence, along with the definition of each field.

• Artificial intelligence is the science that aims to create intelligent machines that work, act, and behave like humans.
• Machine learning is a subfield of artificial intelligence that specializes in algorithms and statistical models to learn from data and have machines perform tasks without explicit instructions.
• Deep learning is a subfield of machine learning that focuses on tasks that mimic the human brain in terms of processing data and creating patterns. It is comprised of neural networks that learn from unstructured and unlabeled data.

What is supervised, unsupervised and reinforcement learning? What are the most common algorithms for each type?

Candidates should explain the following:

• Supervised learning is the practice of using algorithms that are trained by using labeled examples to map an input to an output. Some of the most common algorithms include linear regression, logistic regression, decision trees, random forests, Naïve-Bayes, etc.
• Unsupervised learning uses unlabeled data to find structure within data sets. Some of the most common algorithms include k-means, hierarchical clustering, PCA, anomaly detection, etc.
• Reinforcement learning is the practice of having machines discover through trial and error which actions yield the best results and maximize the notion of cumulative reward. Some of the most common algorithms include q-learning, state-action-reward-state-action, deep q network, deep deterministic policy gradient, etc.

How do you select the best algorithm for a unique dataset scenario?

Here, candidates will do well to clarify that the decision of selecting an algorithm is mainly based on the type of data involved. If the data presents characteristics of linearity, then linear regression would be a good option. If the scenario calls for the use of images, audio, and other complex items, a neural network would help build a comprehensive model.

List some of the most widely-used technologies and tools used in machine learning.

Candidates can use this question to highlight their knowledge and expertise in key technologies that are applicable to machine learning. As seen in this article’s section about the technology stack used in machine learning, engineers can talk about TensorFlow, Pandas, Keras, PyTorch, Scikit learn, Hadoop, Numpy, and more. An added bonus would be to mention cutting-edge technologies that are just emerging into the machine learning landscape.

What is regularization?

Candidates should refer to this concept as the practice of using techniques that aim to improve the validation score, sometimes at the cost of reducing the training score.

What is data augmentation?

Here, applicants can describe that it is the practice of increasing the number of data points derived from internal and external sources within an enterprise to add value. It synthesizes new data by modifying existing data in a way that the target doesn’t change or it changes in a familiar way.

What is overfitting?

Candidates should explain that overfitting occurs when a statistical model describes random errors or noise instead of the underlying relationship. If models are complex, there are too many parameters in relation to the number of training data types, which results in overfitting.

What is a Bayesian network?

Candidates must explain that Bayesian networks are a probabilistic model that represents a set of variables and their conditional dependencies via a directed acyclic graph.

What is support vector machines and what are the two classification models that it supports?

They are supervised learning models that analyze data used for classification and regression analysis. This model can support combining binary classifiers and modifying binary to include multiclass learning.

What is data visualization and which are the libraries you use?

Candidates must talk about the fact that data visualization is the graphical representation of information and data. Through the use of visual elements such as graphs, charts, and maps, users can understand trends and patterns in data. Additionally, the candidate must list some of the most popular tools which include R’s ggplot, Python’s seaborn and matplotlib, Plot.ly, Bokeh, and Tableau, to name a few.

When is a genetic algorithm used?

Genetic algorithms solve constrained and unconstrained optimization problems based on natural selection, which is part of the larger class of evolutionary algorithms. Genetic algorithms are typically used when there is little knowledge about the search space and they are applied to any optimization problem.

These are some basic interview questions for machine learning engineers that serve as a basis of what to focus on, but there are more in-depth questions that may take place. Let’s explore them in the subsequent section.

Machine learning engineer interview questions

Here is a list of machine learning engineer interview questions to deepen the practice of getting to know more about the skills of the machine learning engineer.

What is the bias-variance tradeoff?

With this question, candidates have the opportunity to prove their knowledge about predictive model knowledge. They should talk about how predictive models have a tradeoff between how well the model fits the data (bias) and how the model changes based on changes in the inputs (variance). Adding to this, they should explain that simple models are stable (low variance) but don’t come near a truthful outcome (high bias), while complex models get too close to a particular data set (high variance) and are expressive enough to get closer to the truth (low bias).

What is stratified cross-validation?

Candidates should explain that it’s a technique that divides data between training and validation sets. In a typical cross-validation, the split is done randomly but in a stratified cross-validation, the split keeps the ratio of the categories on both datasets. Applicants will do well do add that this technique is used on datasets with multiple categories and on datasets with different distributions.

What is the predicament of dimensionality? List some best practices to deal with it.

With this question, the aim is to identify that the greatest challenge of dimensionality is that with more dimensions available, it is harder to search through a solution space. In essence, the dataset does not have enough samples for a model to learn from since there are too many features. To address dimensionality, candidates should mention key best practices such as:

• Feature selection: Selecting a subset of features to work on.
• Reduction of dimensionality: Using techniques to reduce dimensionality include principal component analysis and autoencoders.
• Regularization: Create sparse parameters to deal with dimensionality.
• Feature engineering: Create new features to group together existing features.

When talking about dimensionality reduction, what is feature extraction?

Candidates should talk about how feature extraction is a process by which an initial set of data is reduced by identifying key features of the data.

What is an imbalanced dataset?

It’s one that has different proportions of target categories. The different ways to deal with imbalanced datasets include oversampling or undersampling, data augmentation, and the use of appropriate metrics.

What are parametric models?

In this question, applicants should focus on describing how these types of models have a finite number of parameters, which should be clearly defined in order to predict new data. As a bonus, applicants can add examples which include linear regression, logistic regression, etc. Also, it’s noteworthy to mention that nonparametric models not bound to parameters allow more flexibility and mention examples such as linear decision trees, k-nearest neighbors, etc.

What’s your method to select important variables when working on a dataset?

This question is designed to examine the way the engineer goes about their role. The wanted answer should include these steps:

• Remove correlated variables.
• Use linear regression and select variables.
• Use forward selection, backward selection, and stepwise selection.
• Use a random forest and a plot variable importance chart.
• Use the lasso regression.
• Measure information on the available features and select the top features accordingly.

What is convex hull?

In the context of linearly separable data, convex hull represents the outer boundaries of the two groups of data points.

What is the Principle Component Analysis?

Candidates should explain that it’s a method that transforms features in a dataset by combining them into correlated linear combinations. These new features, namely the principal components, maximize the variance in sequential order. It is typically used for dimensionality reduction.

What are the delivery models for machine learning platforms that are cloud-based?

Applicants can showcase their knowledge of cloud solutions and delivery models by mentioning cognitive services, ML Platform as a Service, ML Infrastructure Services, machine learning model management.

What are the key benefits of machine learning in the cloud?

• The cloud has a pay-per-use model that is a good fit for machine learning workloads.
• The cloud helps organizations experiment with different machine learning capabilities and scale up as the projects go in production and demand rises.
• The cloud delivers intelligent capabilities that are easily accessible without the need for advanced skills in artificial intelligence, machine learning, or data science.
• Platform as services cloud provide SDKs and APIs that are readily available to embed machine learning functionalities directly into applications, as they support most programming languages.

These are just a sample of the type of questions that engineers may encounter during the interview process. Depending on the level of seniority in the role, the questions may vary in complexity.

Additional skills that are helpful for a machine learning interview

On top of a strong and sophisticated background in machine learning knowledge, engineers should also possess crucial soft skills that will help him/her perform at their best in a machine learning environment.

Communication is a fundamental skill that will make or break the value that machine learning engineers extract from datasets. The ability to communicate insights that are derived from machine learning tasks and algorithms is vital for organizations to take timely actions. Findings should be clearly translated and communicated in a way that highlights the recommended business decisions that should be executed.

Critical thinking and a problem-solving attitude will go a long way in the career of a machine learning engineer as they face real-life problems on a daily basis and have to select the best methods and practices to approach each complex scenario.

Importantly, machine learning engineers must be highly proficient in mathematics and statistics as these two sciences are the foundation upon which machine learning is built. Machine learning sits at the intersection between statistical, probabilistic, computer science, algorithms, and mathematical aspects that are inherent to extract value from data, which is why it’s critical that machine learning engineers possess extensive knowledge of these fields to thrive.

Along with math and statistics, machine learning engineers will do well to have essential programming skills specific in Python and R. These two programming languages are long known to be successful in environments of statistical computing and with the use of machine learning algorithms as they provide an easy syntax that works well with prediction, pattern recognition, and more.

Prioritization is key in helping machine learning engineers separate the highest priorities from the not so critical priorities. Engineers must decide which problems to tackle first and how much effort should be allocated to them.

A business-oriented attitude helps machine learning engineers translate their findings in an easy-to-understand fashion that enables decision-makers to understand the actions that should be taken to harness the potential and value from machine learning.

How Svitla Systems can help with machine learning projects

Svitla Systems is an experienced and effective partner for machine learning projects. With over 15 years of experience, Svitla has speared its way into the latest technologies, staying up to date with the latest trends about emerging and proven methods that help uncover value for customers.

Machine learning is one of those methods that has gained a significant amount of traction over the years. With numerous applications and an increasing number of companies embracing machine learning, it’s no wonder that we count in our ranks experienced professionals who are competent in machine learning practices and algorithms.