AI is permeating almost everything from consumer products to industrial equipment. As companies leverage AI to improve their competitiveness, more companies are leveraging machine learning to accomplish more in less time, reduce costs, and tap into market potential, whether it's for pharmaceuticals. I've discovered something though.
Non-data scientists don't need to understand how machine learning (ML) works, but they should know enough to use basic terminology correctly.
The scope of ML is far beyond what can be covered in this short article, but here are some of the basics:
Basic terms
Before understanding the concept of machine learning, we need to understand the meaning of machine learning terms. Commonly used terms include:
-
A/B testing – Test two machine learning techniques to determine which performs better.
-
clustering – Group objects based on similarity. For example, within an M&M population, individuals within that population may be grouped by color or type. (Example: Comparison of Peanut M&Ms and regular His M&Ms).
-
decision tree – A hierarchy of binary values used to make decisions (e.g., is the customer athletic or not? Does the athletic customer ski?).
-
false negative – Results that appear negative are actually positive. (Example: Cybersecurity breaches that evade detection.)
-
False positives – Results that appear positive are actually negative. (Example: A facial recognition system that mistakenly identifies a lawmaker as a murder suspect.)
-
Features – Input variables used for prediction. (Example: Female (1) Under 25 years old (2) Smokes cigarettes (3).
-
Features engineering – Decide which features to use in your model.
-
feature set – A group of features used to train the model.
-
holdout data – Data subtracted from the training data that is later used to test the model.
-
inference – Make predictions using a model trained on unlabeled data.
-
K-means – A clustering method that uses Euclidean geometry (more specifically Euclidean distance) to determine the similarity of examples.
-
label – Human-defined outcomes. (Example: cat, dog, tall, short)
-
model – The result of running the algorithm on the training data.
-
neural network – A collection of artificial neurons (aka nodes) that typically uses multiple inputs to produce an output.
-
proxy – Data that can be used to infer sensitive attributes. (Example: Using zip code to determine race, likelihood, or recidivism.)
-
random forest – Create multiple decision trees with random features (using different parts of the dataset) and determine the average prediction of a single decision tree. Random forests are more accurate than a single decision tree, but they are not as interpretable.
-
reinforcement learning – A type of machine learning that uses rewards and penalties.
-
semi-supervised learning – Use labeled data to infer labels for unlabeled data.
-
supervised learning – Use labeled data to learn by example. (Example: day, night.)
-
training set – Subset of data used for training.
-
unsupervised learning – Often used to infer data labels and discover things that humans have not yet discovered. (Example: Discovering the main causes of readmissions.)
-
verification – The process used to determine the quality of the model.
Machine learning and deep learning
Deep learning is a subset of machine learning that utilizes multiple layers of algorithms. The algorithm forms neural network nodes arranged in his three basic layers: input layer, hidden layer and output layer. If a network has multiple hidden layers, it is considered a deep neural network.
Brooke Wenig, Databricks
“Deep learning is simply a series of matrix multiplications and nonlinear transformations,” said Brooke Wenig, machine learning practice leader at the cloud data platform provider. data brick. “Do a number of matrix multiplications on the input features, give each one a corresponding weight, and then add a nonlinear transformation.”
There are many different types of neural network architectures available today, and the list continues to grow.
One thing to keep in mind with deep learning is that it requires a lot of data and storage, so it's expensive. It also requires a lot of computing power. This can be expensive not only from a resource point of view, but also from an environmental (carbon footprint) point of view. There are other considerations as well.
Wayne Thompson, SAS
“People need to minimize models not based on some error criterion, but based on some economic impact of the model,” said Wayne Thompson, chief data scientist at the analytics software provider. SAS. “The problem is, we don't know what numbers to put in on the economic side. When we talk to some customers, we find that the price associated with acquiring a customer and retaining a customer once acquired is I can't tell you the revenue.”
Which type of ML technique a data scientist uses depends on several factors, including the business problem that needs to be solved, the data available, the level of accuracy required, time, efficiency, and more. Sometimes the most elegant solution is also the simplest. , not the most sophisticated or complex.
Some popular neural networks
There are many different types of neural network architectures, all of which have an input layer, an output layer, and one or more hidden layers. Generative adversarial networks (GANs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs) are just three examples.
GANs are used in cybersecurity and gaming because both cases involve an adversary. GAN involves two networks, one of which is adversarial.
Cameron Fenn, AI Capital Management
“We've tried just about everything with GANs. GANs work very well, but they're a little problematic because they're difficult to train,” said Co-Founder and Head of Research at a venture capital firm. Cameron Fenn said. AI Capital Management. “People don't like training GANs, so they're replacing them with other generative models that work just as well or better.”
Convolutional neural networks (ConvNets or CNNs) are modeled after the visual cortex of animals, so it makes sense that they would be used for image recognition. The goal of CNNs is to reduce the image size for processing without sacrificing the features needed for good predictions.
CNN is used for a variety of use cases, including advertising, climate change, natural disaster prediction, and self-driving cars.
Recurrent neural networks (RNNs) use continuous or time-series data. They are called “iterative” because they perform the same task on every step of the sequence. In practice, RNNs are used for handwriting and speech recognition, time series prediction, time series anomaly detection, and even robot control.
Related content:
How to explain AI, ML, and NLP in plain language to business leaders
How to upskill your staff for AI and machine learning
AI and Machine Learning: An Enterprise Guide
