Glossary of key terms

With all the recent buzz around artificial intelligence, it’s hard to keep up with all the terms used to define AI, and even harder to understand how it works. Here are some key terms related to AI, Machine Learning, Natural Language Processing (NLP), Natural Language Understanding (NLU), and subfields such as Generative AI, Conversational AI, Ethical AI, and Explainable AI. I will introduce

Related article: How artificial intelligence can break through data silos

Definition of AI

activation function: Activation functions are important because they help neural networks learn complex things better. Transform simple output into something more complex.

Artificial intelligence (AI): Artificial Intelligence is a subdomain of computer science focused on developing systems or machines that exhibit human-like intelligence in tasks such as problem solving, learning, reasoning, perception, and natural language understanding.

artificial neural network: Inspired by the biological neural networks found in animal brains, artificial neural networks (ANNs) serve as the basis for numerous machine learning and deep learning models. They consist of interconnected nodes (neurons) arranged in layers to create complex computing systems.

attention mechanism: attention mechanisms improve neural networks by paying attention to important input factors. This improves performance for tasks such as machine translation, text summarization, and image captioning.

error backpropagation: Backpropagation is an important algorithm for training neural networks, especially deep learning models, by adjusting weights to minimize the error between actual and predicted outputs. It computes the loss function gradient (a mathematical concept used to optimize the performance of neural networks) for each weight using the chain rule, and iterates layers backwards for optimization.

Bag of Words (BoW): The document is represented as an unordered set of words with frequency information preserved, ignoring grammar and word order.

bias: For neural networks, the bias is an additional parameter that shifts the activation function along the input axis. In the broader AI context, bias refers to systematic errors in model predictions due to biases or assumptions in the training data.

Convolutional Neural Network (CNN): CNN is a deep learning model for processing grid-like data such as images using convolutional layers with filters that detect spatial hierarchy and recognize patterns at different scales.

cross entropy loss: A loss function commonly used in classification tasks to evaluate the difference between predicted probabilities and true probabilities.

Conversational AI: Conversational AI includes techniques that use NLP, NLU, and natural language generation (NLG) techniques to enable computers to have human-like conversations. Common applications that use conversational AI include chatbots, voice assistants, or other applications that enable natural language-based human-machine interaction.

clustering: An unsupervised machine learning technique that clusters data points by similarity without relying on existing labels. Common algorithms include K-means, hierarchical clustering, and density-based clustering algorithms (DBSCAN).

DALL-E: AI model developed by OpenAI. It combines the capabilities of generative and large-scale language models to create images from textual descriptions. DALL-E is based on a variant of the GPT-3 architecture, modified to produce images instead of text.

deep learning: Deep learning is a subfield of machine learning that emphasizes multilayer artificial neural networks that learn complex patterns from large datasets, advancing applications such as image recognition, speech recognition, and natural language processing.

dimensionality reduction: Dimensionality reduction techniques minimize features in datasets while preserving important information, increasing computational efficiency and addressing the curse of dimensionality.

Ethical AI: The practice of designing artificial intelligence systems that adhere to ethical principles and values, such as fairness, accountability, transparency, and the impact of AI on society.

Functional engineering: Feature engineering involves creating or modifying input variables to improve the performance of machine learning models. This includes tasks such as scaling, normalization, and encoding of categorical variables. Domain knowledge is often required to select relevant and beneficial features.

function selection: Identify and select important features from a dataset for building machine learning models, reducing overfitting, improving performance, and minimizing computational complexity. Feature selection techniques include filters, wrappers, and embedding methods.

Generative Adversarial Networks (GANs): A deep learning model containing two neural networks: a generator to create fake data and a discriminator to distinguish between real and fake data. Networks compete with each other in a game-like fashion, with generators trying to fool discriminators by creating more realistic data.

Generative AI: A class of machine learning models that can generate new data samples (i.e. dialogues, answers to questions, or images) similar to the training data. These models, such as generative adversarial networks, are used in various applications such as image synthesis, text generation, data augmentation, and chatbots.

gradient descent: an optimization algorithm that minimizes a function. It is often used in training machine learning models to reduce the loss function. Iteratively tune the model parameters according to the negative slope and converge to a local minimum.

Hyperparameter: User-defined model parameters not learned from data, such as learning rate, batch size, hidden layers of neural networks. Hyperparameter tuning seeks the optimal values ​​for a given model and dataset.

image and text pair: These consist of images and their associated text descriptions or labels, and are used to train AI models, such as image captioning systems and visual question-answering models, to understand relationships between visual and text data. make it possible.

input token: The units of meaning in text for training AI models or NLP tasks are words, phrases, or characters. They are processed by Seq2Seq models or encoders in other NLP architectures to capture sequence structure and meaning.

large scale language model: AI models that heavily use deep learning, aimed at understanding and producing human-like text. Trained on extensive text data sets to capture complex linguistic patterns. Examples include OpenAI’s GPT series (GPT-3, ChatGPT-4), Microsoft Bing, and Google Bard.

loss function: A feature that measures the difference between the predicted output of a machine learning model and the actual output or target. The training aims to minimize the loss function, which is achieved using optimization algorithms such as gradient descent or stochastic (randomly determined) gradient descent.

output token: A single unit of meaning in text produced by a sequence-to-sequence model or other NLP task. They can be words, phrases, or characters, and the Seq2Seq model’s decoder builds them one by one to form a complete output sequence.

overfitting: Overfitting occurs when a machine learning model learns too well the training data, which contains not only the dominant pattern, but also noise and random variations. This makes the model perform poorly on new data.

Reinforcement Learning (RL): This is when a computer program learns to make decisions by interacting with a virtual environment. Take feedback as rewards or penalties and try to get the best total rewards over time. It is used in areas such as video games, robotics, and recommendation systems.

Inter-sequence model (Seq2Seq): A deep learning construct used to transform the data in the input series into the output series. There is an encoder that processes the input and a decoder that creates the output.

soft weight: A soft weight indicates that within a neural network there is a probability assigned to an element when computing the attention score. They help the model pay attention to multiple inputs instead of just one, making the attention process smoother and more adaptive.

stable diffusion: An open-source deep learning model released in 2022, developed by Stability AI in collaboration with academic researchers. This model is primarily used to generate detailed images from text descriptions, but can be applied to other tasks such as restoration, restoration, and image-to-image conversion, guided by text prompts. The current version of Stable Diffusion is called Dream Studio.

supervised learning: Supervised learning is machine learning in which a model is trained using labeled data and learns to map inputs to outputs by minimizing the difference between predicted and actual targets approach. This includes common tasks such as classification and regression.

tokenization: Tokenization refers to the process of breaking text into tokens, the smallest units of meaning. These tokens can be words, subwords, phrases, or characters, depending on the level of granularity you choose. Tokenization is an important preprocessing step for many natural language processing tasks.

token: As mentioned earlier, a token is the smallest unit of meaning in a large language model. Tokens serve as input for AI models that process text, such as large language models, sequence-to-sequence models, and classifiers.

unsupervised learning: A machine learning approach that trains a model using only input data and no corresponding target output. The goal of unsupervised learning is to independently discover patterns, structures, or relationships in data without prior knowledge or guidance.

value vector: A neural network’s attention mechanism involves a value vector that stores information about the input elements. Combine these vectors with the attention scores to create the context vector used to generate the output. Attention scores determine the importance of input factors, allowing the model to focus on the most relevant information.

weight: A neural network parameter that affects neuron connections and is tuned during training to optimize output production that closely matches the target data distribution.

Related article: 5 AI applications to help marketers work more efficiently

AI articles to dig deeper

Now that you understand the terminology used to define the inner workings of AI applications, it’s time to explore some of the specific aspects of AI, particularly those that affect marketing, advertising, content creation, and SEO, as well as AI’s implications for privacy. There are also some interesting articles on issues such as , legal implications, and regulation.

Quick AI link for additional information

Here are quick links to some of the most popular AI organizations and generative AI models.