Skip to content

Instantly share code, notes, and snippets.

@matthew-d-white
Last active February 4, 2024 14:08
Show Gist options
  • Save matthew-d-white/9ee37698bf16a2528267bf429c977221 to your computer and use it in GitHub Desktop.
Save matthew-d-white/9ee37698bf16a2528267bf429c977221 to your computer and use it in GitHub Desktop.
Generative AI Glossary
+----------------------------------------------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Term | Definition |
+----------------------------------------------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| A/B Testing | A carefully planned, practical test intended to contrast A and B, two different iterations of a system or model. |
| Accuracy | Calculated as (True Positives + True Negatives) / (True Positives + True Negatives + False Positives + False Negatives), accuracy is a binary classification score system that determines whether a response or output of AI Model is right or not. |
| Action | The behavior or choice that the agent can make in a given state. |
| Actionable Intelligence | Information you can leverage to support decision making. |
| Activation Function | An artificial neural network function that produces an output value to turn on the following layer by taking the weighted sum of all the inputs from the preceding layer. |
| Active Learning | a unique form of semi-supervised machine learning where a learning agent can interactively ask an oracle (typically a human annotator) for labels at fresh data points. |
| Adam Optimizer | A optimization algorithm for stochastic gradient descent that adapts the learning rate for each parameter based on the historical gradient information. |
| Adversarial Autoencoder | Adversarial Autoencoder (AAE) is a brilliant approach that combines the autoencoder architecture with the adversarial loss concept firstdescribed by GAN. |
| Adversarial Machine Learning | A kind of machine learning known as adversarial machine learning aims to make models more resilient by using adversarial input. With applications including spam filtering, virus detection, and biometric identification, it aims to make machine learning approaches more practical in challenging situations. |
| Agent | An entity that takes actions in an environment in order to achieve a certain goal or maximize a reward signal. |
| AI Ethics | Artificial intelligence is a topic that technology ethics are specifically interested in. Machine learning algorithms that are trained from data may be greatly impacted by biases, which can range from gender to race to age to economic status and anything in between. |
| AI Framework | By offering pre-made solutions, artificial intelligence (AI) frameworks make it simpler and quicker to develop applications for machine learning/deep learning, neural networks, and natural language processing (NLP). TensorFlow, Theano, PyTorch are some examples. |
| Algorithm | A series of steps or instructions given to an AI, neural network, or other device to help it learn on its own. The four most popular types are classification, clustering, suggestion, and regression. |
| Annotation | The method of tagging language data by locating and marking grammatical, semantic, or phonetic components. |
| Anomaly detection. | the method of locating instances of an observation that are unusual or significantly depart from the overall pattern of data. |
| Application Programming Interface (API) | A variety of instructions, procedures, protocols, and objects may be used by software developers to create applications or interact with external systems. |
| Area Under the Curve | a process used in machine learning to choose the best performing model from a set of available models. |
| Artificial General Intelligence (AGI) | A computational system called AGI is capable of performing any intellectual task that a human can. Also known also as Strong AI. |
| Artificial Intelligence (AI) | The simulation of human intelligence in machines that are programmed to think and learn like humans. |
| Artificial Narrow Intelligence (ANI) | ANI, also referred to as weak AI, is a form of artificial intelligence that can only concentrate on one task or issue at a time (e.g. playing a game against a human competitor). This is the type of AI that is currently in use. |
| Artificial Neural Network (ANN) | This arrangement, also known as a neural network, consists of a number of nodes or units that sporadically resemble the processing power of the human brain. |
| Association Rule Learning | a machine learning technique based on rules for identifying interesting relationships between variables in huge data sets. |
| Audio Synthesis | The process of generating new audio signals or sound. |
| Auto-classification | Automated text classification is the automatic classification of text using machine learning, natural language processing (NLP), and other AI-guided approaches in a quicker, more efficient, and more accurate manner. |
| Auto-regressive Models | A type of generative model that generates data by conditioning the probability of each element on the previous elements in a sequence. Examples of auto-regressive models include PixelCNN and WaveNet. |
| Autoencoder | A kind of artificial neural network that generates effective data representations unsupervised and non-linearly, usually to reduce dimensionality. |
| Automated Speech Recognition | a branch of computational linguistics concerned with techniques that allow computers to recognize and convert spoken language into text. |
| Autonomy | The quality of being autonomous is the absence of outside restrictions. Human input is not required for an autonomous machine or vehicle to function. |
| Autoregressive Generative Models | In each step, the distribution of the subsequent sequence element is predicted using the previous sequence elements in autoregressivegenerative models, which implicitly define a distribution across sequences using the Chain Rule for Conditional Probability. Causal conv nets and RNNs are the two primary autoregressive architectures. |
| Backpropagation | Backpropagation is a technique for training neural networks based on a predetermined, desired result for specific test conditions. |
| Backward chaining | The reverse technique involves moving the intended objective or output forward in time to see if there is any evidence to back up those aims or outcomes. |
| Bagging | ML ensemble technique with a focus on stability and accuracy that makes use of several weak learners to enhance the performance of astrong learner. |
| Batch | The set of examples used in one gradient update of model training. |
| Bayesian networks | A graph-based model called the Bayes network, also known as the Bayes model, belief network, and decision network, represents a set of variables and their dependencies. |
| BERT | A pre-trained transformer-based language model developed by Google, which stands for Bidirectional Encoder Representations from Transformers |
| Bias | The model includes assumptions to facilitate learning. Since these presumptions could harm results, the majority of supervised machine learning algorithms perform best with little bias. |
| Bias-Variance Tradeoff | A conflict that develops when data scientists attempt to simultaneously minimize bias and variance prevents supervised algorithms from generalizing outside of their training set. |
| Biases | Additional parameters of a model, besides weights, that are used to adjust the output of the model. They are scalar values that are added to the output of a neuron in a neural network, and they can be thought of as a way to shift the activation function to the left or right. |
| Big data | Big data cannot be processed by standard data processing applications due to its size or complexity. It ranks among the most popular searchterms for artificial intelligence. |
| BigGAN | A specific kind of generative adversarial network called BigGAN was created with the goal of scaling generation to high-resolution, high-fidelity images. |
| Bloom | A large-scale language model developed by OpenAI that can perform a wide range of natural language understanding and generation tasks with high accuracy. |
| Boosting | A family of machine learning algorithms that transform weak learners into strong ones, as well as a machine learning ensemble meta-algorithm for reducing bias and variance in supervised learning. |
| Bounding box | The bounding box is created using visual data that is frequently used for tagging in images or videos. To help the model identify the box as a particular kind of object, the contents are labelled. |
| Brute force search | A query that explores all inputs and is not constrained by clustering or approximations. Usually more expensive and time-consuming, but morethorough. |
| Chatbots | A chat robot that can converse with a human user through text or voice commands. Used by the e-commerce, education, health, and business sectors to facilitate communication and provide user support. |
| ChatGPT | A conversational language model developed by OpenAI that can generate human-like text in response to a given prompt. |
| Classification | A category-assignment algorithm technique that enables computers to classify data points. |
| Clustering | Machines can organize similar data into larger data categories using this algorithmic technique. |
| Cognitive Computing | An automated model that uses data mining, natural language processing, and pattern recognition to imitate human thought processes. |
| Collaborative Filtering | A technique used in recommender systems to gather user preferences from a larger user base in order to predict a user's interests. |
| Computer Vision | A multidisciplinary scientific field that looks into the high-level programming of computers to comprehend digital images or videos. |
| Confidence Interval | An interval estimate that is most likely to contain the actual value of a population parameter that is unknown. The interval has a confidence level attached to it that expresses how confident we are that this parameter is within the interval. |
| Convolutional Neural Network (CNN) | Deep artificial neural networks called convolutional neural networks are used to classify images (e.g., identify what they see), organize theminto groups based on similarities (photo search), and identify objects in scenes. |
| Corpus | A sizable body of text or audio that can be used to train a computer to perform linguistic functions. |
| Cross Validation | A method used in machine learning to assess a model's generalizability by comparing it to one or more validation datasets. |
| CycleGAN | In order to enable training without the necessity for paired data, CycleGAN implements a cycle consistency loss. With no one-to-one mapping between the source and target domains, it can translate from one domain to another. |
| DALL-E 2 | An updated version of DALL-E, a model developed by OpenAI that generates images from text descriptions. |
| Data Discovery | The method of extracting data insights and delivering those insights to the users who require them at the appropriate time. |
| Data Extraction | The process of gathering or retrieving various types of data from various sources, many of which may be erratically organized or entirely unstructured, is known as data extraction. |
| Data Ingestion | A method for assembling disparate data from various sources, reformatting it, and then importing it into a standard format or repository to make it more accessible. |
| Data Labelling | A method of data marking that allows machines to recognize objects. To create metadata that is used to train AI models, information is added to different data types (text, image, etc.). |
| Data Scarcity | The absence of data that could potentially satiate the system's need to improve the predictive analytics' level of accuracy. |
| Deep Convolutional GAN (DCGAN) | An architecture for generative adversarial networks is called Deep Convolutional GAN (DCGAN). It follows the following rules in particular strided convolutions (discriminator) and fractional-strided convolutions (generator) are used to replace any pooling layers. Utilizing batchnorm in the discriminator and generator. |
| Deep Learning | A subfield of machine learning that deals with neural networks with multiple layers that can learn representations of the data with multiple levels of abstraction. |
| Descriptive Analysis | The process of going back and looking at old information, usually to report on it, explain it, or come up with new models for past or presentevents. |
| Deterministic | A process that follows a definite set of rules or laws and is predictable. |
| Diffusion Probabilistic Models | A class of generative models that use a diffusion process to generate new samples by gradually transforming a noise input. |
| Dimensionality Reduction | A data preparation method that converts high-dimensional data to a low-dimensional representation in order to decrease the amount of input features in a dataset. |
| Discriminative Model | A group of models that frequently predict labels from a set of features when used for classification or regression. Comparable to supervised learning. |
| Discriminator | The network in a GAN that classifies the input as real or generated. |
| Energy Based Models | A class of generative models that define a probability density function over the data, and use an energy function to assign a scalar value to each point in the data space. |
| Ensemble methods | Multiple learning algorithms are used in ensemble methods to produce better predictive performance than any one of the individual learning algorithms could. A machine learning ensemble is typically much more flexible than a statistical ensemble in statistical mechanics, which typically consists of an infinite number of alternative models. Instead, it only includes a concrete finite set of alternative models. |
| Entity Annotation | Information extraction is the process of labelling unstructured phrases with data so that a computer can understand them. This might entail naming every individual, group, and place mentioned in a document. |
| Entity Extraction | a catch-all phrase used to describe the process of organizing data so that computers can understand it. Entity extraction may be carried out by people or a machine learning algorithm. |
| Environment | The system or domain in which an agent interacts. It provides the agent with observations and rewards and responds to the agent's actions. |
| Epoch | In the context of training Deep Learning models, one pass of the full training data set. |
| F-Score | A way to calculate the score that accounts for both precision and recall when evaluating a model's accuracy. The F-Score, in more precise terms, is the harmonic average of precision and recall, reaching a maximum value of 1 (perfect precision and recall) and a minimum value of 0. |
| False Negative | An error due to the fact a result did not dismiss the null hypothesis when it should have. |
| False Positive | An error coupled with the fact a result did dismiss the null hypothesis when it shouldn't have. |
| Feature | A variable that serves as a model's input. |
| Feature Learning | A collection of methods designed to extract from raw data the representations required forfeature detection or classification. |
| Flow-based Generative Models | An explicit modelling of a probability distribution using normalizing flow, a statistical technique that applies the change-of-variable law of probabilities to turn a simple distribution into a complex one, is what is known as a flow-based generative model. This type of model is used in machine learning. |
| Gated Recurrent Unit (GRU) | A type of recurrent neural network that uses gating mechanisms to control the flow of information through the network. |
| Generative Adversarial Networks (GANs) | Using two neural networks in competition with one another (hence the name adversarial), generative adversarial networks (GANs) are algorithmic architectures that produce new, artificial instances of data that can be mistaken for real data. They are extensively used for voice, video, and image generation. |
| Generative Adversarial Networks (GANs) | A type of deep learning model that consists of two neural networks, a generator and a discriminator, that are trained to compete against each other. The generator tries to generate samples that are similar to real data, while the discriminator tries to distinguish between real and generated data. |
| Generative AI | Artificial intelligence (AI) that can create new content, as opposed to just analyzing or acting on pre-existing data, is known as generative AI. |
| Generative Flow Networks (GFlowNets) | Generative Flow Networks (GFlowNets) are newly developed models for learning stochastic policies that build compositional items throughsequences of actions with the probability proportional to a particular reward function. |
| Generative Latent Optimization (GLO) | A system for employing straightforward reconstruction losses to train deep convolutional generators. |
| Generative Models for Images | Generative models that are trained on image data and can generate new images that are similar to the training data. Examples include GANs, VAEs, DALL-E, and StyleGan |
| Generative Models for Text | Generative models that are trained on text data and can generate new text that is similar to the training data. Examples include GPT-3, BERT and CLIP. |
| Generative Pre-trained Transformer (GPT) | A deep learning language model called a Generative Pre-Trained Transformer (GPT) can produce texts that resemble human writing from text input. |
| Generative Pre-training | A technique where a generative model is pre-trained on a large dataset and then fine-tuned on a smaller task-specific dataset. This allows the model to leverage the knowledge learned from the large dataset to improve its performance on the smaller task-specific dataset. |
| Generator | The network in a GAN that generates new samples by transforming a random noise input. |
| Genetic algorithm | An approach based on genetics known as a genetic algorithm, or GA, is used to quickly andeffectively solve complex problems. |
| GPT-1 | A language model developed by OpenAI that uses a transformer architecture to generate text. |
| GPT-2 | A improved version of GPT-1, a language model developed by OpenAI with more parameters and a larger dataset, that uses a transformer architecture to generate text. |
| GPT-3 | A state-of-the-art language model developed by OpenAI that uses a transformer architecture and can perform a wide range of natural language processing tasks with high accuracy. |
| GPT-J | A language model developed by OpenAI, similar to GPT-3, that is fine-tuned on a specific set of tasks. |
| GPT-Neo | A variant of GPT-3, that can be fine-tuned on any number of parameters, that allows for more flexibility, and can be fine-tuned on a specific task or dataset with less computational resources. |
| Gradient boosting | An ML technique that trains a group of weak prediction models, like decision trees, iterativelyin order to strengthen or produce the output. |
| Gradient Descent | An optimization algorithm that iteratively adjusts the parameters of a model to minimize a loss function.
| Hallucination | A concept in large natural language processing (NLP) models or large language models (LLMs) when the model produces non-existent or wrong responses but does so in a manner that appears confident of the response.
| Heuristic | A method developed in computer science for rapid, effective, solution-based problem solving. |
| Hidden Layer | A structure within a neural network that performs a specific function, such as an activation function for model training, between the input and output layers. |
| Hyperparameters | Hyperparameters are variables that affect your model's learning capabilities. Usually, theyare manually adjusted outside of the model. |
| Image Segmentation | In order to make the representation of an image easier to analyze, image segmentation is the process of breaking a digital image into several parts or fragments. Segmentation divides entire images into labelled and categorized groups of pixels. |
| Image Synthesis | The process of generating new images from existing images or other input data. |
| Image-to-Image | A task where the goal is to transform one image into another, such as style transfer or super-resolution. |
| Inference | The process of using a trained model on data to produce a model output, such as a score,prediction, or classification, in machine learning. |
| Inpainting | A task where the goal is to fill in missing or corrupted parts of an image. |
| Intelligent Process Automation (IPA) | A group of technologies used to automate some digital processes, such as robotic processautomation (RPA) and artificial intelligence (AI). |
| Intent | This kind of label, which describes the intention or goal of what is said, is frequently used in training data for chatbots and other natural language processing tasks. |
| K-nearest Neighbors | An approach to classification and regression using supervised learning, where the model input is the k closest training examples in the data setand the output is either a class assignment (classification) or a property value (regression), is used to calculate the likelihood that a data point belongs to a group. |
| Keras | An open-source neural network library written in Python that can run on top of TensorFlow, Theano, and Microsoft Cognitive Toolkit (CNTK). |
| Label | A part of training data that indicates the intended result for a given piece of training data. |
| Latent Space | The space of the encoded representation of the data in a generative model, such as VAEs or GANs. It is usually a lower-dimensional space than the original data space. |
| Layer | A group of neurons or computational units that are connected and perform a specific computation on the input data. In deep learning, layers are stacked on top of each other to form a hierarchy of representations, where each layer extracts increasingly complex features or patterns from the data. |
| Learning Rate | A scalar value multiplied by the gradient at each iteration of the training phase of an artificial neural network using the gradient descent algorithm. |
| Least Squares GAN | The generative adversarial network known as LSGAN, or least squares GAN, uses the least squares loss function for the discriminator. ThePearson divergence is minimized by minimizing the objective function of LSGAN. |
| Long Short-Term Memory Networks | A variation of Recurrent Neural Network suggested as a solution to the vanishing gradientproblem. |
| Machine Learning | A field of study that gives computers the ability to learn without being explicitly programmed. |
| Markov Chain | A mathematical model for a sequence of possible events in which the probability of each event depends only on the state attained in the previous event. |
| Maximum Likelihood Estimation | A method to estimate the parameters of a statistical model, by finding the parameter values that maximize the likelihood of the observed data. |
| MidJourney | A dataset and model developed by OpenAI for image synthesis, where the model generates images of outdoor scenes during the day. |
| Mode Collapse | A problem in generative models where the model generates only a limited number of distinct examples, rather than producing a diverse set of outputs. |
| Model | A model is a simplified representation of a system or process that is used to make predictions or understand the underlying relationships and mechanisms. It can be mathematical or computational and is trained on a dataset to make predictions or decisions on new data. |
| Model (Machine Learning) | A set of mathematical algorithms and parameters that is trained on a dataset to make predictions or decisions on new data. It is a tool that can learn from data and improve its performance over time. |
| Monte Carlo | A rough methodology that creates fake, simulated data by repeatedly sampling at random. |
| Monte Carlo Tree Search | A best-first search algorithm that uses simulated playouts to guide the search for the best move in a game. |
| Multi-Modal Learning | A branch of machine learning that aims to combine multimodal signals, process information from various types of data, and create models thatcan relate it. Modes can be image, text or audio. |
| Natural Language Processing | A subfield of AI that deals with the interaction between computers and human languages, including speech recognition, natural language understanding, and natural language generation. |
| Neuron | a computational unit that receives input, performs a computation, and produces an output. In an artificial neural network, a neuron is responsible for processing the input data and passing it along to the next layer of the network. Each neuron has an associated weight and bias, which are learned during the training process. |
| Normalizing Flow Models | A class of generative models that use a series of invertible transformations to map a simple random noise input to a sample from the target distribution. |
| Optimization | Choosing the best component (based on some criteria) from a range of available options. |
| Outpainting | A task where the goal is to generate new but coherent content beyond the boundaries of an image. |
| Output Layer | The final layer of a neural network, marking the conclusion of a model's workflow, is in charge of providing the outcome or response, such as ascore, a classification, or a prediction. |
| Overfitting | The phrase overfitting is frequently used in AI. When a machine learning algorithm can only use or recognize specific examples from the training data, it is said to be overfit. A working model should be able to extrapolate data patterns to address completely new situations. |
| Parameter | A factor within the model that aids in prediction. Data can be used to estimate a parameter's value because they are typically not set by theperson running the model. |
| Policy | A function that maps states to actions and determines the behavior of the agent. |
| Precision | In machine learning, an evaluate of model accuracy computing the ratio of true positives against all true and false positives in a given class. |
| Predictive Analysis | The method of generating predictions, insights, recommendations, or other types of assessments of the likelihood of future outcomes using historical patterns and trends in data. |
| Principle Component Analysis (PCA) | A method known as principal component analysis transforms a set of observations of potentially correlated variables into a set of linearly uncorrelated variables. |
| Probability | A measure of the likelihood of an event occurring. It is a number between 0 and 1, where 0 represents an impossible event and 1 represents a certain event. |
| ProGAN | A generative adversarial network that uses a progressively growing training method is called ProGAN, or Progressively Growing GAN. Thegoal is to gradually increase both the generator and discriminator; starting with a low resolution, then gradually add new layers that simulate increasingly finer details. |
| Prompt Engineering | The process of carefully crafting a prompt for a language model to generate coherent and creative text. |
| PyTorch | An open-source machine learning library developed by Facebook's AI Research lab. It provides a flexible neural network library for applications such as computer vision and natural language processing. |
| Q-function | A function that estimates the expected long-term reward of an agent when taking a particular action in a given state and following a particular policy thereafter. |
| Q-Learning | A type of reinforcement learning algorithm that uses a table or function to approximate the optimal action-value function. |
| Recurrent Neural Network (RNN) | A type of artificial neural network that can exhibit dynamic temporal behavior for a time sequence and process sequential signals using internal state (memory) thanks to connections between neurons that form a directed graph along a sequence. |
| Reinforcement Learning | A type of machine learning that involves training agents to make decisions by interacting with an environment and receiving feedback in the form of rewards or punishments. |
| Reinforcement Learning from Human Feedback (RLHF) | A subfield of reinforcement learning that deals with the integration of human feedback into the learning process of an agentt, which involves incorporating feedback from human evaluators, such as supervisors or users, into the agent's learning process to improve the performance of the model. |
| Reward | A scalar value that the environment provides to the agent as feedback for its actions. It represents the degree to which the agent's actions are successful in achieving its goal. |
| SARSA | A type of reinforcement learning algorithm that uses the current state and action, as well as the next state and action, to update the action-value function. |
| Score-Based Models | A class of generative models that define a likelihood function of the data and use a scoring function to assign a scalar value to each point in the data space. |
| Sequence Based Models | A type of machine learning model that is designed to handle sequential data, such as time series or natural language text. |
| Simluation | A mathematical model that represents the behavior of a system or process over time. It allows for the prediction of future outcomes and the testing of different scenarios without the need for real-world experimentation. It can be used in various fields such as physics, engineering, finance, and social sciences, and it can also be used in virtual training or gaming. |
| Stable Diffusion | A method to improve the stability of generative models, such as GANs, by introducing a diffusion process that gradually transforms a noise input into a sample from the target distribution. |
| State | The representation of the current condition or situation of the environment. It includes all the information that the agent needs to make a decision. |
| State of the Art (SOTA) | The highest level of performance for a specific task or benchmark as reported in academic literature or industry. |
| Stochastic | A process that involves randomness or probability. |
| Style Transfer | A technique that uses deep learning to transfer the style of one image to another image. |
| StyleGAN | Researchers at NVIDIA created the machine learning framework known as StyleGAN in December 2018. It demonstrated a paradigm shift inthe consistency and quality of realistic images produced by AI, particularly in terms of its ability to produce realistic human faces. |
| Synthetic Data | Artificial data generated by a generative model to augment a dataset. Synthetic data can be used to increase the size of a dataset, to balance class distributions, or to generate samples of rare events. |
| Synthetic Reality | The technology that creates realistic simulations of environments, objects and scenarios, which can be used for various applications like gaming, virtual reality, industrial design, and more. |
| TensorFlow | An open-source software library for machine learning developed by Google Brain Team. |
| Testing | The process of evaluating a model's performance on a set of unseen data, to estimate its generalization error. |
| Text-to-Image | A task in which a model generates an image from a text description or caption. |
| Text-to-Video | A task in which a model generates a video from a text description or caption. |
| Training | The process of adjusting the parameters of a model to optimize its performance on a given task, using a set of labeled training data. |
| Transformers | A type of neural network architecture that is designed to handle sequential data and has been successfully applied to natural language processing tasks. |
| Value Function | A function that estimates the expected long-term reward of an agent when following a particular policy in a given state. |
| Variance | A statistical measure that quantifies the spread of a set of data around its mean. |
| Variational Autoencoders (VAEs) | A type of generative model that learns to encode data into a latent representation and decode the latent representation back into data. VAEs use a probabilistic approach to modeling the data, where the encoded representation is assumed to be a sample from a latent distribution. |
| Video Synthesis | The process of generating new videos from existing videos or images. |
| Vision Transformer (ViT) | A type of neural network architecture that is designed to handle image data by treating the image as a sequence of patches, similar to how a transformer architecture handles text data. |
| Voice Synthesis | The process of generating artificial speech or voice. |
| Wasserstein GAN (WGAN) | The Wasserstein Generative Adversarial Network is a variation on the generative adversarial network with the goal of improving learning stability, removing issues like mode collapse, and providing meaningful learning curves useful for debugging and hyperparameter searches. |
| Weights | The parameters of a deep learning model that are used to make predictions. They are learned during the training process, and they represent the strength of the connections between the different layers of artificial neural networks. They are used to determine the output of the model for a given input, and are adjusted during the training process to improve the model's accuracy. |
+----------------------------------------------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment