Glossary of Categories

Glossary of Categories

Term	Description
Active learning	A machine learning approach where the model selectively queries an oracle (e.g., a human annotator) for labels on the most informative samples, reducing the amount of labeled data needed for training.
Agentic coding	Using AI coding agents (e.g., Claude Code, GitHub Copilot’s coding agent, Cursor, OpenCode) that can read a repository, run commands, edit multiple files, and revise their own output, rather than only suggesting code. Best practices emphasize limiting what the agent can access, working feature by feature, and verifying everything the agent produces.
Agriculture	The application of machine learning in agricultural practices, including crop prediction, pest detection, and yield optimization. Often involves satellite imaging and IoT data.
AI safety	Making ML systems behave as intended under adversarial conditions: robustness to data poisoning and backdoors, resistance to prompt injection, and alignment with user goals. Used for resources on attacks against models and their defenses.
AIF360 (AI Fairness 360)	An open-source toolkit from IBM for detecting and mitigating bias in machine learning models. Provides metrics for checking fairness and algorithms for bias mitigation across the ML pipeline.
Audio data	Data in audio format (e.g., speech, music, environmental sounds) used as input for machine learning tasks such as classification, transcription, and generation.
Audio search	Using machine learning to search, identify, and retrieve audio content based on acoustic features, metadata, or content-based queries.
AWS (Amazon Web Services)	Amazon’s cloud computing platform offering a wide range of services for ML, including SageMaker for model training and deployment, and Bedrock for foundation model access.
Azure	Microsoft’s cloud computing platform, providing ML services such as Azure Machine Learning, Azure AI services, and Azure OpenAI for building and deploying AI solutions.
Bedrock	Amazon Bedrock, an AWS service that provides access to foundation models from leading AI companies through a unified API for building generative AI applications.
Benchmarks	Evaluating and comparing models or tools on standardized tasks and metrics. Tags both the Benchmarks section (pages about a specific benchmark or leaderboard) and any resource where benchmark results or leaderboard readings matter.
Bias	Systematic errors or unfairness in ML models that can arise from training data, algorithm design, or evaluation methods. Addressing bias is critical for building equitable AI systems.
Biodiversity	Machine learning applied to the variety of life, such as species identification and population monitoring, often using ecological image datasets like iNaturalist and BioTrove.
Biology	The application of machine learning in biological sciences, including areas like genomics, ecology, protein modeling, and drug discovery.
Blogs	Blog-format resources providing accessible explanations, tutorials, and commentary on machine learning topics.
Books	Comprehensive resources for learning and reference. Often written by experts, these provide in-depth coverage of machine learning topics.
Boosting	An ensemble machine learning technique that combines multiple weak learners sequentially, where each new model focuses on correcting errors made by the previous ones. Includes algorithms like AdaBoost, Gradient Boosting, and XGBoost.
Camera trap	The application of machine learning to classify and analyze images from remote wildlife cameras, aiding ecological research and conservation monitoring.
Carpentries	Lessons from The Carpentries, a volunteer community that teaches foundational coding and data skills through live-coded, hands-on workshops. UW-Madison has a local Carpentries community that develops and teaches ML/AI lessons.
Citizen science	Public participation in scientific research, often aided by ML for tasks like data collection, labeling, and analysis at scale.
Classical ML (Classical Machine Learning)	Refers to traditional ML algorithms like SVMs, decision trees, and k-means clustering. These methods are well-established, interpretable, and often less resource-intensive than deep learning.
Classification	Supervised learning that assigns inputs to discrete categories, such as labeling tweets as disaster-related or not. Compare with regression, which predicts continuous values.
CLIP (Contrastive Language-Image Pre-training)	An OpenAI model that learns visual concepts from natural language supervision, connecting text and images in a shared embedding space. Enables zero-shot image classification and cross-modal retrieval.
Cloud	Cloud computing platforms and services (e.g., AWS, GCP, Azure) used for ML model training, deployment, and scaling. Provides on-demand access to GPUs and other resources.
Clustering	The process of grouping a set of objects in such a way that objects in the same group (cluster) are more similar to each other than to those in other groups. Common algorithms include k-means and DBSCAN.
CNN (Convolutional Neural Network)	A class of deep learning models well suited to image data. Known for automatically learning spatial hierarchies of features. Implemented in PyTorch, TensorFlow, and Keras.
Code-along	Resources with code to run as you go, from live-coded workshops to books and notebooks with runnable examples. Learning by doing rather than by reading alone.
Compute	Refers to computational resources, often in the context of machine learning, such as CPUs, GPUs, and cloud computing. Critical for training models, especially deep learning models.
Computer vision	A field of machine learning focused on enabling computers to interpret and make decisions based on visual data. Applications include image recognition, object detection, and facial recognition. Libraries include OpenCV, PyTorch, and TensorFlow.
Conformer	A convolution-augmented Transformer architecture that combines self-attention with convolutional layers, commonly used in speech and audio processing tasks.
Contrastive learning	A self-supervised learning approach that trains models by contrasting positive pairs (similar samples) against negative pairs (dissimilar samples) to learn meaningful representations without labeled data.
CSI (Cover Song Identification)	A task in music information retrieval that involves identifying whether one song is a cover version of another. Machine learning models for CSI often analyze harmonic, melodic, and rhythmic similarities across versions.
Data	Datasets and data sources for ML projects, with notes on access, licensing, and example applications.
Decision trees	A tree-structured ML model that makes predictions by learning decision rules from data features. Interpretable and widely used for both classification and regression tasks. Forms the basis for ensemble methods like Random Forests and Gradient Boosting.
Deep learning	A subset of machine learning involving neural networks with many layers, used for complex tasks like image recognition and natural language processing. Commonly used libraries include TensorFlow, PyTorch, and Keras.
Diffusion	Diffusion models are a class of generative models that learn to create data by reversing a gradual noising process. Used for high-quality image, audio, and video generation. Examples include Stable Diffusion and DALL-E.
Earth science	Applications of ML to the Earth’s physical systems, including hydrology, climate, geology, and remote sensing.
Ecology	The application of machine learning in ecological research, including species distribution modeling, ecosystem monitoring, and biodiversity analysis.
EDA (Exploratory Data Analysis)	The process of analyzing and visualizing datasets to summarize their main characteristics, discover patterns, and inform subsequent modeling decisions. A critical first step in any ML project.
Education	Resources related to teaching and learning about machine learning, including curricula, pedagogical approaches, and educational tools.
EHR (Electronic Health Records)	The application of machine learning to electronic health record data for tasks such as clinical prediction, patient risk stratification, and treatment recommendation.
Energy	The application of machine learning in the energy sector, including energy consumption forecasting, grid optimization, and renewable energy management.
Ethical AI	Practices and principles for developing AI systems that are fair, transparent, accountable, and aligned with human values. Encompasses bias mitigation, privacy, and responsible deployment.
Explainability	Methods and techniques for making ML model decisions understandable to humans, such as feature importance, attention visualization, and SHAP values. Related to interpretability and trustworthy AI.
Exploring AI@UW	A series exploring AI research, applications, and initiatives at the University of Wisconsin-Madison, showcasing interdisciplinary work across campus.
Fairness	Ensuring ML models treat different demographic groups equitably. Involves metrics for measuring disparate impact and algorithms for mitigating unfair outcomes.
Fine-tuning	Continuing to train a pretrained model on a smaller task-specific dataset. Parameter-efficient methods such as LoRA update only a small set of added weights, so large models can be adapted on a single GPU.
First-steps	Introductory resources designed for beginners getting started with machine learning concepts, tools, and workflows.
Foundation models	Large-scale pre-trained models that serve as a base for fine-tuning or prompting on specific tasks. They underpin most current NLP and computer vision systems. Examples include GPT-4, Llama, CLIP, and Segment Anything.
GCP (Google Cloud Platform)	Google’s cloud computing platform offering ML services such as Vertex AI, TPU access, and AutoML for building and deploying machine learning models.
Gemini	Google’s family of multimodal AI models capable of processing and generating text, images, audio, and code. Successor to earlier models like PaLM.
GenAI (Generative AI)	AI models that can create new content including text, images, audio, and code. Encompasses technologies like large language models, diffusion models, and GANs.
Genomics	The study of genomes, often involving the analysis of DNA sequences. Machine learning aids in tasks like gene prediction, mutation analysis, and personalized medicine.
Geospatial data	Geographic and spatial data used in ML applications such as remote sensing, environmental monitoring, and location-based analytics.
Git/GitHub	Version control system (Git) and the associated platform (GitHub) for hosting and sharing code. Essential tools for collaboration and project management in software development, including ML projects.
Google	Tools and services from Google used in ML workflows, such as Google Colab, Google Cloud Platform (GCP), Gemini, and NotebookLM.
GPU (Graphics Processing Unit)	A specialized processor widely used in deep learning for its parallel processing capabilities. Essential for training large neural networks efficiently.
Guides	Detailed instructions or explanations, often in the form of tutorials or documentation, aimed at helping users understand and apply specific concepts or tools.
Healthcare	The application of machine learning in the healthcare industry, including areas like medical imaging, diagnostics, and personalized treatment plans. Common challenges include data privacy and interpretability.
HPC (High-Performance Computing)	Tightly coupled compute clusters where a single job can span many nodes and GPUs, such as CHTC’s HPC cluster. Suited to large distributed training runs and simulations.
HTC (High-Throughput Computing)	Running many independent jobs across a pool of machines, the model behind CHTC’s HTC system. Suited to hyperparameter sweeps, batch inference, and other embarrassingly parallel workloads.
Hugging Face	A popular platform for sharing pre-trained models, datasets, and other machine learning resources, especially in NLP. Provides tools like the Transformers library for easy model deployment.
Humanities	The application of machine learning in humanities research, including digital humanities, text analysis of historical documents, and computational approaches to cultural studies.
Image classification	The ML task of assigning a label or category to an entire image based on its visual content. A foundational task in computer vision with applications across many domains.
Image data	Data in image format (e.g., photographs, medical scans, satellite imagery) used as input for computer vision and other ML tasks.
Image restoration	Recovering a clean image from a degraded one by removing noise, blur, compression artifacts, or signal-specific distortions. Deep learning approaches are typically trained on paired degraded and clean images; examples include microscopy denoising, low-dose CT, and archival video cleanup.
Image segmentation	The ML task of partitioning an image into meaningful regions or segments, assigning a label to each pixel. Used in medical imaging, autonomous driving, and satellite image analysis.
Industry applications	Refers to the use of machine learning across various industries such as finance, manufacturing, and logistics, for tasks like predictive maintenance, fraud detection, and supply chain optimization.
Interpretability	The degree to which humans can understand the behavior, predictions, and decision-making process of an ML model. Critical for trust, debugging, and regulatory compliance.
Jupyter	An open-source interactive computing environment for creating notebooks that combine live code, equations, visualizations, and narrative text. Widely used in data science and ML workflows.
Keras	A high-level neural networks API in Python. Keras 3 runs on top of TensorFlow, PyTorch, or JAX and allows fast prototyping of deep learning models in a few lines of code.
Libraries	Software libraries and toolkits that provide pre-built functions and tools for machine learning development, such as PyTorch, TensorFlow, Scikit-learn, and Hugging Face Transformers.
LLaVA (Large Language and Vision Assistant)	A multimodal AI model that combines language and vision understanding, capable of processing and generating both text and images.
LLM (Large Language Model)	A transformer-based model trained on large text corpora to generate and reason over text. Examples include GPT-4, Claude, Gemini, Llama, and Qwen.
LMM (Large Multimodal Model)	A large language model extended to take images, audio, or video as input alongside text. Examples include LLaVA, Qwen-VL, and GPT-4o.
LSTM (Long Short-Term Memory)	A type of recurrent neural network (RNN) designed to learn long-term dependencies in sequential data. Widely used in time-series prediction, speech recognition, and text generation.
Medical imaging	The application of machine learning techniques to analyze and interpret medical images. Common tasks include segmentation, classification, and anomaly detection. Libraries include MONAI, PyTorch, and TensorFlow.
Meteorology	The application of machine learning in weather and climate science, including weather forecasting, climate modeling, and extreme event prediction.
Microsoft Copilot	Microsoft’s AI assistant powered by large language models, integrated across Microsoft products for tasks like code completion, document drafting, and data analysis.
ML Marathon	The Machine Learning Marathon, ML+X’s annual 12-week ML/AI hackathon. Tags project pages for challenges run during a Marathon season.
ML+X (Machine Learning + X)	The concept of combining machine learning with domain expertise across disciplines. The core focus of the ML+X Nexus community at UW-Madison.
ML4MI (Machine Learning for Medical Imaging)	A UW-Madison initiative focused on applying machine learning methods to medical imaging problems, fostering collaboration between computer scientists and medical researchers.
MLM24	Challenges and materials from the 2024 Machine Learning Marathon season.
MLM25	Challenges and materials from the 2025 Machine Learning Marathon season.
MLM26	Challenges and materials from the 2026 Machine Learning Marathon season.
MLOps (Machine Learning Operations)	Practices and tools for deploying, monitoring, and maintaining ML models in production environments. Bridges the gap between ML development and reliable system operation.
Model exploration	Libraries that facilitate trying out different model architectures and pretrained models.
Model sharing	Practices and platforms for sharing trained ML models with the community, enabling reproducibility and reuse. Platforms include Hugging Face Hub and PyTorch Hub.
Models	Resources providing access to pre-trained or reference ML model implementations for various tasks and domains.
Multilingual	Machine learning approaches and models that work across multiple languages, including multilingual NLP, translation, and cross-lingual transfer learning.
Multimodal learning	Machine learning approaches that process and integrate information from multiple data modalities (e.g., text and images, audio and video) to improve understanding and predictions.
Music	The application of machine learning in music analysis, generation, classification, and information retrieval.
Neuroscience	Applications of ML to brain and nervous system data, such as decoding speech from neural recordings or analyzing neuroimaging.
NLP (Natural Language Processing)	A field of machine learning focused on the interaction between computers and humans using natural language. Common libraries include NLTK, SpaCy, and Hugging Face Transformers.
Notebooks	Jupyter or Quarto notebook-format resources that combine executable code, visualizations, and narrative text for interactive learning and reproducible analysis.
Object detection	The ML task of identifying and localizing objects within images by drawing bounding boxes and assigning class labels. Common models include YOLO, Faster R-CNN, and DETR.
OCR (Optical Character Recognition)	The use of machine learning to convert images of text (handwritten, printed, or scanned) into machine-readable text. Applications include document digitization and text extraction.
OOD detection (Out-of-Distribution Detection)	Techniques for identifying data points that do not belong to the distribution on which a model was trained. Important for building robust and trustworthy models. Common methods include Mahalanobis distance and energy-based models.
Physics	The application of machine learning in physics research, including particle physics, condensed matter, and physics-informed neural networks.
Projects	Hands-on ML/AI projects from the UW-Madison community, including Machine Learning Marathon challenges and lab collaborations.
Prompt engineering	Techniques for designing and optimizing input prompts to elicit desired behaviors from large language models and other generative AI systems.
Protein engineering	Using machine learning to design, modify, and optimize protein sequences and structures for desired properties, accelerating drug design and biotechnology.
Protein language models	Language models trained on protein amino acid sequences to learn evolutionary and structural patterns, used for predicting protein function, structure, and fitness.
PyTorch	An open-source deep learning framework originally developed by Meta AI and now governed by the PyTorch Foundation. The most common framework for research in computer vision and natural language processing.
PyTorch-OOD	A PyTorch library providing implementations of out-of-distribution detection methods, making it easy to benchmark and apply OOD detection techniques.
Quantization	Storing model weights, and sometimes activations, in lower-precision formats such as 8-bit or 4-bit integers to cut memory use and speed up inference, at some cost in accuracy.
RAG (Retrieval-Augmented Generation)	A technique that combines information retrieval systems with large language models, allowing the model to access and reference external knowledge to generate more accurate and grounded responses.
Regression	A supervised ML technique for predicting continuous numerical values based on input features. Common methods include linear regression, polynomial regression, and neural network-based regression.
Remote sensing	The application of machine learning to analyze satellite, aerial, or drone imagery for tasks like land use classification, environmental monitoring, and disaster response.
Reproducibility	Ensuring that ML experiments and results can be consistently replicated by others, a key principle in scientific research. Often involves detailed documentation, version control, and the use of containers.
Retrieval	Information retrieval techniques that search and rank relevant documents or data, often used in conjunction with LLMs for retrieval-augmented generation and semantic search.
RNN (Recurrent Neural Network)	A neural network architecture that processes sequences one step at a time while carrying a hidden state. Used for time-series and language data before transformers became standard; LSTM and GRU are common variants.
Science communication	The practice of communicating scientific and ML research findings to broader audiences, including visualization, writing, and presentation techniques.
Security	Practices for protecting code, data, and credentials, including keeping API keys and secrets out of files an AI agent could read, sandboxing or containerizing agent execution, vetting third-party packages and models, and never running unvetted tools on machines holding restricted data.
Signal processing	Techniques for analyzing, modifying, and synthesizing signals (e.g., audio, sensor data), often combined with ML methods for tasks like noise reduction, feature extraction, and pattern recognition.
SILO	UW-Madison’s Systems, Information, Learning and Optimization seminar, a weekly cross-departmental talk series on machine learning, optimization, and information theory.
Simulations	Using computational simulations to generate synthetic data or validate ML models, common in fields like physics, biology, and engineering where observed data may be scarce.
Sklearn (Scikit-learn)	A machine learning library for Python that provides simple and efficient tools for data mining and data analysis, built on NumPy, SciPy, and Matplotlib.
Spectral analysis	The analysis of frequency-domain representations of signals or data, used in audio processing, remote sensing, and other domains alongside ML methods.
Sustainability	The application of machine learning to environmental sustainability challenges, including climate modeling, resource optimization, and green computing practices.
Tabular	Machine learning methods and resources focused on structured tabular data (rows and columns), including techniques like gradient boosting, random forests, and neural approaches for tabular prediction.
Tensorflow	Google’s open-source deep learning framework. Keras is its high-level API, and many libraries, including Hugging Face Transformers, support both TensorFlow and PyTorch.
Text analysis	The process of deriving information from text data, often involving techniques from NLP. Used in sentiment analysis, topic modeling, and more. Common libraries include NLTK, SpaCy, and Gensim.
Time-series	Machine learning methods for analyzing and predicting sequential, time-dependent data. Applications include forecasting, anomaly detection, and signal processing.
Transfer learning	A machine learning technique where knowledge gained from training on one task is applied to improve performance on a different but related task, reducing the need for large labeled datasets.
Transformer	A neural network architecture based on self-attention mechanisms, foundational to modern NLP and increasingly used in computer vision. Originally introduced in the “Attention Is All You Need” paper.
Trustworthy AI	Approaches and practices that ensure AI systems are reliable, fair, transparent, and robust, especially in critical applications like healthcare or finance. Encompasses fairness, explainability, and safety.
U-Net	A convolutional neural network architecture originally designed for biomedical image segmentation, featuring a symmetric encoder-decoder structure with skip connections for precise localization.
Unsupervised learning	Machine learning approaches that learn patterns and structure from unlabeled data, including clustering, dimensionality reduction, and generative modeling.
UW-Madison (University of Wisconsin-Madison)	Resources, research, and initiatives from the University of Wisconsin-Madison related to machine learning and AI.
Videos	Video-format resources including recorded lectures, tutorials, conference talks, and workshop recordings for learning machine learning topics.
Visualization	Techniques for visually representing data, model architectures, training progress, and ML model outputs to aid understanding and communication.
ViT (Vision Transformer)	Based on the transformer architecture, adapted for vision tasks. Images are split into patches (usually 16x16), which are flattened and treated as input tokens, then processed similarly to how words are processed in NLP transformers like BERT.
VLM (Visual-Language Model)	A type of machine learning model that understands and generates content based on both visual and textual inputs, often used in tasks like image captioning and visual question answering. Examples include CLIP and LLaVA.
Workshops	Workshop-format educational resources providing structured, hands-on learning experiences on specific machine learning topics and tools.
Zero-shot learning	A machine learning paradigm where models can classify or generate outputs for classes or tasks not seen during training, often leveraging semantic knowledge or natural language descriptions.