Glossary

Search results

Parameter-Efficient Fine-Tuning(also: PEFT, Lightweight Fine-Tuning)

Parameter-efficient fine-tuning is a family of techniques (LoRA, adapters, prefix tuning, prompt tuning) that adapt a large pretrained model to a new task or domain by updating only a small fraction of its parameters - typically under 1% - while freezing the rest. This…

Participatory AI(also: Community-Centered AI, Participatory Machine Learning)

An approach to artificial intelligence development that actively involves the communities affected by AI systems in defining problems, setting priorities, designing solutions, collecting data, evaluating outcomes, and governing deployment. Participatory AI goes beyond…

Pattern Recognition

A branch of machine learning and artificial intelligence focused on identifying regularities, patterns, and structures in data such as images, sounds, or sensor readings. In accessibility, pattern recognition is fundamental to technologies like sign language recognition systems…

Pedestrian Detection(also: Person Detection, Human Detection)

A computer vision task that identifies and locates people in images or video frames, typically using deep learning models such as convolutional neural networks. In accessibility applications, pedestrian detection is used in wearable assistive technologies for blind and low…

Perceptual Gap

A design failure identified by Choudhury (2026) in which an AI system's explanation is delivered through exactly the sensory channel that its user cannot access. For example, a Grad-CAM heat map overlaid on an image tells a blind user where the model looked but cannot be seen by…

Perturbation testing(also: Counterfactual testing, Template-based testing)

A bias evaluation methodology for NLP models that systematically substitutes identity-related terms (e.g., disability phrases) in otherwise identical sentences to measure whether the model produces different predictions based on the identity mention alone. By holding all other…

Predictive AI(also: Predictive Analytics, Predictive Algorithm)

AI systems that use machine learning to identify patterns in data and anticipate future outcomes, behaviors, or events. In the context of disability, predictive AI systems have been documented causing significant harm by making decisions about resource allocation (welfare…

Proactive description(also: Proactive notification, Unsolicited description)

The ability of an assistive system to automatically provide relevant visual or environmental information without requiring the user to explicitly request it. For blind and visually impaired users navigating real-world environments, proactive description is critical — a human…

Prompt Chaining(also: Chained Prompting, Sequential Prompting)

A technique for interacting with large language models where multiple prompts are issued in sequence, with each prompt building on the output of the previous one to achieve a more refined or accurate result. In accessibility and bias mitigation contexts, prompt chaining enables…

Prompt Engineering(also: Prompt Design, Prompt Crafting)

The practice of designing and structuring input prompts to guide large language models (LLMs) toward producing more accurate, relevant, and useful outputs. In accessibility contexts, prompt engineering techniques such as role-play prompting (assigning expert personas),…

Provenance Indicator(also: Source Attribution)

Information that identifies which AI model, trial, or prompt produced a particular piece of content in a multi-model comparison. Provenance indicators help users understand which models generate which claims, enabling them to build mental models of individual model strengths and…

11 results.