dropout

Dropout is a term used in several fields to describe a person who leaves a formal program before completion and a regularization technique in machine learning.

In education, a dropout is a student who leaves school before earning a degree or diploma. Definitions vary by jurisdiction and level of study, with some relying on age thresholds and others on non-completion within a specified time. Dropout rates are influenced by factors such as socioeconomic status, family support, academic preparation, and school resources. The consequences can include lower lifetime earnings, reduced employment opportunities, and increased social costs. Efforts to reduce dropout include early warning systems, tutoring, financial aid, counseling, and offering alternative pathways such as GED programs or vocational training.

In higher education, dropout is often used interchangeably with attrition, referring to students who withdraw or fail to persist to degree completion. Persistence measures help institutions monitor progress and target interventions to support at-risk students.

In machine learning, dropout is a regularization technique designed to prevent overfitting. During training, neurons are randomly deactivated with probability p; at test time, all units are used with their outputs scaled to account for the training-time dropout. This encourages redundancy and robustness, allowing larger networks to be trained on limited data. Typical p values are around 0.5 for hidden layers, with smaller values for input layers. Variants include spatial dropout, DropConnect, and other noise-based methods.