Artificial Intelligence
Model Drift
Plain English
When a glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term gets worse over time.
Definition
glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term drift occurs when a model’s glossaryPerformancePerformance refers to how quickly and efficiently a system responds to user actions and processes tasks.Open glossary term declines over time due to changes in glossaryDataData is raw, uninterpreted information collected and stored so it can be analysed, processed, or used to inform decisions. On its own it carries no meaning; context and interpretation are what make it useful.Open glossary term or real-world conditions.
In practice
Monitored by comparing glossaryModel OutputModel output is the result or response generated by a model after processing input data.Open glossary term over time and retraining when glossaryPerformancePerformance refers to how quickly and efficiently a system responds to user actions and processes tasks.Open glossary term drops.
In context
It is silent. Nothing errors, nothing alerts, and the glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term simply gets slowly worse until someone notices the outputs stopped being useful months ago.
The reality
Drift is inevitable in dynamic glossaryEnvironmentAn environment is a specific configured setup where software runs, such as development, staging, or production, each with its own data and settings. Keeping them consistent is what makes behaviour in one predict behaviour in another.Open glossary term and requires ongoing monitoring and maintenance.
Compared with
Model Drift vs Data Drift
glossaryDataData is raw, uninterpreted information collected and stored so it can be analysed, processed, or used to inform decisions. On its own it carries no meaning; context and interpretation are what make it useful.Open glossary term drift is the input distribution changing. glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term drift is the resulting decline in glossaryPerformancePerformance refers to how quickly and efficiently a system responds to user actions and processes tasks.Open glossary term. Data drift is the cause you can monitor directly; model drift is the effect you usually notice late.
FAQ
Common questions
A few practical answers to the questions that usually come up around this term.
What is model drift?
It is when a glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term’s glossaryPerformancePerformance refers to how quickly and efficiently a system responds to user actions and processes tasks.Open glossary term declines due to changing glossaryDataData is raw, uninterpreted information collected and stored so it can be analysed, processed, or used to inform decisions. On its own it carries no meaning; context and interpretation are what make it useful.Open glossary term or conditions.
Why does model drift happen?
Because real-world glossaryDataData is raw, uninterpreted information collected and stored so it can be analysed, processed, or used to inform decisions. On its own it carries no meaning; context and interpretation are what make it useful.Open glossary term changes over time.
How do you detect model drift?
By monitoring glossaryPerformancePerformance refers to how quickly and efficiently a system responds to user actions and processes tasks.Open glossary term and comparing outputs over time.
How do you fix model drift?
By retraining or updating the glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term with new glossaryDataData is raw, uninterpreted information collected and stored so it can be analysed, processed, or used to inform decisions. On its own it carries no meaning; context and interpretation are what make it useful.Open glossary term.
Related Services
Related Guides
Related Terms
