model monitoring
In plain wordsContinuously checking an AI model after it goes live — what data it receives, what it predicts and whether it is still right — so that problems are caught before they cause harm.
The continuous measurement of a deployed model's inputs, outputs, performance and operating health, with alerts when they depart from expectations. Monitoring covers operational metrics (latencylatency, n.: How long you wait between asking an AI system for something and getting the answer. Lower latency means a snappier, more responsive system., error rates, throughputthroughput, n.: How much work an AI system gets done per second across all its users at once — for a chatbot service, roughly how much text it can produce in total each second., cost); data-quality checks (missing values, schema violations, out-of-range values); data driftdata drift, n.: When the information flowing into an AI model in real use gradually stops resembling the data it was trained on — new customers, new slang, new cameras — making its predictions less reliable. in input distributions and prediction drift in outputs; and — where ground-truth labels arrive, often with a delay — accuracy and calibrationcalibration, n.: How well a model's stated confidence matches reality: when a well-calibrated system says it is 80% sure, it turns out to be right about 80% of the time., broken down by segment to expose degradation for particular groups. Drift is commonly scored with statistics such as the population stability index, Kolmogorov–Smirnov tests or divergencesKullback–Leibler divergence, n.: A score for the cost of believing the wrong odds: on average, how much more surprised you will be by real outcomes than someone who knew the true probabilities. It is zero only when your odds are exactly right. computed against a training or reference window. Alerts trigger investigation, rollback or retraining, and the records supply evidence for audits and regulators. For language-modellarge language model, n.: A computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code. applications, monitoring extends to traces, user feedback, automated quality grading and safety signals, usually under the name LLM observabilityLLM observability, n.: Recording and inspecting what an AI application actually did on each request — prompts sent, documents found, tools used, answers given, time and cost — so problems can be found and fixed..
Model monitoring flagged a jump in prediction drift on the day the partner app changed its signup form.
Origin
Ongoing monitoring has long been expected for statistical models in regulated finance; the US Federal Reserve's 2011 supervisory guidance on model risk management (SR 11-7) lists it as a core element. As machine learning moved into production in the late 2010s, startups and cloud platforms packaged drift and performance monitoring as products, often marketed as 'ML observability'.
Usage note
Detecting drift does not show that accuracy has fallen; that requires labels or reliable proxies. Dashboards nobody reads, or alerts that fire constantly, provide little protection.
Related
- data driftdata drift, n.: When the information flowing into an AI model in real use gradually stops resembling the data it was trained on — new customers, new slang, new cameras — making its predictions less reliable.
- concept driftconcept drift, n.: When the real-world link between what a model sees and the right answer changes over time — as when spammers change tactics — so lessons the model learned no longer hold.
- LLM observabilityLLM observability, n.: Recording and inspecting what an AI application actually did on each request — prompts sent, documents found, tools used, answers given, time and cost — so problems can be found and fixed.
- MLOpsMLOps, n.: The working habits and tools that keep machine-learning models dependable inside real products once the experiment is over: automated testing, release, monitoring and retraining; the factory and maintenance crew around a model, not the model itself.
- model deploymentmodel deployment, n.: Putting a trained AI model to work in a real product so that people or other software can use it — and doing so carefully enough that the change can be undone if something goes wrong.
- service-level objectiveservice-level objective, n.: A specific, measurable target for how well a service should perform — for example, 99.9 percent of requests answered within two seconds — used to decide when engineers must act.
- distribution shiftdistribution shift, n.: When the data an AI system meets in real use differs from the data it was trained on — new users, places, times or habits — often making it perform worse than tests suggested.
Mentioned in
anomaly detectionanomaly detection, n.: Automatically spotting the rare data points that don't fit the usual pattern, such as a fraudulent card payment, a machine about to fail or an intruder on a network., concept driftconcept drift, n.: When the real-world link between what a model sees and the right answer changes over time — as when spammers change tactics — so lessons the model learned no longer hold., data driftdata drift, n.: When the information flowing into an AI model in real use gradually stops resembling the data it was trained on — new customers, new slang, new cameras — making its predictions less reliable., machine learning engineermachine learning engineer, n.: An engineer who turns machine-learning ideas into working systems: preparing data, training and testing models, and running them reliably inside real products., MLOpsMLOps, n.: The working habits and tools that keep machine-learning models dependable inside real products once the experiment is over: automated testing, release, monitoring and retraining; the factory and maintenance crew around a model, not the model itself., model deploymentmodel deployment, n.: Putting a trained AI model to work in a real product so that people or other software can use it — and doing so carefully enough that the change can be undone if something goes wrong., canary deploymentcanary deployment, n.: Releasing a new model or software version to a small slice of users first and watching closely, so that any problem hits only a few people before the change reaches everyone., data flywheeldata flywheel, n.: A cycle in which a product's users generate data that improves its AI, the better AI attracts more users, and those users generate still more data., LLM observabilityLLM observability, n.: Recording and inspecting what an AI application actually did on each request — prompts sent, documents found, tools used, answers given, time and cost — so problems can be found and fixed., machine learning technical debtmachine learning technical debt, n.: The hidden future cost of quick shortcuts in machine-learning systems — tangled data dependencies, messy glue code, unchecked feedback loops — that later make them hard and risky to change., post-market monitoringpost-market monitoring, n.: The duty of an AI system's maker to keep watching how it performs after release, gathering evidence from real use so that problems are found and fixed., service-level objectiveservice-level objective, n.: A specific, measurable target for how well a service should perform — for example, 99.9 percent of requests answered within two seconds — used to decide when engineers must act., shadow deploymentshadow deployment, n.: Quietly running a new AI model alongside the old one on real requests, recording its answers without showing them to anyone, to check it behaves well before it takes over., training–serving skewtraining–serving skew, n.: A mismatch between how data is prepared when a model is trained and when it is used for real, which makes the live model quietly worse than its tests promised.