# MLDemon: Deployment Monitoring for Machine Learning Systems

arXiv: https://arxiv.org/abs/2104.13621 · HTML: https://mlsanity.com/paper/2104.13621/

## Key facts

- Cited by 14 papers in the map (data through 14 Oct 2026).
- Builds on 14 of its 66 references that are in the map.
- First posted on arXiv 28 Apr 2021.

## Builds on (most cited first, 14 of 14)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Learning under Concept Drift: A Review](https://mlsanity.com/paper/2004.05785/) (2020)
- [Detecting and Correcting for Label Shift with Black Box Predictors](https://mlsanity.com/paper/1802.03916/) (2018)
- [Failing Loudly: An Empirical Study of Methods for Detecting Dataset Shift](https://mlsanity.com/paper/1810.11953/) (2018)
- [A Unified View of Label Shift Estimation](https://mlsanity.com/paper/2003.07554/) (2020)
- [The Effect of Natural Distribution Shift on Question Answering Models](https://mlsanity.com/paper/2004.14444/) (2020)
- [Lifelong Machine Learning with Deep Streaming Linear Discriminant Analysis](https://mlsanity.com/paper/1909.01520/) (2019)
- [Deep Online Learning via Meta-Learning: Continual Adaptation for Model-Based RL](https://mlsanity.com/paper/1812.07671/) (2018)
- [Masked Face Recognition Dataset and Application](https://mlsanity.com/paper/2003.09093/) (2020)
- [Robust Machine Learning Systems: Challenges, Current Trends, Perspectives, and the Road Ahead](https://mlsanity.com/paper/2101.02559/) (2021)
- [FrugalML: How to Use ML Prediction APIs More Accurately and Cheaply](https://mlsanity.com/paper/2006.07512/) (2020)
- [Request-and-Reverify: Hierarchical Hypothesis Testing for Concept Drift Detection with Expensive Labels](https://mlsanity.com/paper/1806.10131/) (2018)
- [Automatic Model Monitoring for Data Streams](https://mlsanity.com/paper/1908.04240/) (2019)
- [Increasing Trustworthiness of Deep Neural Networks via Accuracy Monitoring](https://mlsanity.com/paper/2007.01472/) (2020)

## Built on by (most cited first, 14 of 14)

- [Calibrated Selective Classification](https://mlsanity.com/paper/2208.12084/) (2022)
- [GSCLIP : A Framework for Explaining Distribution Shifts in Natural Language](https://mlsanity.com/paper/2206.15007/) (2022)
- [Confidence-based Estimators for Predictive Performance in Model Monitoring](https://mlsanity.com/paper/2407.08649/) (2024)
- [Keeping Medical AI Healthy and Trustworthy: A Review of Detection and Correction Methods for System Degradation](https://mlsanity.com/paper/2506.17442/) (2025)
- [An investigation of challenges encountered when specifying training data and runtime monitors for safety critical ML applications](https://mlsanity.com/paper/2301.13476/) (2023)
- [Randomized Confidence Bounds for Stochastic Partial Monitoring](https://mlsanity.com/paper/2402.05002/) (2024)
- [MLHOps: Machine Learning for Healthcare Operations](https://mlsanity.com/paper/2305.02474/) (2023)
- [Adversarial Learning for Feature Shift Detection and Correction](https://mlsanity.com/paper/2312.04546/) (2023)
- [A Tutorial on Clinical Speech AI Development: From Data Collection to Model Validation](https://mlsanity.com/paper/2410.21640/) (2024)
- [Self-Supervised Data Generation for Precision Agriculture: Blending Simulated Environments with Real Imagery](https://mlsanity.com/paper/2502.18320/) (2025)
- [Active Learning Framework to Automate NetworkTraffic Classification](https://mlsanity.com/paper/2211.08399/) (2022)
- [An Efficient Framework for Monitoring Subgroup Performance of Machine Learning Systems](https://mlsanity.com/paper/2212.08312/) (2022)
- [Combining Embeddings and Domain Knowledge for Job Posting Duplicate Detection](https://mlsanity.com/paper/2406.06257/) (2024)
- [Reliably Detecting Model Failures in Deployment Without Labels](https://mlsanity.com/paper/2506.05047/) (2025)

## Abstract (arXiv)

Post-deployment monitoring of ML systems is critical for ensuring reliability, especially as new user inputs can differ from the training distribution. Here we propose a novel approach, MLDemon, for ML DEployment MONitoring. MLDemon integrates both unlabeled data and a small amount of on-demand labels to produce a real-time estimate of the ML model's current performance on a given data stream. Subject to budget constraints, MLDemon decides when to acquire additional, potentially costly, expert supervised labels to verify the model. On temporal datasets with diverse distribution drifts and models, MLDemon outperforms existing approaches. Moreover, we provide theoretical analysis to show that MLDemon is minimax rate optimal for a broad class of distribution drifts.