Drift-to-Action Controllers: Budgeted Interventions with Online Risk Certificates
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910046547345408 |
|---|---|
| author | Lamaakal, Ismail Yahyati, Chaymae Makkaoui, Khalid El Ouahbi, Ibrahim Maleh, Yassine |
| author_facet | Lamaakal, Ismail Yahyati, Chaymae Makkaoui, Khalid El Ouahbi, Ibrahim Maleh, Yassine |
| contents | Deployed machine learning systems face distribution drift, yet most monitoring pipelines stop at alarms and leave the response underspecified under labeling, compute, and latency constraints. We introduce Drift2Act, a drift-to-action controller that treats monitoring as constrained decision-making with explicit safety. Drift2Act combines a sensing layer that maps unlabeled monitoring signals to a belief over drift types with an active risk certificate that queries a small set of delayed labels from a recent window to produce an anytime-valid upper bound $U_t(δ)$ on current risk. The certificate gates operation: if $U_t(δ) \le τ$, the controller selects low-cost actions (e.g., recalibration or test-time adaptation); if $U_t(δ) > τ$, it activates abstain/handoff and escalates to rollback or retraining under cooldowns. In a realistic streaming protocol with label delay and explicit intervention costs, Drift2Act achieves near-zero safety violations and fast recovery at moderate cost on WILDS Camelyon17, DomainNet, and a controlled synthetic drift stream, outperforming alarm-only monitoring, adapt-always adaptation, schedule-based retraining, selective prediction alone, and an ablation without certification. Overall, online risk certification enables reliable drift response and reframes monitoring as decision-making with safety. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_08578 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Drift-to-Action Controllers: Budgeted Interventions with Online Risk Certificates Lamaakal, Ismail Yahyati, Chaymae Makkaoui, Khalid El Ouahbi, Ibrahim Maleh, Yassine Machine Learning Computation and Language Deployed machine learning systems face distribution drift, yet most monitoring pipelines stop at alarms and leave the response underspecified under labeling, compute, and latency constraints. We introduce Drift2Act, a drift-to-action controller that treats monitoring as constrained decision-making with explicit safety. Drift2Act combines a sensing layer that maps unlabeled monitoring signals to a belief over drift types with an active risk certificate that queries a small set of delayed labels from a recent window to produce an anytime-valid upper bound $U_t(δ)$ on current risk. The certificate gates operation: if $U_t(δ) \le τ$, the controller selects low-cost actions (e.g., recalibration or test-time adaptation); if $U_t(δ) > τ$, it activates abstain/handoff and escalates to rollback or retraining under cooldowns. In a realistic streaming protocol with label delay and explicit intervention costs, Drift2Act achieves near-zero safety violations and fast recovery at moderate cost on WILDS Camelyon17, DomainNet, and a controlled synthetic drift stream, outperforming alarm-only monitoring, adapt-always adaptation, schedule-based retraining, selective prediction alone, and an ablation without certification. Overall, online risk certification enables reliable drift response and reframes monitoring as decision-making with safety. |
| title | Drift-to-Action Controllers: Budgeted Interventions with Online Risk Certificates |
| topic | Machine Learning Computation and Language |
| url | https://arxiv.org/abs/2603.08578 |