Edge-to-Cloud Swarm Coordination for precision oncology clinical workflows with embodied agent feedback loops

Описание и применение

Система распределённого выполнения задач на границе, облаке и локальном кластере использует агенты с локальными наблюдениями для быстрой и безопасной обработки больших медицинских изображений.

Задача

Centralized cloud inference cannot meet the latency, privacy, heterogeneity, and burstiness constraints of precision oncology workflows, such as whole‑slide image analysis and real‑time radiomics.

Результат

The embodied‑feedback swarm achieved nearly three times faster policy convergence than a centralized scheduler in a robotics simulator, and the architecture scales to real oncology workloads by respecting latency, privacy, heterogeneity, and burstiness constraints.

Рабочий процесс

• 1. A task (e.g., tile a WSI region) is announced to the broker.
• 2. Each embodied agent in Tier 1 gathers its local state (compute headroom, thermal headroom, RTT to Tier 2, privacy tier, clinical urgency).
• 3. Agents compute a bid using the provided weighted formula.
• 4. The broker awards the task to the lowest bidder.
• 5. All agents observe the outcome of tasks they did not win, providing counterfactual feedback.
• 6. Agents update their bid policy (actor‑critic loop) based on observed outcomes.
• 7. The process repeats for subsequent tasks, dynamically balancing load across tiers.

Доказательства

When I gave each agent a small "embodied" feedback signal — a local observation of its own compute, thermal, and memory state — the coordination policy converged nearly three times faster than a purely centralized scheduler.

Источник

A single whole-slide image (WSI) from a modern digital pathology scanner can exceed 100,000 × 100,000 pixels — several gigabytes of data.

Источник

HIPAA, GDPR, and institutional IRBs often prohibit raw patient data from leaving the premises.

Источник