Machine Learning, Networks, and IoT
summary
Three short papers: one frames image offloading as a DQN decision problem that learns correlations, enforces per-camera token buckets and shows gains on synthetic sequences and a Raspberry Pi testbed with DQN with token bucket (rate 0.1, bucket=4) and minimal runtime overhead versus MDP and heuristics. Another gives an online reroute-and-reschedule method for time‑triggered TSN flows using fixed‑point search and redundancy to recover deadlines after link failures. A verification paper trains a sensitivity net to drive guided falsification with provable sensitivity‑guided exploration (linear convergence). Finally Tinkertoy: C++17 functors, 4× smaller kernels composes kernel modules for ARM Cortex‑M3 with <200 lines to assemble app-specific OSes.
AI-written summary · about
chapters (10)
- 0:00Start
- 3:34Key Challenges for Developing a Solution
- 6:35Mdp Policy
- 10:04Summary
- 30:46Reachability Analysis
- 38:44Convergence Theorem
- 43:55Introduction
- 47:43Scheduler
- 48:06Event Handler
- 57:03System Programming Languages
related talks
57mMachine Learning in the Age of IoT
59mML and the IoT: Living on the Edge
48mLeveraging the Power of C++ for Efficient Machine Learning on Embedded Devices
21mThings I learned running neural networks on microcontrollers
39mLeveraging the power of C++ for efficient machine learning on embedded devices
01h 01mML and the IoT: Living on the Edge
12mPreview: ML Serving
49mML: Teaching machine to learn from its own mistakes (...)