Surprise-Inspired Networking
summary
Argues that information value comes from surprisal-driven fusion of edge-generated “new” data with cloud “prior” models, and proposes pushing computation and especially storage into a layer near the edge within a roughly millisecond propagation bubble (aiming for 1–10 ms human-grade latency). Recommends concrete primitives: upstream aggregation / inverse CDN, declarative materialized views with incremental computation, and a training/compressed-model → inference pattern; cites AWS Wavelength, WebAssembly deployments, and CDN practices as practical starting points.
AI-written summary · about
chapters (12)
- 0:00Start
- 12:25Circuit Switched Telephony 1890-2020
- 14:41Packet Switching/Internet
- 17:29Tools of the Networking Trade
- 20:20Spoiler Alert
- 30:05Where Should We Leverage Surprise?
- 31:46Near the edge!
- 39:06Example: Federated Learning
- 41:15What's the Programming Model?
- 41:30What Plumbing is Required?
- 46:31Incremental computation
- 48:35Implications of Surprise for life beyond E2E and SG
