Keynote: Tiny Data Collectors, Vastly (...) & the Land of Tiny Challenges
summary
Claims a concrete path offload cloud work by running ML and preprocessing on constrained devices: a desktop edge schedules robot-car tasks by battery CPU and RAM to trade accuracy for runtime; a Raspberry Pi micro-cluster shows measurable latency gains for local image/sound workloads; tiny lamppost-class processors (smaller than a Pi) share road warnings; statistical on-device filters (TIOT paper) avoid heavy ML; an Elixir federated framework aggregates models by environment conditions rather than one global model, but startup data sparsity and power limits remain unsolved.
AI-written summary · about
chapters (9)
- 0:0006:12 - Tiny Data collectors
- 6:1311:33 - On the edge
- 11:3416:18 - Vastly distributed systems
- 16:1920:47 - Federated Learning
- 20:4823:24 - Federated Learning Frameworks
- 23:2528:37 - The Land of Tiny Challenges
- 28:3833:41 - Erlang/ Elixir ecosystem
- 33:4237:21 - The Land of Opportunities
- 37:2242:54 - QnA
related talks
57mKeynote: The Physics of Thought and the Architecture of Intelligence
12mLightning Talk: Distributed AI Without the Infrastructure Tax
26mExo: Run your own AI cluster at home
16m[Opening Keynote] Making Distributed Computing Easy
01h 15mKeynote: Feeling the Local-First Elephant: A Roadmap, Hidden Gems, and New Puzzles...
50mBig Data on Smallish Machines
16mPreview: Cloud Computing
38mMPI Meets Machine Learning: Unlocking PyTorch distributed for scaling AI workloads