Doctoral research, just starting
ARGENTIC
Every AI workload has to run somewhere. My PhD is about teaching a system to decide where, and to keep deciding as conditions change.
01 The question
Training, inference, moving data: each job could run close to the people using it, in the cloud, or on a supercomputer. Something has to choose, every time, while the conditions keep changing.
- At the edge
- In the cloud
- On a supercomputer
ARGENTIC is a Horizon Europe project on agentic AI across that whole range. My part is the decision: a scheduler that learns, with reinforcement learning, where each workload should go.
02 What it will weigh
Four things pulling in different directions
-
01
Latency
How long someone waits for the answer.
-
02
Energy
How much energy the job uses where it runs.
-
03
Data locality
Whether the data can stay where it already is.
-
04
Privacy
Whether the data is allowed to leave at all.
03 What comes next
In progress
More here soon
The work is starting now. Design notes, experiments and results will appear on this page as they happen.
- Design notes To follow
- First experiments To follow
- Results To follow