Open Models Almanac
Methodstraining objective

specific method · filed under training objective

Winner-takes-all action-candidate supervision

Updates or includes only the action candidate with the lowest distance or loss in action supervision.

The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See training objective for the methods filed beside it.

Also called winner-takes-all scheme, winner-takes-all paradigm.

sources
2
model
1
lab adopt it
1
strongest
used

Documented in

Evidence

2 quotes from 2 sources, one card per model, each opening with its strongest.

How sources treat it
used 2

Xiaomi-Robotics-1

Xiaomiadopted2 sources

Action prediction is supervised via a winner-takes-all scheme: only the candidate with the lowest distance is updated via backpropagation.

usedtraining objectivein Xiaomi-Robotics-0Xiaomi Robotics
1 more from this model

We adopt a winner-takes-all paradigm as in [59], where only the candidate with the smallest loss is included in action loss computation

usedtraining objectivein Xiaomi-Robotics-1Xiaomi
↑