10 cards. Cooperating with partners you did not train with.
Every card shows its question first. Reveal one at a time to test yourself, or
reveal all of them to use the deck as a revision sheet.
10 cards
Concept
What is partner dependence?
Concept
Partner dependence occurs when an agent performs well because it has specialized to the behaviour or conventions of particular training partners.
Scenario
Agent A achieves high reward with B but fails when paired with C. What might this reveal?
Scenario
Partner dependence.
High reward may reflect specialization rather than general cooperation, so the agent fails when paired with a different partner.
Distinction
What is the difference between learning general teamwork and learning one partner's habits?
Distinction
General teamwork transfers across different partners and behaviours.
Learning one partner's habits can produce strong familiar-partner performance but poor generalization.
Concept
Why can training with diverse partners improve partner generalization?
Concept
Different partners expose the agent to varied behaviours and conventions.
This reduces the opportunity to rely on one fixed partner strategy.
Concept
What is agent modelling, and why can it help adaptation?
Concept
Agent modelling means inferring useful properties of another agent from its behaviour or interaction history.
The inferred information can help the agent adjust its own actions.
Equation
What does this represent?
p^t=P(FETCH∣ht)
Equation
The estimated probability that the partner will choose FETCH, given the interaction history ht.
- ht
- everything observed about the partner so far
- p^t
- the current belief, updated as evidence arrives
Intuition
How does the Observe, Infer, Adjust, Cooperate loop describe adaptation?
Intuition
The agent:
- observes the partner's behaviour;
- infers something about the partner;
- adjusts its own strategy;
- uses that adjustment to cooperate more effectively.
Concept
What is ad hoc teamwork?
Concept
Ad hoc teamwork studies how an agent can cooperate with partners without relying on prior coordination with those specific agents.
Concept
What is zero-shot coordination?
Concept
Zero-shot coordination evaluates whether agents that were not trained together can cooperate successfully when paired for the first time.
Equation
What do cross-play and the generalization gap help us understand?
Δgen=Jfamiliar−Junseen
Equation
Cross-play evaluates how policies perform with different partners, especially unfamiliar pairings.
The gap compares familiar-partner and unseen-partner performance. A larger gap suggests stronger dependence on familiar partners or conventions.
Δgen is a teaching diagnostic used in this resource, not a canonical MARL metric.
No card in this deck matches that search.
Back to all decks ·
Read the Adapt material