3 problems
- 0 votes0 replies0 views
Ye et al.'s log-linear learning conjecture for social convention dynamics
Ye et al.'s conjecture. Individuals revise their state according to the log-linear learning rule, with payoff functions that extend those of a coordination game to account for iner…
- 0 votes0 replies0 views
The natural-gradient reformulation conjecture for effective parameter-update sequences
Consider a sequence of parameter updates and a loss function whose value shows overall improvement along the sequence, without requiring monotone decrease at every update. A sequen…
- 0 votes0 replies0 views
The natural-gradient representation conjecture for effective learning rules
Let a sequence of parameter updates be given, and suppose that the updates improve a scalar measure of performance. A learning rule is in natural gradient form when its updates can…