Behavioral considerations suggest an average reward TD model of the dopamine system

36 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Towards Continual Reinforcement Learning: A Review and Perspectives (2020, positive): “It has been suggested that an average reward per step objective is more consistent with human studies than the discounted cumulative reward objective that has become more popular for RL research (Daw and Touretzky, 2000).”
Dopamine: generalization and bonuses (2002, background): “…& Dayan, 2002) are exploring opponent process models (Grossberg & Schmajuk, 1987; Solomon & Corbit, 1974) involving an putative opponent interaction between dopamine and the dorsal raphe serotonin system, in the computational context of average-case reinforcement learning (Daw & Touretzky, 2000).”
Opponent interactions between serotonin and dopamine (2002, background): “We start by considering long-term aspects, arguing that opponency emerges naturally (Daw & Touretzky, 2000, 2002) from TD learning in the case of predicting long-run average rewards rather than summed future rewards (Mahadevan, 1996; Puterman, 1994; Schwartz, 1993; Tadepalli & Ok, 1998).”