Separable value functions across time-scales
Joshua Romoff · Peter Henderson · Ahmed Touati · Yann Ollivier · Joelle Pineau · Emma Brunskill

Tue Jun 11th 04:00 -- 04:20 PM @ Room 104

In many finite horizon episodic reinforcement learning (RL) settings, it is desirable to optimize for the undiscounted return - in settings like Atari, for instance, the goal is to collect the most points while staying alive in the long run. Yet, it may be difficult (or even intractable) mathematically to learn with this target. As such, temporal discounting is often applied to optimize over a shorter effective planning horizon. This comes at the cost of potentially biasing the optimization target away from the undiscounted goal. In settings where this bias is unacceptable - where the system must optimize for longer horizons at higher discounts - the target of the value function approximator may increase in variance leading to difficulties in learning. We present an extension of temporal difference (TD) learning, which we call TD($\Delta$), that breaks down a value function into a series of components based on the differences between value functions with smaller discount factors. The separation of a longer horizon value function into these components has useful properties in scalability and performance. We discuss these properties and show theoretic and empirical improvements over standard TD learning in certain settings.

Author Information

Joshua Romoff (McGill University)
Peter Henderson (Stanford University)
Ahmed Touati (MILA / FAIR)
Yann Ollivier (Facebook Artificial Intelligence Research)
Joelle Pineau (McGill University / Facebook)
Emma Brunskill (Stanford University)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors