Reinforcement Learning
Robust On-Policy Sampling for Data-Efficient Policy Evaluation in Reinforcement Learning
On-policy algorithms learn about a particular target policy using data collected by behaving according to the target policy.
LAPO: Latent-Variable Advantage-Weighted Policy Optimization for Offline Reinforcement Learning Xi Chen
In practice, offline datasets are often heterogeneous, i.e., collected in a variety of scenarios, such as data from several human demonstrators or from
Explain My Surprise: Learning Efficient Long-Term Memory by Predicting Uncertain Outcomes
In many sequential tasks, a model needs to remember relevant events from the distant past to make correct predictions.
Learning Representations via a Robust Behavioral Metric for Deep Reinforcement Learning
As behavioral metrics are expensive or even intractable to computers, approximation or relaxation are necessary.
When to Trust Y our Simulator: Dynamics-Aware Hybrid Offline-and-Online Reinforcement Learning
H2O introduces a dynamics-aware policy evaluation scheme, which adaptively penalizes the Q-function learning on simulated state-action pairs with large dynamics gaps, while also simultaneously allowing learning from a fixed real-world dataset.
Object-Category Aware Reinforcement Learning Qi Yi
In OORL, the agent's observation is a set of object representations, and the task can be solved via reasoning based on these objects.