Modern deep reinforcement learning methods have departed from the incremental learning required for eligibility traces, rendering the implementation of the λ-return difficult in this context. In particular, off-policy methods that utilize experience replay remain problematic because their random sampling of minibatches is not conducive to the efficient calculation of λ-returns. Yet replay-based methods are often the most sample efficient, and incorporating λ-returns into them is a viable way to achieve new state-of-the-art performance. Towards this, we propose the first method to enable practical use of λ-returns in arbitrary replay-based methods without relying on other forms of decorrelation such as asynchronous gradient updates. By promoting short sequences of past transitions into a small cache within the replay memory, adjacent λ-returns can be efficiently precomputed by sharing Q-values. Computation is not wasted on experiences that are never sampled, and stored λ-returns behave as stable temporal-difference (TD) targets that replace the target network. Additionally, our method grants the unique ability to observe TD errors prior to sampling; for the first time, transitions can be prioritized by their true significance rather than by a proxy to it. Furthermore, we propose the novel use of the TD error to dynamically select λ-values that facilitate faster learning. We show that these innovations can enhance the performance of DQN when playing Atari 2600 games, even under partial observability. While our work specifically focuses on λ-returns, these ideas are applicable to any multi-step return estimator.
more »
« less
Improving Computational Efficiency in Visual Reinforcement Learning via Stored Embeddings
Recent advances in off-policy deep reinforcement learning (RL) have led to impressive success in complex tasks from visual observations. Experience replay
improves sample-efficiency by reusing experiences from the past, and convolutional
neural networks (CNNs) process high-dimensional inputs effectively. However,
such techniques demand high memory and computational bandwidth. In this paper,
we present Stored Embeddings for Efficient Reinforcement Learning (SEER), a
simple modification of existing off-policy RL methods, to address these computational and memory requirements. To reduce the computational overhead of gradient
updates in CNNs, we freeze the lower layers of CNN encoders early in training
due to early convergence of their parameters. Additionally, we reduce memory
requirements by storing the low-dimensional latent vectors for experience replay
instead of high-dimensional images, enabling an adaptive increase in the replay
buffer capacity, a useful technique in constrained-memory settings. In our experiments, we show that SEER does not degrade the performance of RL agents while
significantly saving computation and memory across a diverse set of DeepMind
Control environments and Atari games. Finally, we show that SEER is useful for
computation-efficient transfer learning in RL because lower layers of CNNs extract
generalizable features, which can be used for different tasks and domains.
more »
« less
- Award ID(s):
- 2024675
- NSF-PAR ID:
- 10300405
- Date Published:
- Journal Name:
- Advances in neural information processing systems
- ISSN:
- 1049-5258
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
null (Ed.)Off-policy deep reinforcement learning (RL) has been successful in a range of challenging domains. However, standard off-policy RL algorithms can suffer from several issues, such as instability in Qlearning and balancing exploration and exploitation. To mitigate these issues, we present SUNRISE, a simple unified ensemble method, which is compatible with various off-policy RL algorithms. SUNRISE integrates two key ingredients: (a) ensemble-based weighted Bellman backups, which re-weight target Q-values based on uncertainty estimates from a Q-ensemble, and (b) an inference method that selects actions using the highest upper-confidence bounds for efficient exploration. By enforcing the diversity between agents using Bootstrap with random initialization, we show that these different ideas are largely orthogonal and can be fruitfully integrated, together further improving the performance of existing off-policy RL algorithms, such as Soft Actor-Critic and Rainbow DQN, for both continuous and discrete control tasks on both low-dimensional and high-dimensional environments.more » « less
-
null (Ed.)Conveying complex objectives to reinforcement learning (RL) agents can often be difficult, involving meticulous design of reward functions that are sufficiently informative yet easy enough to provide. Human-in-the-loop RL methods allow practitioners to instead interactively teach agents through tailored feedback; however, such approaches have been challenging to scale since human feedback is very expensive. In this work, we aim to make this process more sample- and feedback-efficient. We present an off-policy, interactive RL algorithm that capitalizes on the strengths of both feedback and off-policy learning. Specifically, we learn a reward model by actively querying a teacher’s preferences between two clips of behavior and use it to train an agent. To enable off-policy learning, we relabel all the agent’s past experience when its reward model changes. We additionally show that pre-training our agents with unsupervised exploration substantially increases the mileage of its queries. We demonstrate that our approach is capable of learning tasks of higher complexity than previously considered by human-in-the-loop methods, including a variety of locomotion and robotic manipulation skills. We also show that our method is able to utilize real-time human feedback to effectively prevent reward exploitation and learn new behaviors that are difficult to specify with standard reward functions.more » « less
-
null (Ed.)Conveying complex objectives to reinforcement learning (RL) agents can often be difficult, involving meticulous design of reward functions that are sufficiently informative yet easy enough to provide. Human-in-the-loop RL methods allow practitioners to instead interactively teach agents through tailored feedback; however, such approaches have been challenging to scale since human feedback is very expensive. In this work, we aim to make this process more sample- and feedback-efficient. We present an off-policy, interactive RL algorithm that capitalizes on the strengths of both feedback and off-policy learning. Specifically, we learn a reward model by actively querying a teacher’s preferences between two clips of behavior and use it to train an agent. To enable off-policy learning, we relabel all the agent’s past experience when its reward model changes. We additionally show that pre-training our agents with unsupervised exploration substantially increases the mileage of its queries. We demonstrate that our approach is capable of learning tasks of higher complexity than previously considered by human-in-the-loop methods, including a variety of locomotion and robotic manipulation skills. We also show that our method is able to utilize real-time human feedback to effectively prevent reward exploitation and learn new behaviors that are difficult to specify with standard reward functions.more » « less
-
Human learning and decision-making are supported by multiple systems operating in parallel. Recent studies isolating the contributions of reinforcement learning (RL) and working memory (WM) have revealed a trade-off between the two. An interactive WM/RL computational model predicts that although high WM load slows behavioral acquisition, it also induces larger prediction errors in the RL system that enhance robustness and retention of learned behaviors. Here, we tested this account by parametrically manipulating WM load during RL in conjunction with EEG in both male and female participants and administered two surprise memory tests. We further leveraged single-trial decoding of EEG signatures of RL and WM to determine whether their interaction predicted robust retention. Consistent with the model, behavioral learning was slower for associations acquired under higher load but showed parametrically improved future retention. This paradoxical result was mirrored by EEG indices of RL, which were strengthened under higher WM loads and predictive of more robust future behavioral retention of learned stimulus–response contingencies. We further tested whether stress alters the ability to shift between the two systems strategically to maximize immediate learning versus retention of information and found that induced stress had only a limited effect on this trade-off. The present results offer a deeper understanding of the cooperative interaction between WM and RL and show that relying on WM can benefit the rapid acquisition of choice behavior during learning but impairs retention. SIGNIFICANCE STATEMENT Successful learning is achieved by the joint contribution of the dopaminergic RL system and WM. The cooperative WM/RL model was productive in improving our understanding of the interplay between the two systems during learning, demonstrating that reliance on RL computations is modulated by WM load. However, the role of WM/RL systems in the retention of learned stimulus–response associations remained unestablished. Our results show that increased neural signatures of learning, indicative of greater RL computation, under high WM load also predicted better stimulus–response retention. This result supports a trade-off between the two systems, where degraded WM increases RL processing, which improves retention. Notably, we show that this cooperative interplay remains largely unaffected by acute stress.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
