Improved deep reinforcement learning for robotics through distribution-based experience retention

By Tim de Bruin, Jens Kober, Karl Tuyls, Robert Babuška. 2016

58 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Cognitive Neuroscience, Management Science and Operations Research

How researchers see this paper

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

In Artificial Intelligence

1. Sergey Levine
2. Nei Kato
3. Fengxiao Tang
4. Pieter Abbeel
5. Doina Precup
6. Scott Fujimoto
7. Bomin Mao
8. Ofir Nachum

In Cognitive Neuroscience

1. Chengtai Cao
2. Fan Zhou
3. Tim de Bruin
4. Doina Precup
5. Jens Kober
6. Scott Fujimoto
7. Robert Babuška
8. Karl Tuyls

In Management Science and Operations Research

1. Scott Fujimoto
2. Sergey Levine
3. Chengtai Cao
4. Fan Zhou
5. Doina Precup
6. Karl Tuyls
7. Robert Babuška
8. Tim de Bruin

Citations in the corpus

• Off-Policy Deep Reinforcement Learning without Exploration (2018, background): “It has been observed that large experience replay (Lin, 1992) can be detrimental to performance (de Bruin et al., 2015; Zhang & Sutton, 2017) and the diversity of states in the buffer is an important factor for performance (de Bruin et al., 2016).”
• State-of-the-Art Deep Learning: Evolving Machine Intelligence Toward Tomorrow’s Intelligent Network Traffic Control Systems (2017, methodology): “For example, deep neural networks as function approximators in reinforcement learning for physical systems like robots were used in [89].”
• Overcoming Catastrophic Forgetting in Graph Neural Networks with Experience Replay (2021, methodology): “Drawing inspiration from the prior work (de Bruin et al. 2016) on continual reinforcement learning, we hypothesize that approximating a uniform distribution over all nodes from the training set D tr i in each task T i can facilitate choosing experience nodes.”
• Continual Graph Learning (2020, unknown): “Drawing inspiration from [de Bruin et al. , 2016], we suggest that approximating a uniform distribution over all nodes from the training set D tr i in each task T i can facilitate choosing experience nodes.”
• Off-policy experience retention for deep actor-critic learning (2016, methodology): “This work is related to our earlier work [1] which attempts to influence the distribution of the samples in the experience database over the state-action space by changing the way the database is overwritten when it is full.”