A Time Leap Challenge for SAT Solving

Fichte, Johannes Klaus; Hecher, Markus; Szeider, Stefan

doi:10.48550/arxiv.2008.02215

Cited by 1 publication

(1 citation statement)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Grace (2013) found that hardware improvements account for half of the progress for several problems in the vicinity of AI research like computer chess, computer go and physics simulations. More recently, Fichte et al (2020) observed the same for progress in SAT solvers. In the realm of deep learning, Hernandez & Brown (2020) investigated improvements in algorithmic efficiency by holding final performance on various benchmarks constant and analyzing how the number of FLOPS needed to reach that level changed over time.…”

Section: Algorithmic Progress and Increased Efficiencymentioning

confidence: 60%

Measuring Progress in Deep Reinforcement Learning Sample Efficiency

Dorner¹

2021

Preprint

View full text Add to dashboard Cite

Sampled environment transitions are a critical input to deep reinforcement learning (DRL) algorithms. Current DRL benchmarks often allow for the cheap and easy generation of large amounts of samples such that perceived progress in DRL does not necessarily correspond to improved sample efficiency. As simulating real world processes is often prohibitively hard and collecting real world experience is costly, sample efficiency is an important indicator for economically relevant applications of DRL. We investigate progress in sample efficiency on Atari games and continuous control tasks by comparing the number of samples that a variety of algorithms need to reach a given performance level according to training curves in the corresponding publications. We find exponential progress in sample efficiency with estimated doubling times of around 10 to 18 months on Atari, 5 to 24 months on state-based continuous control and of around 4 to 9 months on pixel-based continuous control depending on the specific task and performance level.

show abstract

Section: Algorithmic Progress and Increased Efficiencymentioning

confidence: 60%

Measuring Progress in Deep Reinforcement Learning Sample Efficiency

Dorner¹

2021

Preprint

View full text Add to dashboard Cite

show abstract

A Time Leap Challenge for SAT Solving

Cited by 1 publication

References 29 publications

Measuring Progress in Deep Reinforcement Learning Sample Efficiency

Measuring Progress in Deep Reinforcement Learning Sample Efficiency

Contact Info

Product

Resources

About