Boosting for Control of Dynamical Systems

Authors: Naman Agarwal, Nataly Brukhim, Elad Hazan, Zhou Lu

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical evaluation on a host of control settings supports our theoretical findings.
Researcher Affiliation Collaboration 1Google AI Princeton 2Department of Computer Science, Princeton University. Correspondence to: <namanagarwal@google.com, {nbrukhim,ehazan,zhoul}@princeton.edu>.
Pseudocode Yes Algorithm 1 Dyna Boost 1; Algorithm 2 Dyna Boost 2
Open Source Code No The paper does not provide an explicit statement or link for open-source code for the methodology described.
Open Datasets No The paper describes experiments within simulated environments (Linear Dynamical Systems, Inverted Pendulum, etc.) where data is generated by the simulation rather than from a pre-existing publicly available dataset. While some environments like OpenAI Gym are open, they are environments for simulation and not datasets in the typical sense that require specific access information.
Dataset Splits No The paper conducts experiments within simulated environments and does not mention specific training, validation, and test dataset splits in terms of percentages or sample counts for model training.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) used for running its experiments.
Software Dependencies No The paper mentions using an "LSTM architecture" but does not specify version numbers for any software libraries, frameworks, or dependencies used in the experiments.
Experiment Setup Yes The GPC weak-controller is designed as in Equation 8, following (Agarwal et al., 2019), with the pre-fixed matrix K set to 0. The RNN weak-controller, using an LSTM architecture, with 5 hidden units. We set the memory length to H = 5, and use N = 5 weak-learners in all the experiments. The cost function used in all settings is c(x, u) = x 2 2 + u 2 2. each noise term wt is normally i.i.d. distributed with zero mean, and 0.12 variance. wt+1 N(wt, 0.32). wt = sin(t)/2π. wt N(wt 1, 5e-3), where the noise values are then clipped to the range [ 0.5, 0.5].