CoMic: Complementary Task Learning & Mimicry for Reusable Skills

Authors: Leonard Hasenclever, Fabio Pardo, Raia Hadsell, Nicolas Heess, Josh Merel

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

Reproducibility Variable Result LLM Response
Research Type Experimental We systematically compare a variety of different network architectures across different data regimes both in terms of imitation performance as well as transfer to challenging locomotion tasks.
Researcher Affiliation Collaboration 1Deep Mind, London 2Imperial College, London, work done during an internship at Deep Mind.
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The acknowledgements mention 'Saran Tunyasuvunakool for infrastructure and opensourcing support' but there is no explicit statement or link confirming the release of the paper's source code.
Open Datasets Yes All motion capture data was obtained from the CMU mocap database. 1mocap.cs.cmu.edu
Dataset Splits No The paper describes using motion capture data for 'training' and then evaluating on 'transfer tasks' and 'out-of-sample tasks,' but it does not specify explicit training, validation, and test dataset splits with percentages or sample counts for the motion capture data used.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory amounts) used for running the experiments.
Software Dependencies No The paper mentions software like V-MPO, RHPO, and Mu Jo Co, but does not specify their version numbers or any other software dependencies with version information.
Experiment Setup Yes We regularize the latent embedding with a standard Gaussian prior by adding a KL loss term to the V-MPO losses: ... where the coefficient β controls the strength of the regularization. ... We trained MLP low-level controllers on the Locomotion clip set with latent space dimensions D = 20, 40 and 60 and a range of KL regularization strengths β.