Delay-Tolerant Online Convex Optimization: Unified Analysis and Adaptive-Gradient Algorithms

Authors: Pooria Joulani, Andras Gyorgy, Csaba Szepesvari

AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical We present a unified, black-box-style method for developing and analyzing online convex optimization (OCO) algorithms for full-information online learning in delayed-feedback environments. Our new, simplified analysis enables us to substantially improve upon previous work and to solve a number of open problems from the literature. Our unified framework builds on a natural reduction from delayed-feedback to standard (non-delayed) online learning. This reduction, together with recent unification results for OCO algorithms, allows us to analyze the regret of generic FTRL and Mirror-Descent algorithms in the delayed-feedback setting in a unified manner using standard proof techniques.
Researcher Affiliation Academia Pooria Joulani1 Andr as Gy orgy2 Csaba Szepesv ari1 1Department of Computing Science, University of Alberta, Edmonton, AB, Canada {pooria,szepesva}@ualberta.ca 2Department of Electrical and Electronic Engineering, Imperial College London, UK a.gyorgy@imperial.ac.uk
Pseudocode Yes Algorithm 1 Single-instance Online Learning In Delayed environments (SOLID) Set x first prediction of BASE. for each time step t = 1, 2, . . . do Set xt x as the prediction for the current time step. Receive the set of feedbacks Ht that arrive at the end of time step t. for each fs Ht do Update BASE with fs. x the next prediction of BASE. end for end for
Open Source Code No The paper is theoretical and focuses on analysis and algorithm design. It does not mention providing open-source code for its methodology or any related implementations.
Open Datasets No The paper is theoretical and does not involve empirical experiments with specific datasets. It describes a 'sequential prediction game' and 'loss functions' as part of its theoretical problem setting, but does not refer to any concrete, publicly available dataset.
Dataset Splits No The paper is theoretical and does not involve empirical experiments, dataset evaluation, or code implementation for performance testing. Therefore, there are no training/validation/test splits mentioned.
Hardware Specification No The paper is theoretical and does not involve empirical experiments that would require specific hardware. No hardware specifications are mentioned.
Software Dependencies No The paper is theoretical and focuses on algorithm analysis and design. It does not mention any specific software dependencies or version numbers required to reproduce experiments, as no empirical experiments are conducted.
Experiment Setup No The paper is theoretical and does not involve empirical experiments. Consequently, there are no details provided about an experimental setup, hyperparameters, or system-level training settings.