Stochastic and Adversarial Online Learning without Hyperparameters

Authors: Ashok Cutkosky, Kwabena A. Boahen

NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical Here we introduce an online optimization algorithm that achieves O(log4(T)) regret in a wide class of stochastic settings while gracefully degrading to the optimal O(T) regret in adversarial settings (up to logarithmic factors). Our algorithm does not require any prior knowledge about the data or tuning of parameters to achieve superior performance. In Section 3, we give explicit pseudo-code and prove our regret bounds for the adversarial setting. In Section 4, we formally define α-acute convexity and prove regret bounds for the acutely convex stochastic setting. Finally, in Section 5, we give some motivating examples of acutely convex stochastic losses.
Researcher Affiliation Academia Ashok Cutkosky Department of Computer Science Stanford University ashokc@cs.stanford.edu Kwabena Boahen Department of Bioengineering Stanford University boahen@stanford.edu
Pseudocode Yes Algorithm 1 FREEREXMOMENTUM Initialize: 1 η2 0 0, a0 0, w1 0, L0 0, ψ(w) = ( w + 1) log( w + 1) w for t = 1 to T do Play wt Receive subgradient gt ℓt(wt) Lt max(Lt 1, gt ). // Lt = maxt t gt 1 η2 t max 1 η2 t 1 + 2 gt 2, Lt g1:t . at max(at 1, 1/(Ltηt)2) Pt 1 t =1 gt wt 1+ g 1:t wt+1 argmin W h 5φ(at(w wt) atηt + g1:t w i ... Algorithm 2 Coordinate-Wise FREEREXMOMENTUM
Open Source Code No The paper does not contain any statements about releasing source code for the described methodology or links to a code repository.
Open Datasets No The paper is theoretical and does not describe experiments using datasets, thus no access information for a public dataset is provided.
Dataset Splits No The paper is theoretical and does not describe experiments using datasets, thus no specific dataset split information for validation is provided.
Hardware Specification No The paper describes theoretical work and does not mention any specific hardware used for running experiments.
Software Dependencies No The paper describes theoretical work and does not list any specific software dependencies with version numbers for replication.
Experiment Setup No The paper describes theoretical work and does not provide specific experimental setup details like hyperparameters or training configurations.