Online Deep Learning: Learning Deep Neural Networks on the Fly

Authors: Doyen Sahoo, Quang Pham, Jing Lu, Steven C. H. Hoi

IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate the efficacy on large data sets (both stationary and concept drifting scenarios).4 Experiments 4.1 Datasets We consider several large scale datasets. Higgs and Susy are Physics datasets from UCI repository. For Higgs, we sampled 5 million instances. We used 5 million instances from Infinite MNIST [Loosli et al., 2007]. We also evaluated on 3 synthetic datasets.
Researcher Affiliation Collaboration Doyen Sahoo1, Quang Pham1, Jing Lu2, Steven C. H. Hoi1 1 School of Information Systems, Singapore Management Univerity, 2 JD.com {doyens,hqpham.2017}@smu.edu.sg, lvjing12@jd.com, chhoi@smu.edu.sg
Pseudocode Yes Algorithm 1 Online Deep Learning (ODL) using HBP
Open Source Code Yes Source code available at https://github.com/LIBOL/ODL
Open Datasets Yes Higgs and Susy are Physics datasets from UCI repository. For Higgs, we sampled 5 million instances. We used 5 million instances from Infinite MNIST [Loosli et al., 2007].
Dataset Splits No The paper describes evaluating performance in different 'windows' or 'stages' of the learning process (e.g., 'first 0.5% of data', '[10-15]%', '[60-80]%'), which is characteristic of online learning where data arrives sequentially. It does not specify explicit train/validation/test dataset splits as used in batch learning for reproduction.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No Implementation was in Keras [Chollet, 2015]. However, no specific version number for Keras or any other software dependencies is mentioned.
Experiment Setup Yes Configuration across all methods: Re LU activation, fixed learning rate of 0.01 (finetuned on the baselines). For momentum, a fixed learning rate of 0.001 was used, and momentum parameters were finetuned to give the best performance on the baselines. For HBP, we set β = 0.99 and the smoothing parameter s = 0.2.