Online and Stochastic Learning with a Human Cognitive Bias

Authors: Hidekazu Oiwa, Hiroshi Nakagawa

AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental results show the superiority of the derived algorithm for problems involving human cognition.
Researcher Affiliation Academia Hidekazu Oiwa The University of Tokyo and JSPS Research Fellow 7-3-1 Hongo, Bunkyo-ku, Tokyo, Japan hidekazu.oiwa@gmail.com Hiroshi Nakagawa The University of Tokyo 7-3-1 Hongo, Bunkyo-ku, Tokyo, Japan nakagawa@dl.itc.u-tokyo.ac.jp
Pseudocode No The paper states "The overall procedure of E-OGD is written in the supplementary material." but no pseudocode or algorithm block is present in the main text.
Open Source Code No The paper does not contain any concrete access information (specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described in this paper.
Open Datasets Yes We used five large-scale data sets from the LIBSVM binary data collections3." and "3http://www.csie.ntu.edu.tw/ cjlin/libsvmtools/datasets/ binary.html" and "We set up a synthetic scene recognition task as a binary classification problem using indoor recognition datasets2." and "2http://web.mit.edu/torralba/www/indoor.html
Dataset Splits No The paper mentions training and testing sets, for example, "we randomly sample 90% data from the dataset and used them as a training set and remaining data as a test set" for the webspam-t dataset. However, it does not explicitly describe specific training/test/validation dataset splits, such as exact percentages for all splits or sample counts, that would be needed for full reproducibility of data partitioning, nor does it explicitly mention a validation set in general experimental setup.
Hardware Specification No No specific hardware details (such as GPU/CPU models, processor types, or memory amounts) used for running the experiments were provided.
Software Dependencies No The paper does not provide specific ancillary software details, such as library names with version numbers, needed to replicate the experiment.
Experiment Setup Yes We used the logistic loss as loss functions. Each algorithm learned the weight vector from the training set through 1 iteration. Learning rates are t = / p t. We varied from 103 to 1.91 10 3 with common ratio 1/2 to obtain the appropriate step width for minimizing cumulative loss.