Adversarial Prediction Games for Multivariate Losses

Authors: Hong Wang, Wei Xing, Kaiser Asif, Brian Ziebart

NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our approach, Multivariate Prediction Games (MPG), on the three performance measures of interest in this work: precision at k, F-score, and DCG. Our primary point of comparison is with structured support vector machines (SSVM)[27] to better understand the trade-offs between convexly approximating the loss function with the hinge loss versus adversarially approximating the training data using our approach. We employ an optical recognition of handwritten digits (OPTDIGITS) dataset [17] (10 classes, 64 features, 3,823 training examples, 1,797 test examples), an income prediction dataset ( a4a ADULT1 [17] (two classes, 123 features, 3,185 training examples, 29,376 test examples), and query-document pairs from the million query TREC 2007 (MQ2007) dataset of LETOR4.0 [23] (1700 queries, 41.15 documents on average per query, 46 features per document).
Researcher Affiliation Academia Hong Wang Wei Xing Kaiser Asif Brian D. Ziebart Department of Computer Science University of Illinois at Chicago Chicago, IL 60607 {hwang27, wxing3, kasif2, bziebart}@uic.edu
Pseudocode Yes Algorithm 1 Constraint generation game solver, Algorithm 2 Lagrangian-augmented F-measure Maximizer for adversary player ˇY
Open Source Code No The paper does not provide explicit statements or links indicating the availability of open-source code for the described methodology.
Open Datasets Yes We employ an optical recognition of handwritten digits (OPTDIGITS) dataset [17] (10 classes, 64 features, 3,823 training examples, 1,797 test examples), an income prediction dataset ( a4a ADULT1 [17] (two classes, 123 features, 3,185 training examples, 29,376 test examples), and query-document pairs from the million query TREC 2007 (MQ2007) dataset of LETOR4.0 [23] (1700 queries, 41.15 documents on average per query, 46 features per document). Footnote 1: http://www.csie.ntu.edu.tw/ cjlin/libsvmtools/datasets/binary.html
Dataset Splits Yes For OPTDIGITS/ADULT, we use a random 1/3 of the training data as a holdout validation data to select the L2 regularization parameter trade-off C {2^6, 2^5, ..., 2^6}. We compare the performance of our approach and comparison methods using five-fold cross validation on the MQ2007 dataset.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments.
Software Dependencies No The paper mentions some software like 'libsvmtools' and 'SVM-struct' but does not specify version numbers for any software dependencies used in its own experimental setup.
Experiment Setup Yes For OPTDIGITS/ADULT, we use a random 1/3 of the training data as a holdout validation data to select the L2 regularization parameter trade-off C {2^6, 2^5, ..., 2^6}.