Fourier Analysis-based Iterative Combinatorial Auctions

Authors: Jakob Weissteiner, Chris Wendler, Sven Seuken, Ben Lubin, Markus Püschel

IJCAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally show that our hybrid ICA achieves higher efficiency than prior auction designs, leads to a fairer distribution of social welfare, and significantly reduces runtime.
Researcher Affiliation Academia 1University of Zurich 2ETH Zurich 3Boston University 4 ETH AI Center weissteiner@ifi.uzh.ch, chris.wendler@inf.ethz.ch, seuken@ifi.uzh.ch, blubin@bu.edu, pueschel@inf.ethz.ch
Pseudocode Yes Algorithm 1: HYBRID ICA
Open Source Code Yes Our code is available on Git Hub: https://github.com/ marketdesignresearch/FA-based-ICAs.
Open Datasets Yes For our experiments, we use the spectrum auction test suite (SATS) [Weiss et al., 2017].
Dataset Splits No The paper states that parameters were optimized on a 'training set of CA instances' and results are averaged over a 'test set of CA instances', indicating a split of instances. However, it does not provide specific details on traditional train/validation/test dataset splits (e.g., percentages, sample counts, or specific files) for the data used within each instance for training the ML models, which are trained iteratively on elicited reports.
Hardware Specification Yes All experiments were conducted on machines with Intel Xeon E5 v4 2.20GHz processors with 24 cores and 128GB RAM or with Intel E5 v2 2.80GHz processors with 20 cores and 128GB RAM.
Software Dependencies Yes We used SATS version 0.6.4 for our experiments. The implementations of GSVM and LSVM have changed slightly in newer SATS versions.
Experiment Setup Yes Table 4: Best configuration of HYBRID ICA. R:[32, 32] | N:[10, 10] WHT 30 21 20 29; R:[32, 32] | N:[10, 10, 10] WHT 30 30 10 30; L,R,N:[16, 16] WHT 30 220 0 250