Faster Greedy MAP Inference for Determinantal Point Processes

Authors: Insu Han, Prabhanjan Kambadur, Kyoungsoo Park, Jinwoo Shin

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

Reproducibility Variable Result LLM Response
Research Type Experimental In our experiments, our algorithms are significantly faster than their competitors for large-scale instances, while sacrificing marginal accuracy.
Researcher Affiliation Collaboration 1School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Republic of Korea. 2Bloomberg LP, 731 Lexington Avenue, New York, NY, 10069.
Pseudocode Yes Algorithm 1 Faster Greedy DPP Inference
Open Source Code Yes The codes are available in https://github.com/insuhan/fastdppmap.
Open Datasets Yes Details of the primaries are provided in http://www.presidency.ucsb.edu/debates.php. We use 39 videos from a Youtube dataset (De Avila et al., 2011), and the trained DPP kernels from (Gong et al., 2014).
Dataset Splits No The paper describes the datasets and experimental parameters but does not explicitly provide details about training, validation, or test dataset splits (e.g., specific percentages or sample counts).
Hardware Specification Yes The experiments are performed using a machine with a hexa-core Intel CPU (Core i7-5930K, 3.5 GHz) and 32 GB RAM.
Software Dependencies No The paper mentions using NLTK for preprocessing but does not provide specific version numbers for NLTK or any other key software libraries or dependencies used in the experiments.
Experiment Setup Yes Unless stated otherwise, we choose parameters of p = 5, k = 10, s = 50, m = 20 and n = 15, regardless matrix dimension, for our algorithms. We also run CG until it achieves convergence error less than 10 10 and typically TCG 30.