Batched Gaussian Process Bandit Optimization via Determinantal Point Processes

Authors: Tarun Kathuria, Amit Deshpande, Pushmeet Kohli

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments on a variety of synthetic and real-world robotics and hyper-parameter optimization tasks indicate that our DPP-based methods, especially those based on DPP sampling, outperform state-of-the-art methods.
Researcher Affiliation Industry Tarun Kathuria, Amit Deshpande, Pushmeet Kohli Microsoft Research t-takat@microsoft.com, amitdesh@microsoft.com, pkohli@microsoft.com
Pseudocode Yes Algorithm 1 GP-BUCB/B-EST Algorithm; Algorithm 2 GP-(UCB/EST)-DPP-(MAX/SAMPLE) Algorithm
Open Source Code No The paper mentions using publicly available code for baselines (BUCB, PE, EST, GPy Opt) but does not provide a link or statement for their own implementation of the DPP-based methods.
Open Datasets Yes The Abalone dataset is provided by the UCI Machine Learning Repository at http://archive.ics.uci.edu/ml/datasets/Abalone; Bibtex [15] and Delicious[32]
Dataset Splits No The paper mentions using synthetic functions and benchmark datasets (Abalone, Bibtex, Delicious) but does not provide specific details on how these datasets were split into training, validation, or test sets for their experiments.
Hardware Specification No The paper mentions general concepts like 'large parallel processing facilities' or 'multiple experimental setups' but does not specify any concrete hardware details (e.g., CPU/GPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions using publicly available code for baselines and implementing algorithms, but does not provide specific software dependencies with version numbers (e.g., Python version, library versions) for reproducibility.
Experiment Setup Yes We perform 50 experiments for each objective function and report the median of the immediate regret obtained for each algorithm. To maintain consistency, the first point of all methods is chosen to be the same (random). The mean function of the prior GP was the zero function while the kernel function was the squared-exponential kernel of the form k(x, y) = γ2 exp[ 0.5 P d(xd y2 d)/l2 d]. The hyper-parameter λ was picked from a broad Gaussian hyperprior and the the other hyper-parameters were chosen from uninformative Gamma priors.