Value-at-Risk Optimization with Gaussian Processes

Authors: Quoc Phong Nguyen, Zhongxiang Dai, Bryan Kian Hsiang Low, Patrick Jaillet

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our V-UCB algorithm empirically demonstrates state-of-the-art performance in optimizing synthetic benchmark functions, a portfolio optimization problem, and a simulated robot task. The performance of our proposed algorithm is empirically demonstrated in optimizing several synthetic benchmark functions, a portfolio optimization problem, and a simulated robot task in Section 4.
Researcher Affiliation Academia 1Department of Computer Science, National University of Singapore, Republic of Singapore 2Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, USA.
Pseudocode Yes Algorithm 1 The V-UCB Algorithm
Open Source Code No The experiments using ρKGapx is performed by adding new objective functions to the existing implementation of Cakmak et al. (2020) at https://github.com/saitcakmak/Bo Risk. The paper does not provide a link to the code for its proposed V-UCB algorithm.
Open Datasets Yes The first problem is portfolio optimization adopted by (Cakmak et al., 2020). [...] The objective function is the posterior mean of a trained GP on the dataset in Cakmak et al. (2020) of size 3000 generated from CVXPortfolio.
Dataset Splits No The paper refers to using a 'trained GP on the dataset in Cakmak et al. (2020)' but does not provide specific details on how this dataset was split into training, validation, or test sets for their experiments.
Hardware Specification No The paper does not specify any particular hardware, such as GPU models, CPU types, or cloud resources, used for running the experiments.
Software Dependencies No The gradient of Vα(ut 1(x, Z)) with respect to x can be obtained easily (e.g., using automatic differentiation provided in the Tensorflow library (Abadi et al., 2015)). The paper mentions 'Tensorflow' and 'CVXPortfolio' but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes The noise variance σ2 n is set to 0.01. The risk level α is 0.1. In the implementation of LNSO, the local region is defined with r = 0.1. The surrogate function is a neural network of 2 hidden layers with 30 hidden neurons each and sigmoid activation functions.