Constrained Preferential Bayesian Optimization and Its Application in Banner Ad Design
Authors: Koki Iwai, Yusuke Kumagae, Yuki Koyama, Masahiro Hamasaki, Masataka Goto
IJCAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our technical evaluation shows that our CPBO method successfully identifies optimal solutions by focusing on exploring feasible regions. As a practical application, we also present a designer-in-the-loop system for banner ad design using CPBO... We conducted a user study with professional ad designers, demonstrating the potential benefits of our approach in guiding creative design under real-world constraints. |
| Researcher Affiliation | Collaboration | Koki Iwai1 , Yusuke Kumagae1 , Yuki Koyama2 , Masahiro Hamasaki2 and Masataka Goto2 1Hakuhodo DY Holdings Inc. 2National Institute of Advanced Industrial Science and Technology (AIST) EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1 CPBO with pairwise comparison 1: H 2: for n = 1, 2, . . . do 3: select x(i) n , x(j) n by optimizing EUBOC: x(i) n , x(j) n = arg max x(i), x(j) X EUBOC(x(i), x(j)) 4: Ask the evaluator to compare x(i) n and x(j) n 5: Evaluate the constraint function to obtain c(i) n and c(j) n 6: H H {(x(selected) n x(not selected) n ), (c(i) n , c(j) n )} 7: Update surrogate models f and c based on H 8: end for |
| Open Source Code | No | We implemented our CPBO technique on Bo Torch [Balandat et al., 2020], a BO library. We used Gaussian process models in Bo Torch as the surrogate models for the objective and constraint functions. The paper mentions using third-party libraries (Bo Torch, XGBoost) but does not state that the authors' own implementation of CPBO or their specific framework code is being released, nor does it provide a link. |
| Open Datasets | No | We employed two synthetic problems for evaluation. The first comes from Gardner et al. [2014]. It consists of 2-dimensional (2D) objective and constraint functions composed of sine and cosine functions. The second comes from Letham et al. [2019]... The CTR prediction model was built using XGBoost [Chen et al., 2016], trained on our dataset of real-world banner ad deployment data. The paper refers to synthetic problems from cited works (which describe function definitions, not specific datasets) and mentions using 'our dataset' for CTR prediction without providing any access details. |
| Dataset Splits | No | In this setting, we simulated both human responses (pairwise comparison) and constraint queries (direct observation of the constraint function values) using known synthetic test functions. The paper uses synthetic test functions and a proprietary dataset. For the synthetic functions, it describes simulation settings rather than fixed dataset splits. For the 'real-world banner ad deployment data', it states 'our dataset' without any information about how it was split into training, validation, or test sets. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware used for running the experiments, such as GPU or CPU models, memory, or cloud computing specifications. It only mentions that 'CTR prediction takes approximately 0.3 milliseconds per image' which implies computation but no hardware specifics. |
| Software Dependencies | No | We implemented our CPBO technique on Bo Torch [Balandat et al., 2020], a BO library... The CTR prediction model was built using XGBoost [Chen et al., 2016]. The paper mentions the use of Bo Torch and XGBoost but does not provide specific version numbers for these software libraries. |
| Experiment Setup | Yes | The number of iterations was set to 50. We ran each method 50 times with random initializations and recorded the mean and standard deviation of the results of all runs. We pre-trained the constraint surrogate model using 200 points randomly sampled from the search space. We set the threshold λ (i.e., the minimum acceptable predicted CTR) to the average of CTR values obtained from 1,000 randomly sampled parameters. |