Contextual Combinatorial Multi-output GP Bandits with Group Constraints
Authors: Sepehr Elahi, Baran Atalar, Sevda Öğüt, Cem Tekin
TMLR 2023 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We lastly show in experiments using synthetic and real-world data and based on a federated learning setup as well as a content-recommendation one that our algorithm performs better then the current non-GP state-of-the-art combinatorial bandit algorithm, while satisfying group constraints. ... We perform two sets of experiments. In the first one, we use a synthetic dataset and a privacy-aware federated learning setup to show that our algorithm outperforms the non-GP state-of-the-art while maintaining user privacy requirements. In the second experiment, we consider a content caching-aware movie recommendation setup. |
| Researcher Affiliation | Academia | Sepehr Elahi EMAIL School of Computer and Communication Sciences EPFL Baran Atalar EMAIL Department of Electrical and Computer Engineering Carnegie Mellon University Sevda Öğüt EMAIL School of Computer and Communication Sciences EPFL Cem Tekin EMAIL Department of Electrical and Electronics Engineering Bilkent University |
| Pseudocode | Yes | Algorithm 1 TCGP-UCB 1: Input: X, K, M, ζ, δ, T; GP Prior: GP(µ0, k) 2: for t = 1, . . . , T do 3: Observe base arms in Mt, their contexts Xt, their groups Gt, and feasible super arms St 4: for xt,m : m Mt do 5: Calculate µJt 1K(xt,m) and σJt 1K(xt,m) using GP posterior 6: Compute indices it(xt,m) and i t(xt,m) 7: end for 8: Gt,good Oraclegrp(i t(xt,m)m Mt, Gt) 9: Form the set ˆS t using Gt,good and St 10: St Oraclespr(it(xt,m)m ˆ S t, ˆS t) 11: Observe base arm outcomes r, collect group rewards VG and super arm reward U 12: end for |
| Open Source Code | No | The paper does not contain an explicit statement about the release of source code nor does it provide a link to a code repository. |
| Open Datasets | Yes | We use the Movie Lens 25M dataset with a slight modification (Harper & Konstan, 2015). |
| Dataset Splits | No | The paper describes dynamic sampling and data generation methods for its experiments (e.g., "In each round, we simulate the number of available clients using a Poisson distribution with mean 50." and "In each round t, we randomly choose Ht movies from the available set of movies, where Ht is Poisson distributed with a mean of 75."), but it does not specify fixed training/validation/test splits for a static dataset. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models or other processor information used for running the experiments. |
| Software Dependencies | No | The paper mentions using 'sparse approximation to GPs' but does not specify any software libraries or tools with version numbers for implementation. It also mentions 'squared exponential kernels' and 'Matern kernels' but not the specific software/library versions used to implement them. |
| Experiment Setup | Yes | In our simulation, we set s = 10. We also set δ = 0.05, ζ = 0.5, and use two squared exponential kernels with both lengthscale and variance set to 1. ... In our simulation, we set the number of inducing points to s = 5. We also set δ = 0.05, ζ = 0.5, and use two Matern kernels with both lengthscale and variance set to 1. We set v1 = 1, v2 = 1, ρ = 0.5, and N = 2, as given in Definition 1 of (Nika et al., 2020). |