FedRec++: Lossless Federated Recommendation with Explicit Feedback
Authors: Feng Liang, Weike Pan, Zhong Ming4224-4231
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive empirical studies clearly show the effectiveness of our Fed Rec++ in providing accurate and privacy-aware recommendation without much additional communication cost. We conduct experiments on three public datasets to study the effectiveness and efficiency of our Fed Rec++. |
| Researcher Affiliation | Academia | Feng Liang, Weike Pan*, Zhong Ming* National Engineering Laboratory for Big Data System Computing Technology College of Computer Science and Software Engineering Shenzhen University, Shenzhen 518060, China |
| Pseudocode | Yes | Algorithm 1 The algorithm of Fed Rec++ in the server. Algorithm 2 Client Training(Vi , i = 1, 2, . . . , m; OPERATION; u; u; U; t), i.e., the algorithm of Fed Rec++ in the client. |
| Open Source Code | No | The paper does not include an unambiguous statement that the authors are releasing the code for the work described, nor does it provide a direct link to a source-code repository. |
| Open Datasets | Yes | Besides using the two datasets in Fed Rec (Lin et al. 2020), i.e., Movie Lens 100K (ML100K) and Movie Lens 1M (ML1M), we also include a subset from Netflix (NF5K5K). |
| Dataset Splits | No | The paper states: 'we randomly divide the dataset into five parts with the same size; (ii) we take four parts as the training data, and the remaining one part as the test data; and (iii) we repeat the second step four times to get five different copies of training data and test data.' and 'All the hyper parameters are searched according to the MAE performance on the first copy of each dataset.'. However, it does not explicitly define a separate validation dataset split from the original dataset, beyond implicitly using part of the 'first copy' for tuning. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models, memory, or type of computing cluster) used for running its experiments. |
| Software Dependencies | No | The paper mentions using 'multi-thread programming in Java' but does not specify any version numbers for Java or any other software libraries or dependencies. |
| Experiment Setup | Yes | For parameter configurations, we mainly follow Fed Rec (Lin et al. 2020). In particular, we fix the number of latent features d = 20 and the number of iterations T = 100. We search the best value of the learning rate γ {0.7, 0.8, . . . , 1.4}, and have γ = 0.8, γ = 0.8 and γ = 1.0 on ML100K, ML1M and NF5K5K, respectively. We search the best value of the tradeoff parameter on the regularization terms α {0.1, 0.01, 0.001}, and have α = 0.001 on all the three datasets. We use different values of the sampling parameter ρ {0, 1, 2, 3}. We choose the best value of the iteration number Tpredict for starting filling the sampled unrated items via Eq.(1) and the iteration number Tlocal for locally training U u both from {5, 10, 15}... |