AdaO2B: Adaptive Online to Batch Conversion for Out-of-Distribution Generalization

Authors: Xiao Zhang, Sunhao Dai, Jun Xu, Yong Liu, Zhenhua Dong

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results have demonstrated that Ada O2B significantly outperforms state-of-the-art baselines on both synthetic and real-world recommendation datasets.
Researcher Affiliation Collaboration Xiao Zhang,1 Sunhao Dai,1 Jun Xu,1,* Yong Liu,1 Zhenhua Dong2 1 Gaoling School of Artificial Intelligence, Renmin University of China, Beijing, China 2 Huawei Noah s Ark Lab, Shenzhen, China EMAIL EMAIL
Pseudocode Yes Algorithm 1: Ada O2B
Open Source Code No The paper does not provide an explicit statement or link for open-source code for the methodology described. It mentions a dataset website.
Open Datasets Yes We used the Kuai Rec dataset3, which provides a fully observed user-item interaction matrix from the popular videosharing app Kuaishou. 3https://kuairec.com
Dataset Splits No For both synthetic and real-world datasets, we split them into two subsets for the online learning phase (as well as the O2B conversion phase) and the batch testing phase, respectively, denoted by OL-Data and BT-Data.
Hardware Specification No The paper does not explicitly describe the hardware used for its experiments, such as specific GPU/CPU models or memory details.
Software Dependencies No The paper mentions 'Adam (Kingma and Ba 2014) is used to conduct the optimization' but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes We trained Ada O2B based on the last 10 (i.e.,, K = 10) data buffers and history policies. We tuned the hyper-parameters as follows: the learning rate was tuned within the range of {1e 2, 1e 3, 1e 4, 1e 5}, the weight decay was tuned among {1e 3, 1e 4, 1e 5, 1e 6}, and the batch size was tuned in {256, 512, 1024, 2048}.