GPUSGD: A GPU‐accelerated stochastic gradient descent algorithm for matrix factorization

Jin, Jing; Lai, Siyan; Su, Hu; Lin, Jing; Lin, Xiaola

doi:10.1002/cpe.3722

Cited by 13 publications

(6 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Compared with previous methods mentioned in the previous section, our method is the fastest one, as it does not require any time for shuffling dataset and/or sorting portions of the dataset before processing it. BSGD processed ratings from first 50,000 users on the first 2,000 movies of Netflix data [19,20]. We used the same dataset to compare the performance of ESGD with BSGD.…”

Section: Experiments and Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Novel GPU-Based Approach for Matrix Factorization using Stochastic Gradient Descent

Nassar¹,

El-Sayed²,

Taha³

2017

IJICR

View full text Add to dashboard Cite

Recommender systems are crucial tools used in most of daily-based web applications. Matrix factorization is an advanced and efficient technique for recommender systems. Recently, Stochastic Gradient Descent (SGD) method is considered to be one of the most popular techniques for matrix factorization. SGD is a sequential algorithm, which is difficult to be parallelized for largescale problems. Nowadays, researches focus on efficiently parallelizing SGD. In this research, we propose a novel GPU approach for parallelizing SGD. The proposed approach is more efficient than recent parallel approaches because it utilizes GPU, reducing non-coalesced access of global memory and achieving load balance of threads. In addition, it does not require any sorting and/or data shuffling as preprocessing phase. Although platform used for implementation of the proposed approach is old, the implementation demonstrates 12.5x speedup over state-of-the-art GPU method, BSGD.

show abstract

Section: Experiments and Resultsmentioning

confidence: 99%

“…First, pipelining the data transfer from/to GPU and kernel execution. Second, using GPU clustering to parallelize the execution [20].…”

Section: Discussionmentioning

confidence: 99%

Novel GPU-Based Approach for Matrix Factorization using Stochastic Gradient Descent

Nassar¹,

El-Sayed²,

Taha³

2017

IJICR

View full text Add to dashboard Cite

show abstract

“…(b) Decompose M into U and I, where U is the user feature matrix and I is the item feature matrix. (c) According to (11), iteratively update U and I to solve the optimal value. Until the stop condition is met, the optimal value is reached.…”

Section: A Improved Funksvd Algorithmmentioning

confidence: 99%

“…Kang [10] runs the non-negative matrix factorization algorithm on the GPU platform, which accelerates the running speed of the algorithm and improves the effectiveness of the algorithm. Jin [11] proposed an efficient GPU algorithm to solve the matrix decomposition problem based on Stochastic Gradient Descent (SGD) method and improve the efficiency of the algorithm.…”

Section: Introductionmentioning

confidence: 99%

Parallel Algorithm of Improved FunkSVD Based on GPU

Xiaochen

Liu

2022

IEEE Access

View full text Add to dashboard Cite

Nowadays, the increasing amount of network data information and the development of big data technology have brought development opportunities and challenges to the recommendation system. The model-based collaborative filtering algorithm has become one of the mainstream algorithms in the recommendation system. For example, the Funk Singular Value Decomposition (FunkSVD) algorithm. However, in the face of big data calculations, data sparseness and iterative oscillations often affect the accuracy of the FunkSVD algorithm. Moreover, when the data volume is in units of GB or more, the FunkSVD algorithm runs slowly and is not effective. Therefore, we propose an improved FunkSVD algorithm (IFABG) based on RMSProp (Root Mean Square Prop) and GPU (Graphics Processing Unit) to solve this problem. Firstly, we use RMSProp algorithm to improve the traditional FunkSVD algorithm, alleviate data sparseness and iterative shock, and improve the prediction accuracy of the algorithm. Next, we implemented the parallelization of the improved FunkSVD algorithm in the GPU, which increased the calculation speed of the algorithm. Finally, we verify the IFABG algorithm under the Movielens dataset. The experimental results show that the IFABG algorithm is very suitable for processing sparse data, and it alleviates the iterative shock, and the prediction accuracy rate is about 30% higher than that of the traditional FunkSVD algorithm. The experimental results also show that the IFABG algorithm has a good acceleration effect. Under the same size data set, the IFABG algorithm is faster than the traditional FunkSVD, and the acceleration ratio can be as high as 19.27.

show abstract

“…However, these methods cannot be accelerated significantly on graphics processing unit (GPU) systems because the serious over‐writing problem and thread divergence may occur. The fourth paper, ‘GPUSGD: a GPU‐accelerated stochastic gradient descent algorithm for matrix factorization’ , by Jing Jin, Siyan Lai, Su Hu, Jing Lin, and Xiaola Lin, proposes an efficient GPU algorithm, named GPUSGD, to solve the matrix factorization problem based on SGD method. The proposed GPUSGD not only can handle the over‐writing problem but also can avoid the performance loss caused by the thread divergence.…”

Section: Themes Of This Special Issuementioning

confidence: 99%