|
Biblioteca Digital do IPG >
Escola Superior de Tecnologia e Gestão (ESTG) >
Artigos em Acta de Conferência Internacional (ESTG) >
Utilize este identificador para referenciar este registo:
http://hdl.handle.net/10314/3250
|
| Título: | GPUMLib Framework: Using the GPU to Empower Machine Learning Research |
| Autores: | Lopes, Noel
Ribeiro, Bernardete |
| Palavras Chave: | GPU Computing
machine learning algorithms |
| Data: | 2015 |
| Resumo: | The amount of information being produced by humans is continuously increasing, to the point that we are generating, capturing and sharing an unprecedented volume of data from which useful and valuable information can be extracted. However, obtaining the information represents only a fraction of the time and effort needed to analyze it. Hence, we need scalable fast Machine Learning (ML) tools that can cope with large amounts of data in a realistic time frame. As problems become increasingly challenging and demanding, they become, in many cases, intractable by traditional CPU architectures. Accordingly, novel real-world ML applications will most likely demand tools that take advantage of new high-throughput parallel architectures. In this context, today GPUs (Graphics Processing Units) can be used as inexpensive highly-parallel programmable devices, providing remarkable performance gains as compared to the CPU (it is not uncommon to obtain speedups of one or two orders of magnitude). However mapping algorithms to the GPU is not an easy task. To mitigate this effort we are in the process of building an open source GPU Machine Learning Library – GPUMLib to help ML researchers and practitioners worldwide. This presentation focus on the challenges of implementing GPU ML algorithms using CUDA. Moreover, it presents an overview of GPUMLib algorithms and tools and highlights its main benefits. |
| URI: | http://hdl.handle.net/10314/3250 |
| Aparece nas Colecções: | Artigos em Acta de Conferência Internacional (ESTG)
|
Ficheiros deste Registo:
| Ficheiro |
Descrição |
Tamanho | Formato |
|---|
| Ata Cientifica_NoelLopes.pdf | | 196Kb | Adobe PDF | Ver/Abrir | |
|
|
