Optimizing deep learning inference on mobile devices with neural network accelerators

Deep learning has now been widely used in intelligent apps of mobile devices.In pursuit of ultra-low power and latency,integrating neural network accelerators(NNA)to mobile phones has become a trend.However,conventional deep learning programming frameworks are not well-developed to support such devices,leading to low computing efficiency and high memory-occupation.To address this problem,a 2-stage pipeline is proposed for optimizing deep learning model inference on mobile devices with NNAs in terms of both speed and memory-footprint.The 1 st stage reduces computation workload via graph optimization,including splitting and merging nodes.The 2 nd stage goes further by optimizing at compilation level,including kernel fusion and in-advance compilation.The proposed optimizations on a commercial mobile phone with an NNA is evaluated.The experimental results show that the proposed approaches achieve 2.8×to 26×speed up,and reduce the memory-footprint by up to 75%.

A new methodology for identification of potential pay zones from well logs: Intelligent system establishment and application in the Eastern Junggar Basin, China

In recent years, as the exploration practices extend into more complicated formations, conventional well log interpretation has often shown its inaccuracy and limitations in identifying hydrocarbons. The Permian Wutonggou Formation hosts typical clastic reservoirs in the Eastern Junggar Basin. The sophisticated lithology characteristics cause complex pore structures and fluid properties. These all finally cause low well testing agreement rate using conventional methods. Eleven years＇ recent statistics show that 12 out of 15 water layers have been incorrectly identified as being oil or oil/water layers by conventional well log interpretation. This paper proposes a methodology called intelligent prediction and identification system （IPIS）. Firstly, parameters reflecting lithological, petrophysical and electrical responses which are greatly related to reservoir fluids have been selected carefully. They are shale content （Vsh）, numbered rock type （RN）, porosity （φ）, permeability （K）, true resistivity （RT） and spontaneous-potential （SP）. Secondly, Vsh, φ and K are predicted from well logs through artificial neural networks （ANNs）. Finally, all the six parameters are input into a neuro-fuzzy inference machine （NFIM） to get fluids identification results. Eighteen new layers of 145.3 m effective thickness were examined by IPIS. There is full agreement with well testing results. This shows the system＇s accuracy and effectiveness.