机译:使用近似计算技术对神经网络分类器进行弹性训练,以实现硬件优化的实现
Univ Fed Minas Gerais Grad Program Elect Engn Av Antonio Carlos 6627 BR-31270901 Belo Horizonte MG Brazil;
DFKI GmbH Cyber Phys Syst Bibliothekstr 5 D-28359 Bremen Germany;
field programmable gate arrays; learning (artificial intelligence); floating point arithmetic; neural nets; resilient training; neural network classifiers; approximate computing techniques; hardware-optimised implementations; high-end processing platforms; research community; arithmetic operations; mathematical functions; customised adaptive artificial neural networks training method; mixed operations; complex rounding methods; optimal simplifications; specific deep learning problems; optimised framework; precise operations; standard training algorithms; network structures; half-precision floating point; square root operations; field programmable gate array implementation complexity; reciprocal square root approximation; software implementation; mixed-precision platform; precision conversions; machine learning applications;
机译:利用弹性神经网络训练算法和质构分析,通过反向神经网络识别计算机断层图像的对数特征:初步结果
机译:使用增强型弹性反向传播神经网络(ERBPNN)和快速自适应神经网络分类器(FANNC)评估加纳共同基金业绩的稳健性
机译:使用基于神经网络的预测控制技术识别近似神经网络的二阶工厂
机译:具有计算错误的加速器的弹性神经网络训练
机译:通过开发和利用新型跨学科软计算理论和技术来实现智能QoS网络
机译:用模拟突触装置使用近似反向衰减的片上训练尖刺神经网络
机译:用于计算错误的加速器的弹性神经网络培训