大数据分析技术将取代人脑功能

根据在美国 IBM Almaden研究中心举行的一场感知运算研讨会上发表之技术简报，大数据(big data)分析学不但能模仿人脑，甚至有可能取而代之。 在该场研讨会上，风险投资业界资深人士、也是Sun Microsystems 共同创办人的Vinod Khosla呼吁，科技演进应该朝向藉由把更多医疗决策交由智能系统进行，以减少人为疏失。他认为，目前的医疗都仰赖医学专家的意见，因此往往是：“根据一连串通常正确比错误来得多的偏见(bias)；我怀疑我们是不是需要让人类跳出该种循环。” Khosla列举了许多研究，量化人为疏失对医疗诊断与治疗方面带来的影响；他主张应该投入更多资源在医疗传感器与分析学方面的研究，而他自己目前也投资多家开发相关技术的新创公司，包括AliveCor、Ginger.io、Kyron、Quanttus等。 “资料科学在未来十年为医疗领域带来的贡献，将比生物科学来得多。”Khosla指出，目前的一些数字医疗产品，像是协助学步幼儿/复健者的装置，未来将会因 为消费者需求而提供更复杂的功能：“我认为改变将来自消费者主导的医疗保健领域，而我希望有少数几个典范能颠覆整个产业利益结构。” 此外，感知运算研究学者Jeff Hawkins则展示了他的新技术成果“Grok”，是一种运用于人脑新皮质(neocortex)的技术，能藉由建立“SDR (sparse distributed representations)”模式追踪大型数据集(large datasets)。 Grok是以Hawkins的公司所开发的一种SDR算法为基础，并以开放源码形式释出：“我们不知道该如何以数学形式来描述它，但我认为这会是感知运算的基本功能区块。” 该 软件已经在亚马逊(Amazon)的云端服务上使用，能针对异常数据流进行快速侦测以及排名，并提供能快速对那些数据流进行调查的工具。Hawkins将 该种工具形容为安装在一个功能非常基本之大脑、也就是尺寸仅“老鼠大脑新皮层的千分之一”上的“单一传感器”──就像是一只耳朵。 “我们看到了一些不寻常的东西──而我们不知道那是如何发生、或是其根源；”他表示，“这需要在一个既定系统中建立物理模型，并去理解一切可能是如何运作。” 无论如何，Hawkins声称Grok是一种能支持包括金融、电子商务，以及制造业等多样化应用领域的巨量数据分析工具；澳洲业者CEPT Systems已经在自然语言识别方案，采用了Grok核心的开放源码算法。 本文授权编译自EE Times，版权所有，谢绝转载 编译：Judith Cheng 参考英文原文：Big Data May Mimic, Replace Brain，by Rick Merritt

相关阅读：
• 智能互联时代移动终端的创新趋势
• 认知计算时代，智能手机将比人更聪明
• 2017年亚太区大数据市场规模将达23.8亿美元YhHesmc

{pagination} Big Data May Mimic, Replace Brain Rick Merritt SAN JOSE, Calif. — Big data analytics may both mimic the human brain and replace it, according to presentations at an IBM symposium here. Veteran venture capitalist Vinod Khosla called for advances that reduce human error by putting more healthcare decisions into the hands of smart systems. Separately, cognitive computing researcher Jeff Hawkins showed advances in applying techniques used in the neocortex to sorting large datasets. Today's medicine relies on doctors' expert opinions, which are often "based on a series of biases that are more often right than wrong," said Khosla, a serial entrepreneur and co-founder of Sun Microsystems. "I suspect we will need humans out of the loop." Khosla cited numerous studies quantifying the impact of human errors in diagnosing and treating health issues. He argued for more work on healthcare sensors and analytics, an area where he is currently investing in several startups including AliveCor, Ginger.io, Kyron, and Quanttus. "Data science will do more for medicine in the next 10 years than biological science," Khosla told a symposium on cognitive computing at the IBM Almaden Research center here. He referred to today's digital medicine products as "clumsy toddler steps" that will lead to more sophisticated offerings that empower consumers. "I think change will come from consumer-driven healthcare, and I hope a few role models will cause an avalanche of interest." Separately, Jeff Hawkins, described his latest product, Grok. It uses a technique employed in the neocortex to track large datasets by creating so-called sparse distributed representations (SDRs). Grok is based on an SDR algorithm Hawkins's company released as an open-source code. "We don't know how to characterize it mathematically, but I'd argue this is a basic building block of cognitive computing," he said. The software, released for use on Amazon's cloud service, handles rapid detection and ranking of anomalies in data streams and provides tools to investigate them quickly. Hawkins described the tool as a single sensor -- like an ear -- attached to a very rudimentary brain "a thousandth the size of a mouse's neocortex." "We see something unusual -- we don't know why it happened or its root cause -- that requires a physics model and an understanding of how things are supposed to work" in a given system, he said. Nevertheless, he claimed the tool is a powerful one that could be applied broadly to big data analytics problems in areas as diverse as finance, web sales, and manufacturing. CEPT Systems in Austria is already using the open-source algorithm at the heart of Grok in its work on natural language recognition.