Back

Speaker "Yuwei Cui" Details Back

 

Topic

Real-time streaming data analysis with HTM

Abstract

Much of the world’s data is becoming streaming, time-series data. It becomes increasingly important to analyze streaming data in real-time. Hierarchal Temporal Memory (HTM) is a detailed computational theory of the neocortex. At the core of HTM are time-based learning algorithms that store and recall spatial and temporal patterns. HTM is well suited to a wide variety of problems; particularly those involve streaming data and time-based patterns. The current HTM systems are able to learn the structure of streaming data, make predictions and detect anomalies. It is distinguished from other techniques in its ability to learn continuously in a fully unsupervised manner. HTM has been tested and implemented in software, all of which is developed with best practices and is suitable for deploying in commercial applications. The core learning algorithms are fully documented and available in an open source project called NuPIC. HTM not only advances our understanding of how the brain may solve the sequence learning problem but is also applicable to real-world sequence learning problems from continuous data streams.

Profile

I am a Research Staff Member at Numenta, a company focused on Machine Intelligence. My professional interests are in the areas of Artificial Intelligence, Computational Neuroscience and Machine Learning. I discovered my interest in AI while I studied physics in the University of Science and Technology of China. I later went on to get a PhD in computational neuroscience, specializing in understanding how our visual system process sensory inputs and contribute to perceptions, from the University of Maryland at College Park. I became fascinated by the brain and reverse engineering its underlying computational principles. I have published numerous peer-reviewed scientific articles in Neuroscience and AI.