Reverse engineering the neocortex to figure out how the brain works is one of humanity's grand scientific challenges. We are fortunate to be living in an era when neuroscience techniques are advancing at an amazing pace, giving rise to a wealth of data on everything from synapses, to neurons, to entire cortical regions. Our research team focuses on theory. We work with experimentalists and published results to derive an understanding of what the neocortex does and how the detailed architecture of the neocortex implements this. We test our theories via simulation, mathematical analysis, and experimental partnerships. While there are many experimental neuroscience laboratories, Numenta is unique in its focus on large-scale cortical theory and simulation.
Although there is much to do, we have made significant progress on several key aspects of cortical theory. Our first focus was on how the neocortex learns the structure in streams of sensory data. This led to a comprehensive theory of why neurons in the neocortex have thousands of synapses and active dendrites, and why they are arranged in minicolumns as observed in the neocortex. We have taken this theoretical advance and applied it to commercially valuable problems in anomaly detection and prediction in streaming data. To further that goal we created a benchmark for anomaly detection in streaming data that for the first time provided a means for comparing the results of different algorithms for anomaly detection. Our current research focus is developing a comprehensive theory of how the neocortex learns through movement — what is often called “sensorimotor” learning.
We document our research in several ways, including peer-reviewed journal papers, conference proceedings, research reports, invited talks, and a living book titled BAMI (for biological and machine intelligence). In addition, we place all of our software, from our commercial grade applications to our daily research commits, in our open source project, NuPIC. We strive to be completely open in everything we do. We welcome collaborations with both neuroscientists and machine learning researchers.