Neuromorphic Spatiotemporal Where-What Machines
Despite a promising beginning, pattern recognition software using a “neural network” approach in general has encountered serious roadblocks limiting the rate of progress. Traditional methods cannot “attend” and “recognize” using the same network structure. For example, a system can find interesting regions, but cannot recognize objects: a system can only recognize objects that have already been segmented and separated from their natural background.
A crucial challenge is to handle the demands for bottom-up “attention” to the presence of given objects in a particular class and efficiently couple that with a top-down knowledge-based “recognition” of the selected objects. This results in efficient pattern recognition. Such a combined top-down and bottom-up architecture is necessary for the kind of information processing that rapidly distinguishes “friend from foe,” spots a looming pothole while ignoring other information, or that analyzes information relating to the relative position of words and features as opposed to merely identifying a string of information.
Description of Technology
Michigan State University’s software algorithm, while of the neural network variety, uses no back propagation (a common feature of neural net architectures). The software is massively parallel, uses a Hebbian Learning method, and mimics the modularity of the brain, including the compartmentalization (in space and function) of various cognitive functions (e.g., auditory, decision making, and positional location). The algorithm incorporates several potential breakthroughs:
- A model of how the human brain focuses attention on designated objects in space and time, allowing the algorithm to zero in on subjects of interest (e.g., a human running in front of a car or a looming pothole) and effectively ignoring all background information (e.g., houses, shadows, and so on).
- A combination of a “top down” and “bottom up” architecture loosely mimicking how the brain handles information processing in the cerebral cortex. It is a system for putting the modular pieces together.
- High effectiveness: The software has already been tested on tasks, including object recognition, and has been found to be superior to existing software alternatives.
- Massive parallelism: The software architecture is compatible with, and runs most effectively with, massively parallel chips.
- Generality of applications: The software incorporates a generalized information processing architecture loosely modeled on the modular architecture and hierarchical information processing of the human brain.
This invention has applications in:
- text understanding
- object and situation recognition (e.g., steering, braking, and collision avoidance in automotive applications)
- defense applications requiring object recognition and situation assessment
- advanced search engines providing the users with an experience like that of talking with a knowledgeable friend
Patented US 8,694,449
Juyang Weng, Matthew Luciw, Mojtaba Solgi, Zhengping Ji
For Information, Contact:
Michigan State University