Information Based Neural Network

Researchers at Purdue University have developed new neural networks (NN) and convolutional neural networks (CNN) for artificial intelligence (AI) that can be used to process images with high efficiency. The neural networking technique utilizes information typically ignored in AI, including optically analyzing single molecules directly from image content. Currently, deep neural networks (DNN) are used to measure single molecules by rotating a double helix point spread function (PSF) or implementing Gaussian models, which requires complex mapping to connect input with output, and depends on domain expertise. The Purdue University approach instead relies on a network architecture that recognizes inherent information from images at a rate of one million PSFs for each task. In addition, this technology can accurately distinguish one hundred to three hundred sub-regions on average, which is equivalent to twenty to sixty raw data frames. Key features of a single molecule such as location, orientation, and wave-front distortion can now be obtained with ease. This enhanced decoupling technique allows researchers to process biological images faster than ever before.

Advantages:

-Measures at single molecule level

-Enhanced imaging

-Easy storage and interpretation of images

Potential Applications:

-Biomedical research

-Biological image processing

-Artificial Intelligence

TRL: 7

Intellectual Property:

Keywords: Neural networks, convolutional neural networks, artificial intelligence, image processing, deep neural networks, single molecule analysis, biological images, enhanced imaging, biomedical research, wave-front distortion, decoupling technique