Random Ensemble Hypernetwork for Pattern Recognition with Enzymatic Weight Update

Pattern recognition is a major division of machine learning which focuses on learning the patterns and regularities in data. It differs to that of pattern matching where only exact matches are found. However in the field of DNA computing, molecular pattern recognition has not been well established due to the lack of control of molecules in liquid state, instability and inaccuracy to solve such problems. Also the cost of designing the mass amount of DNA to represent data is a realistic issue. Here, we propose the random ensemble Hypernetwork as a model for pattern recognition in vitro with handwritten digit data encoded to DNA. For the manipulation of DNA in vitro, this molecular programming model is proposed to build a massively-parallel classification device, with the use of enzymatic weight update. Furthermore, a novel method of encoding vast amounts of data to DNA is introduced, also allowing the production of random hyperedges, a key difference to previous studies or computational implementations which only focused on fixed hyperedges.