The Math Behind Building An AI Using DNA #SoME3

Frank Rosenblatt invented the perceptron in 1958. Geoffrey Hinton, Yann LeCun, and Yoshua Bengio pioneered deep learning earning the 2018 Turing Award. Machine learning researchers develop architectures for image classification, natural language processing, and scientific applications. Neuroscientists study biological neural networks inspiring artificial implementations.

Naka and Rushton characterized sigmoid response curves in biological neurons in the 1960s. Machine learning pioneers adopted sigmoid functions for artificial neural networks. Researchers study activation function properties affecting network training and expressivity. Modern deep learning often employs ReLU and variants though sigmoid remains relevant for specific applications.

Yann LeCun and colleagues developed efficient matrix implementations for neural networks enabling GPU acceleration. Linear algebra underpins all neural network computations. Hardware designers optimize matrix multiplication units in AI accelerators—TPUs, GPUs achieving thousands of trillions of operations per second. Researchers develop low-precision arithmetic reducing computational costs.

Lulu Qian and colleagues demonstrated DNA-based matrix operations implementing neural network computations. Synthetic biologists engineer split-gate architectures enabling selective activation. Theoretical computer scientists analyze DNA computing’s computational complexity. Researchers develop increasingly sophisticated molecular logic circuits.