The first neural network called Perceptron was proposed by Marvin Minsky at MIT in 1951, which paved the way for the new field of Machine Learning:
1960s – 1970s
Machine Learning pioneers throughout the world continued to combine optimization, statistical modeling, dynamic systems, and computer science.
An optimization algorithm called Backpropagation emerged that successfully trained Multi-Layer Perceptrons (MLPs) and is still the dominant training algorithm for neural networks.
The emergence of Recurrent Neural Networks (RNNs) enabled modeling sequences of numbers and symbols. Hidden Markov Models (HMMs) emerged as a standard method for speech recognition with neural networks.
Late 1990s – 2000s
Support Vector Machines (SVMs) emerged as the best predictive models for time-series predictions, classification, and pattern recognition, beating MLPs. Long-Short Term Memory (LSTM) cells emerged to replace memoryless cells of RNNs for modeling sequences, which dramatically improved prediction accuracy. Yann LeCun introduced Convolutional Neural Networks (CNNs) that revolutionized AI.
2010s – “Deep Learning Era”
Deep RNN and CNN architectures led to the emergence of Deep Learning as the most powerful model for image, speech, and text processing. Deep Learning has revolutionized Natural Language Processing (NLP). Ubiquity of fast GPU computing and availability of massive datasets has opened new doors for applications every day.