174.5
From the United States to the Soviet Union and Back Again: A Transatlantic Story of Machine Learning
From the United States to the Soviet Union and Back Again: A Transatlantic Story of Machine Learning
Friday, 20 July 2018: 16:30
Location: 603 (MTCC SOUTH BUILDING)
Oral Presentation
Machine learning algorithms based on so-called "neural networks" are often considered today as the future of the AI (artificial intelligence) and a solution to many major problems faced by the humanity. Highly optimistic and futuristic narratives produced by (and around) this field generally obscure its mid-20th century origins, in particular its deep connection to cybernetics. The concept of a “neural network” itself was introduced by Warren McCulloch and Walter Pitts back in the 1940s. A subsequent career of this idea and of its implementations has been quite turbulent, from high expectations to complete oblivion, and numerous rediscoveries. Although its history begins to be known today, these narratives generally omit the European perspective, and especially the Soviet experiences in this field. This paper intends to bring into focus a transatlantic travel of the idea of Perceptron developed by Frank Rosenblatt in 1958. This machine designed to learn pattern recognition became one the first implementations of the artificial neural network. The model of Perceptron was adopted and creatively appropriated by Soviet scientists early after its first appearance in the United States. I will consider the case of the Soviet Institute of Management Problems (IPU), whose researchers were among the pioneers of algorithms for automatic image recognition in the 1960s (motivated by demand from the military). Inspired by the Perceptron model, members of the IPU led by Mark Aizerman developed a geometric learning method (a so-called method of potential functions) quickly recognized and adopted by American and European scientists and engineers. This work was a part of a larger theoretical research program in the domain of image recognition and self-learning machines ("unsupervised learning"). This episode is not only an important chapter in the history of artificial intelligence, but also an exciting case of the effective intellectual and technological transfer through the Iron Curtain.