Computing was in a state of crisis. As the price of the hardware continued to fall, and the power of processors continued to rise, the cost of programming the devices was becoming a major part of the overall cost. Small wonder. In 1954, although there were only a few dozen computers in existence, their programming was carried by a small army of experts coding at worst in binary code and at best in some form of assembler.
Enter John Backus. This somewhat unorthodox but brilliant mathematician had been hired off the street by IBM in 1949 to work on its Selective Sequence Electronic Calculator (SSEC) computer. As a result of his experience there, he realised that the only way out of this impasse was to enable programmers to code more efficiently using a higher level of abstraction. But that would not, itself, be sufficient. It would be essential that the final binary code produced by any automatic coding system be as efficient as that being produced by the laborious hand coding.
A parallel Fortran framework for neural networks and deep learning
