Torch
The Swiss institute Idiap released a modular machine learning library in C++, from which PyTorch grew fifteen years later.
Why it matters
The beginning of the tool line that reaches today's standard, and it begins in Martigny rather than in California.
The authors include Ronan Collobert, Samy Bengio and Johnny Mariethoz. The defining feature is modularity: layers and criteria assemble like parts. The later Torch7 moved to Lua and became a favourite in Facebook and DeepMind research groups precisely for its flexibility. PyTorch in 2017 inherited both the name and the philosophy of immediate execution.