If C is so strong a programming language why can’t it replace Python in AI and data science?
My (former*) employer is a facial recognition company with our core technology written in C styled C++, meaning we use classes and STL, but the code is basically C. Our machine learning group also does all it’s work and training in C/C++.
The result is an extremely fast product, with a very small footprint. Granted, the memory handling requires extra, significant effort, but the result is a single integrated application with a web server, a database, the facial recognition core, and a modern UI. This is in contrast to the “traditional” machine learning application that is hosted inside a “server stack”, requires multiple servers and hardware instances to operate, and is accessed through an external UI such as a web browser.
We can put our single application on an Intel Compute Stick (those look like a big USB drive) and it runs perfectly fine, requiring no other software.
It takes effort, and old school C programmers, but it can be done. The result of the extra effort is a product which those working with a traditional server stack simply cannot out perform. We can process 25 million faces per second per CPU core, a level exponentially higher than anyone working in a higher level language can achieve. Combine that with servers having 32+ cores, and that is a hell of a lot more facial compares per second than can be achieved with the limited number of GPUs one can have on as single server.
I left the position end of April, ’21 after working there 7 years
