Citadel Securities hired a machine learning performance specialist from Google
Citadel Securities isn't just taking the youngest talent from major technology companies, it's taking their experienced people too. The market-maker's latest hire is a performance optimization specialist in a machine learning language likened to C++.
Alan Kelly has joined Citadel Securities as a high-performance computing engineer. He spent the last five and a half years at Google, where he most recently worked on XNNPack, a "highly optimized library of floating-point neural network inference operators," which serves as the CPU backend for TensorFlow Lite, a Python machine learning library that uses the computational power of on-premise devices.
Before that, Kelly, a PhD engineer, worked directly on TensorFlow's runtime and kernal performance optimization for Google. TensorFlow is the performance-optimized cousin of PyTorch, the much more popular Python library in use at Jane Street. TensorFlow is thought to operate like compiled languages such as C++ and can be difficult to code in, but rewards that difficulty with performance.
Electronic trading firms are thought to have some of the largest fleets of GPUs and CPUs in finance. XTX Markets, which paid its partners £20m ($27m) per head last year, has a fleet of over 25,000 GPUs. Jane Street is thought to have more than 5,000. Last year, Citadel Securities revealed that it could concurrently use one million CPU cores thanks to its partnership with Google Cloud. Kelly will therefore have his work cut out for him.
Have a confidential story, tip, or comment you’d like to share? Contact: Telegram: @AlexMcMurray, Signal: @AlexMcMurrayEFC.88 Click here to fill in our anonymous form, or email editortips@efinancialcareers.com.
Bear with us if you leave a comment at the bottom of this article: all our comments are moderated by human beings. Sometimes these humans might be asleep, or away from their desks, so it may take a while for your comment to appear. Eventually it will – unless it’s offensive or libellous (in which case it won’t.)
