Neural Processing from the Ground Up, & Phoenix Architecture

About this event

The MLAI Meetup is a community for AI researchers and professionals which hosts monthly talks on exciting research. Our format is: • 6:00 - 6:20: Socializing • 6:20 - 6:40: Announcements and AI news • 6:40 - 7:40: Talk(s) and Q&A • 7:40 - 8:00 Networking • 8:00: Head to the nearest pub for dinner Tristan Penman: "Neural Processing from the Ground Up" Talk description: Have you ever wondered how an NPU or TPU works under the hood? A lot is written and said about how these purpose-built chips can accelerate neural networks, but not so much about how they actually work. In this talk we'll see how, starting from basic operations like dot products and matrix multiplications, we can implement non-trivial model architectures such as CNNs and Transformers. From there, we'll see how those same techniques are applied in real embedded NPUs and data-centre scale TPUs. Some concepts we’ll touch upon include systolic arrays, tiling, and the impact of memory hierarchies. Code examples will be shared where possible, along with diagrams to help build intuition. Speaker bio: Tristan is a Principal Engineer at a Melbourne-based EdTech startup. His recent work centres around bringing AI to the classroom, in ways that improve learning outcomes, increase student engagement, and minimise strain on teachers. In his spare time, Tristan dabbles in reverse engineering, hardware hacking and FPGA design. Raúl V: "Phoenix Architecture" Abstract: Drop the code, (re)generate specs that are well defined for (multi) agents to replicate every single time. Machine code is no longer scarce and increasingly disposable. The limiting factor is no longer writing software, but understanding, evaluating, and governing it. Speaker Bio: Raúl is the solopreneur founder of Quantaliz, a private AI software development. He is a neerk (nerd + geek) systems engineer, always looking to learn, explore, exploit the most out of the computers.