Three keynote talks delivered at SNAI 2025
SNAI 2025 featured three keynote talks, one on each day of the symposium. The keynote programme was curated by the General Chair and the Program Chairs to span the breadth of the symposium's scope — learned networking in production, data-plane ML at line rate, and federated learning under bandwidth constraints.
Aisha Rahman — National University of Singapore (General Chair, SNAI 2025)
Delivered on Day 1 (24 October 2025). A retrospective on what the learned-networking research agenda has produced over the past decade, separating research artefacts that have shipped into production networks from those that have remained research curiosities. The talk drew on a decade of operational experience inside a tier-1 ISP and surveyed where learned congestion control, learned traffic engineering, and learned fault diagnosis have actually moved the needle.
Viktor Halász — Principal Researcher
Delivered on Day 2 (25 October 2025). A technical deep-dive on the systems challenges of running machine-learning inference primitives inside programmable data planes at line rate. The talk covered match-action pipeline design for tree-based models, quantization for fixed-precision feature extraction, and the trade-offs between accuracy, throughput, and silicon area on commercial SmartNICs.
Grace Adeyemi — University of Cape Town
Delivered on Day 3 (26 October 2025). A practitioner-oriented talk on federated learning deployed over real, bandwidth-constrained, intermittently connected links in sub-Saharan Africa. The talk covered gradient compression, client selection under mobility, and the operational realities of running federated training across cellular networks with intermittent backhaul.