As roads melted, infrastructure ground to a halt and the UK suffered record-breaking temperatures; at MLOps London, the show triumphantly went on. Data scientists, engineers and ML enthusiasts braved sweltering and stuffy commutes to travel to Shoreditch where, thankfully, the air conditioning was on full blast and two superb talks were delivered.
Alex Persin, a senior software engineer at Wayve kicked us off with a talk about how they deliver massive-scale inference for autonomous vehicles. Unlike traditional self-driving car companies, who focus on assembling lots of different sensors and algorithms to control a vehicle, Wayve believes that the whole process of driving can be handled by large neural networks with a few camera inputs – much the same way that we humans learn to drive.
Training the self-driving networks requires millions of images which get enhanced by adding depth and terrain information. As you can imagine, running inference on millions of images is a complex scenario to handle. Wayve uses Kubernetes and PySpark to manage this. You can learn all of the details, including how Wayve made massive infrastructure cost savings in the recording of Alex’s talk.
Our second talk of the day saw the introduction of a new format. Rather than the usual slide presentation (sometimes plus demo), we held a “fireside chat” style discussion with Byron Allen.
Byron is the AI and ML practice lead for Contino, a consultancy who focus on cloud native and digital transformation projects. His wealth of experience delivering machine learning projects for clients means he can talk about almost anything but the session focused heavily on a recent blog Byron published on how to build cross functional data science teams.
Our next MLOps London will take place on the 6th September where we’ll have talks on Kubeflow Pipelines and reducing technical debt. As always, it’ll be broadcast live on YouTube if you aren’t able to come along in person.