No Data Scientists Were Harmed: ML Inference for Regular Game Devs

Viviane and Xavier are visiting from the Unity Montreal office, and they have kindly offered to give a talk about using machine learning inference in Unity games (actually same talk which will also be given internally)

Xavier Désilets - Senior Software Developer, Unity
Viviane Rochon Montplaisir - Senior Software Developer, Unity

Are you a game dev who keeps hearing words like "tensors" and "logits" and quietly alt-tabs back to your prefab folder? This session is for you.

In the first half, we'll do a fast, practical ML 101 specifically for game developers. We'll cover what tensors actually are (in human language), quantization, how model graphs work, and what to make of different model formats.

In the second half, we'll focus on the decisions you have to make to ship something: how to pick a model size that won't set your players' fans on fire, what formats and hardware are supported, and how to think about speed vs. quality. We'll walk through real-world input and output pipelines. We'll ground this with currently available Unity Sentis samples.

By the end, you won't be a data scientist, but you will know how to choose, run, and wire up ML models at runtime in Unity without breaking your frame budget or your brain.

Program:

17:30 - Sandwiches, sodas and mingling
18:00 - Welcome
18:10 - Presentation
19.15 (approx.) - Mingling continues
21:30 - Venue closing