Agenda
Machine Learning Roundtable: ML in Gameplay
Mathieu Nayrolles (Director of Applied AI/ML, Supercell)
Pass Type:
Academic Pass, Festival Pass, Game Changer Pass, Indie Pass, Start-Up Pass -
Get your pass now!
Track:
Programming, Production & Team Leadership
Audience Level:
All
Format:
Roundtable
Vault Recording: TBD
Audience Level: All
Models that touch the player in the moment. Local or remote, classic or generative, they all have to survive a frame budget instead of a slide. Sometimes it works. Sometimes it barely works. Sometimes it should not have shipped.
This round table is about what actually reached players in the last twelve months. On-device small models now answer an NPC line in tens of milliseconds. Neural texture compression turns gigabytes of VRAM into hundreds of megabytes in demos, and almost nobody has put it in a shipping game. Generative NPCs got smarter in trailers faster than in builds. We will talk about what made it past certification, what the player actually felt, and what got quietly toggled off because a cheaper model stopped being fun.
This is not a demo reel. It is a conversation about real systems, real latency, real cost per player, and the gap between what the hardware can do and what you can afford to run.
This round table is about what actually reached players in the last twelve months. On-device small models now answer an NPC line in tens of milliseconds. Neural texture compression turns gigabytes of VRAM into hundreds of megabytes in demos, and almost nobody has put it in a shipping game. Generative NPCs got smarter in trailers faster than in builds. We will talk about what made it past certification, what the player actually felt, and what got quietly toggled off because a cheaper model stopped being fun.
This is not a demo reel. It is a conversation about real systems, real latency, real cost per player, and the gap between what the hardware can do and what you can afford to run.
Takeaway
Attendees leave with a clearer picture of what it takes to ship ML in the game loop today. This includes on-device versus cloud tradeoffs, memory and frame budgets, the latency players tolerate, and when a small local model beats a frontier one.
Rather than best practices, attendees gain practical intuition from peers about what worked at runtime, what broke at scale, and why. The goal is to help you decide what to put in front of a player next and what to leave in the lab.
Rather than best practices, attendees gain practical intuition from peers about what worked at runtime, what broke at scale, and why. The goal is to help you decide what to put in front of a player next and what to leave in the lab.
Intended Audience
This round table is intended for ML practitioners and game developers already shipping or seriously considering ML in gameplay. This includes gameplay programmers, AI engineers, technical designers, rendering and animation engineers, and tech leads.
An intermediate understanding of ML helps, but hands-on experience matters more than a formal background. If you have shipped something ML-driven to players, or you are about to, this session is for you.
An intermediate understanding of ML helps, but hands-on experience matters more than a formal background. If you have shipped something ML-driven to players, or you are about to, this session is for you.