Engaging new and diverse audiences with ML
summary
Pandemic-era player growth shows 38% report increased playtime and 65% longer sessions, and only 1–2% are persistent trolls, so the solution focuses on context-aware detection and adaptive experiences. Uses include BERT fine-tuned on game chat, AutoML/BQML models, autoencoders for anomaly (cheat/grief) detection, and Vertex AI deployment at planet scale for millions concurrent. Live pipelines push per-event 0–1 disruption scores through Pub/Sub and Dataflow for aggregation, while the What‑If tool and fairness monitoring control bias and model drift.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 0:24Current gaming landscape
- 4:07Disruptive behavior
- 7:57Toxicity
- 13:04Google Cloud
- 14:38Skillbased matchmaking
- 16:17Dynamic experiences
- 17:49Word embeddings
- 18:54Unidirectional vs Bidirectional
- 20:02Auto Encoders
- 20:53Putting it all together
- 22:51Summary
related talks
37mAI/ML Strategies & Techniques for a Post Pandemic World
50mThe tech behind modern AI/ML use cases in Gaming
23mCustom-made games with machine learning and big data
32mMachine Learning for Game Developers
41mMachine Learning Models in a Dynamic Environment
18mScaling ML workflows for real-time moderation challenges
33mProduction ML Models in the Mobile Gaming Industry
10mML-based Log Analysis for Faster Debugging