←
Sam Snodgrass299 posts
Sam Snodgrass avatar

Sam Snodgrass

@SamPSnodgrass

Machine Learning Lead at Escape Velocity Entertainment; AI, ML, and PCG Researcher; Academic in my free time;

◷ Joined October 2016▧ 17 Photos and videos
Sam Snodgrass avatar
Sam Snodgrass@SamPSnodgrass·14 May 2024

Be sure to chat with @sebastianberns about this super cool work if you're at CHI this week! Had a lot of fun with this team looking into how human perceptions align (or don't) with common similarity measures used in PCG and other domains.

This week, I am at #chi2024 to present our latest paper. “Not All the Same: Understanding and Informing Similarity Estimation in Tile-Based Video Games” arxiv.org/abs/2402.18728. Co-authors: @CIGbalance, Laurissa Tokarchuk, @SamPSnodgrass and @creativeEndvs.

Matthew Guzdial avatar
⟳ Sam Snodgrass retweeted
Matthew Guzdial@MatthewGuz·30 May 2023

Tired of AI hype but interested in reading about how machine learning (generative AI if you like) can be used to generate content in games? Well do I have the book for you!

Post attachment
Matthew Guzdial avatar
⟳ Sam Snodgrass retweeted
Matthew Guzdial@MatthewGuz·30 Apr 2023

New paper on Procedural Content Generation via Knowledge Transformation (PCG-KT) from @riffsircar, me, @SamPSnodgrass, @Autumnsburg, Tiago Machado, and @gillianmsmith. We propose a framework for transforming knowledge between domains, allowing for generation for unseen games!

Post attachment
Sam Snodgrass avatar
Sam Snodgrass@SamPSnodgrass·16 Dec 2022

From subtweet to published book in ~2 years? That's not a bad turnaround. A PCGML textbook now exists! I had a great time writing this with @MatthewGuz and @Autumnsburg And I'm really hoping this book helps bring more people into the field! Link to book: link.springer.com/book/10.10…

I’ve had to explain to a few students recently that there isn’t a PCGML textbook because the topic only coalesced ~4 years ago. But if any academic publishers want to pay @Autumnsburg, @SamPSnodgrass, and I to write a textbook I certainly wouldn’t say no.

Sam Snodgrass avatar
Sam Snodgrass@SamPSnodgrass·25 Oct 2022

Well, well, well. Look who got a title bump. I'll still be bringing PCGML to game devs, and exploring new ways that PCGML can impact game design, and improve bots and testing. Now with some seniority sprinkled in! Looking forward to growing alongside @modl_ai in my new role

0:22
Sam Snodgrass avatar
Sam Snodgrass@SamPSnodgrass·21 May 2021

Thank you if you stuck with me through the thread! There is some really important and interesting work to be done in this space, and this thread just talks about a few ways. But by doing this work we can get closer to one of PCGML's goals: making PCG more broadly accessible 18/18

Sam Snodgrass avatar
Sam Snodgrass@SamPSnodgrass·21 May 2021

We should also be further developing other methods for assisting non-ML experts in PCGML systems. e.g., automatically learning/suggesting data representations; allowing data from similar domains to be used; and deploying models that are more robust to suboptimal setups. 17/n

Sam Snodgrass avatar
Sam Snodgrass@SamPSnodgrass·21 May 2021

So where do we go from here? As researchers & designers of PCGML systems, enabling characterization at each stage is key. Characterizing input data eases model setup. Characterizing the model space eases parameter tuning. Characterizing output eases content eval & selection. 16/n

Sam Snodgrass avatar
Sam Snodgrass@SamPSnodgrass·21 May 2021

For the PCGML systems we develop/deploy at modl.ai, the Content Evaluator role is the external/designer-facing role in our tools. We leverage the ideas above, giving intuitive control over the models and characterizations of generated content to aid the users 15/n

Post attachmentPost attachment