@[email protected]

Ass’t Prof. of Computer Science at Pomona. Videogames, formal methods, symbolic AI. As a rule, I won’t follow current 5C students.
He/him.
Black Lives Matter.

[@[email protected] @JoeOsborn]

@zarawesome The fact that it’s techbros of all possible groups of people with this belief… who could look at web dev or C and think “throw stuff at the wall and see what sticks” could fix racism or whatever? A whole philosophical edifice of “well if you just…”

[@[email protected] @JoeOsborn]

I'm feeling pretty good about my work on adding accessibility features to retro games (for now, NES) with a minimum of fuss. I'll hopefully have some screenshots for the next post.
https://t.co/NaRSI3i6Kw

[@[email protected] @JoeOsborn]

@exag20xx Will exag sessions be streamed by any chance?

[Irene Zhang @schemeprincess]

@CompArchSA The most important barriers to grad school that I’ve heard are not knowing: 1. Grad school is paid for and you get a small stipend! You do not have to pay for a PhD; and 2. You can get fee waivers from American schools. The way to apply varies. Ask grad students for help.

[Lara J. Martin | @laramartin.net @LangTechLara]

Hiya! I'm a #CIFellow postdoc at UPenn, and I'm on the faculty job market this year! I do human-centered #AI with a focus on language technologies/#NLProc. I've worked on automated story generation (think AI #DnD dungeon master), speech processing, and improving #AAC apps.

[Moola @MoolaMixtape]

"screw it," he said, "lets make squiddy holly real"
(feat. harmony)

[@[email protected] @JoeOsborn]

Melissa pointed out that an important part of a research journal is showing that I am a whole person who exists outside of my research output. I hope students understand that they matter and that their health matters.

https://t.co/kyEI9LZruP

[@[email protected] @JoeOsborn]

@BillB I feel better about my unread copies now. Thank you!

[@[email protected] @JoeOsborn]

Happy Monday, debugging convergence issues in neural networks is a pain. Overfitting would be a great problem for me to have.

https://t.co/XkGP9Txyid

[@[email protected] @JoeOsborn]

If we can overfit on four examples, maybe we can overfit on 130? Today I got to re-familiarize myself with convolutional neural networks, loading datasets, and running ML experiments.

https://t.co/kx0Fpo4DtP

[@[email protected] @JoeOsborn]

@rndmcnlly Yeah, it’s about ten or a hundred times slower on CPU, but I didn’t do any model compilation or anything. The model itself is pretty wide and shallow, the first thing I’ll throw away tomorrow (no convolutions or patches).
I think eventually this will be a piece of a larger model?

[@[email protected] @JoeOsborn]

Today I convinced myself that a simple ML model for 2D game camera movement can overfit to four data points, a fact which is both useless in a general sense and extremely important as evidence that my data processing and pipeline are working.

https://t.co/KF3fmAy0k1

[@[email protected] @JoeOsborn]

Once you've got a research direction in mind, you need to do some feasibility checking. Is it possible to get the data we need? Is all our code from a year ago still working? Is it ergonomic to use? It's hard to get it right, but I hate to re-do work because of data bugs.

[@[email protected] @JoeOsborn]

While thinking about neural network architectures, I got into the weeds to get ready to collect the data I'll need for the most straightforward next steps for bootstrapping Mappy towards working purely from vision.
https://t.co/KiIDO5pyil

[@[email protected] @JoeOsborn]

Pomona College computer science is hiring at all ranks! We have a very collegial department with a focus on teaching, solid research support, superb undergrads, and great benefits. I’d love to chat about my experience and answer questions.

https://t.co/7RSdKrJhGc

[@[email protected] @JoeOsborn]

I know very little about accessibility except that I've seen these features added to games and they seem to make sense. I'm looking for feedback for what would be useful in making, e.g. The Legend of Zelda or Metroid playable for someone for whom these artifacts are inaccessible.

[@[email protected] @JoeOsborn]

I did another one of these.

While more and more games made today are implemented with accessibility features, so many games of major cultural import will never be patched to include such features. So, how can we hack existing games to add them?

https://t.co/IXeAcwamNc

[@[email protected] @JoeOsborn]

Mappy is this thing, it's a feature extraction deal for NES games that is just super useful:

https://t.co/atRHic8Ijk

https://t.co/8U1OF54TCp

You run it on NES games and get live data about the terrain, sprites on screen (and over time), camera movement, tilesheets, and more.

[@[email protected] @JoeOsborn]

Oh yeah, I’m sure the post is missing tons of context. Happy to elaborate on anything!

[@[email protected] @JoeOsborn]

I never know what to put here, so at my wife Melissa's suggestion I'll be posting up summaries of my research process on days I do that. Today I sample some CV papers and think about applying them to reverse engineering game designs from examples of play.

https://t.co/ufA0pnD3nD