Erik Härkönen

PhD student at Aalto University, previously Universitat Politècnica de València & 東京大学.

[Nam Hee Gordon Kim @NamHeeGordonKim]

Can we build AI players that are not just great at the game, but play *like people*?

New #Eurographics 2026 paper: Robo-Saber: Generating and Simulating Virtual Reality Players

(Links in the reply!)

[AK @_akhaliq]

Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models

Guidance is a crucial technique for extracting the best performance out of image-generating diffusion models. Traditionally, a constant guidance weight has been applied

[Markus Kettunen @nmkettunen]

Thrilled to announce our work "Conditional Resampled Importance Sampling and ReSTIR" (SIGGRAPH Asia 2023). We extend RIS and ReSTIR into conditional probability spaces, enabling novel forms of sample reuse. 1/7

[Erik Härkönen @harskish]

Huge thanks to Károly for covering our work!
Make sure to play around with our pretrained models - links in the video description :)

[Two Minute Papers @twominutepapers]

New Video - NVIDIA’s New AI: Next-Level Nature Videos!
https://t.co/TX5oR5qq5f

#nvidia #ai

[Ming-Yu Liu @liu_mingyu]

Wondering what next after #StyleGAN-x? Check out this awesome paper from my colleagues on #AliasFreeGAN. It addresses the texture sticking issue in image synthesis and further advances the rendering quality
https://t.co/ZKPQG8YVPb

[Erik Härkönen @harskish]

@arfafax @DabhiLevin @AydaoAI I agree, it does take some work. Maybe generating grids for predefined early/middle/late layer ranges for every component could speed the searching up a bit?

[Erik Härkönen @harskish]

@arfafax @DabhiLevin @AydaoAI Have you tried limiting the layer range at which the components are applied? That usually helps a lot with entanglement!

[Erik Härkönen @harskish]

@ecologician Hi Russell, I'm considering including a still frame of this clip in my NeurIPS presentation on GANSpace, would you be OK with that? I'd include your Twitter handle of course!

[Erik Härkönen @harskish]

@Buntworthy You might be interested in https://t.co/0Cmp54gSos. Figure 2 shows a clustering of channels that correspond to meaningful object parts!

[Erik Härkönen @harskish]

@quasimondo @ak92501 Yep, this is GANSpace at work! 😀

[Erik Härkönen @harskish]

@quasimondo @ak92501 Just manipulating the latent, no first order models here! Still haven't figured out how to add that elusive spark though :D

[Erik Härkönen @harskish]

@quasimondo @ak92501 Sure you can, but they're a bit self-conscious about it 😄

[Erik Härkönen @harskish]

@jongranskog Gratulerar! :)

[Erik Härkönen @harskish]

@dvsch @AaronHertzmann It is!
https://t.co/lcv4AcxwTk