Luca Eyring
@LucaEyring
Followers
268
Following
3K
Statuses
34
@ELLISforEurope PhD student @ExplainableML
Munich, Germany
Joined October 2022
Can we enhance the performance of T2I models without any fine-tuning? We show that with our ReNO, Reward-based Noise Optimization, one-step models consistently surpass the performance of all current open-source Text-to-Image models within the computational budget of 20-50 sec!
6
30
155
RT @Joshua_Bambrick: ๐ก๐ฒ๐๐ฟ๐๐ฃ๐ฆ ๐ฎ๐ฌ๐ฎ๐ฐ: ๐๐ถ๐ณ๐ณ๐๐๐ถ๐ผ๐ป ๐ง๐ต๐ฒ๐บ๐ฒ๐ ๐ฎ๐ป๐ฑ ๐ ๐ฒ๐บ๐ฒ๐ ๐ The moment you've all been waiting for... I have a blog! The first post sโฆ
0
21
0
Check out our work on how Quadratic Optimal Transport can help enhance Disentangled Representation Learning! Now accepted at #ICLR2025, see you in Singapore :) Full details in the thread below! ๐
Curious about the potential of optimal transport (OT) in representation learning? Join @CuturiMarco's talk at the UniReps workshop today at 2:30 PM! Marco will notably discuss our latest paper on using OT to learn disentangled representations. Details below โฌ๏ธ
0
5
38
RT @Dominik1Klein: Intrigued? Also go check out our follow-up work @NeurIPS ( w @theo_uscidda and @ICLR24 w @LucaEโฆ
0
1
0
RT @Dominik1Klein: Good to see moscot finally published in @Nature! Check out the paper (, the research briefing (hโฆ
0
13
0
RT @confusezius: Great to see this work being accepted at #ICLR2025 - it provides a wonderful new perspective on disentanglement through thโฆ
0
5
0
RT @theo_uscidda: Our work on geometric disentangled representation learning has been accepted to ICLR 2025! ๐See you in Singapore if you wโฆ
0
18
0
RT @theo_uscidda: Curious about the potential of optimal transport (OT) in representation learning? Join @CuturiMarco's talk at the UniRepsโฆ
0
27
0
Thanks to the help of @fffiloni and @natanielruizg, we have a running live Demo of ReNO, play around with it here: ๐ค: Paper (updated with FLUX-Schnell + ReNO results):
1
3
11
RT @zeynepakata: We are looking for two postdoctoral researchers in our @ExplainableML group @TU_Muenchen @HelmholtzMunich @ELLISforEurope,โฆ
0
43
0
RT @zeynepakata: The deadline for PhD applications in our @ExplainableML group @TU_Muenchen @HelmholtzMunich is tomorrow! Checkout the @ELLโฆ
0
4
0
RT @Dominik1Klein: 1/6 Looking for neural estimators of entropic #OptimalTransport or simply cool applications of #FlowMatching? Excited byโฆ
0
15
0
RT @zeynepakata: We are looking for two PhD students via the @ELLISforEurope PhD program fully funded by @TU_Muenchen and @HelmholtzMunich!
0
49
0
RT @fffiloni: Iโve been lowkey working with @LucaEyring and @natanielruizg on a @Gradio demo for ReNO ๐งโ๐ฌ You can try the ReNO @huggingfacโฆ
0
15
0
@RamanDutt4 @zeynepakata This is particularly significant for medical data as it usually requires different feature sets, as e.g. discussed here: Thus, realigning with respect to human-preference rewards might not help on a model that is not trained on any medical data. 2/2
1
0
1
@ShyamgopalKart1 @RisingSayak @ai_ucl For full details on the computational cost, see Table 4 in the paper. I would see ReNO as a more powerful method compared to Attend&Excite (see Table 6 for a comparison) with a similar computational cost, where one also does not need to specify any subject tokens.
1
0
2