Lukas Gosch (@lukgosch) profile picture

Researcher. Optimization. Graph ML. PhD Student @TU_Muenchen, DAML. relAI Fellow. Quirk for Philosphy. He/Him

Munich
Joined September 2020
Photo by Lukas Gosch (@lukgosch)Photo by Lukas Gosch (@lukgosch)Photo by Lukas Gosch (@lukgosch)Photo by Lukas Gosch (@lukgosch)Photo by Lukas Gosch (@lukgosch)Photo by Lukas Gosch (@lukgosch)Photo by Lukas Gosch (@lukgosch)Photo by Lukas Gosch (@lukgosch)
Excited to announce our #ICLR2025 spotlight work deriving the first exact certificates for neural networks against label poisoning 🎉. Joint work w/ @maha81193, @guennemann & Debarghya. For more details check out the thread below👇 or check out our paper arxiv.org/abs/2412.00537.
🎉Excited to announce our #ICLR2025 Spotlight! 🚀@lukgosch and I will be presenting our paper on the first exact certificate against label poisoning for neural nets and graph neural nets. Joint work with @guennemann and Debarghya 👇[1/6]
14
22
1,185
Lukas Gosch retweeted
Discrete flow matching usually changes one component per edit in a fixed-size state. Can we extend it to general actions, like inserting whole molecular fragments? Introducing Trajectory Matching 🧵 New paper with M. Lienen, @ludke_david, @guennemann
3
24
1
101
5,300
A plot of ICLR papers by country is making the rounds, showing no EU + Japan papers and people are drawing all kinds of conclusions. ..but the plot excludes all (!) EU institutions due to a cutoff. China + USA still dominant of course but the full picture looks a bit different.
someone analyzed all 5000+ accepted papers at ICLR 2026, and it's a good signal who's pushing the research of AI: > China has surpassed the US with 43.7% of the papers > Europe's contribution is surprisingly small (5.3% including UK)
7
62
16
329
68,240
Lukas Gosch retweeted
New preprint: "Transferable SCF-Acceleration through Solver-Aligned Initialization Learning (SAIL)" * Standard ML initial guesses target the converged density. * We target the solver trajectory instead.
4
8
16
621
Very happy to present our work on a principled robustness evaluation of graph transformers at #ICLR2026! You can talk to us at tomorrow's poster session from 10:30 am to 1 pm in Pavilion 3 - #816. Joint work with Philipp Foth, @geisler_si, @SchwinnLeo, and @guennemann.
1
4
9
290
This work was published in TMLR, and I'm very happy it received a J2C-Certification, allowing us to present it at #ICLR! Also, feel free to reach out if you want to talk about graph machine learning in general, or machine learning for combinatorial optimization!
1
37
Lukas Gosch retweeted
Training on sensitive images? The random cropping you're already using is quietly protecting privacy. We formalize and quantify exactly how much. "Amplified Patch-Level Differential Privacy for Free via Random Cropping", published at @TmlrOrg arxiv.org/abs/2603.24695 w/ @SchuchardtJan, @SebastianSch35, @guennemann [1/4]🧵
3
8
14
709
How private is DP-SGD for self-supervised training on sequences? Our #ICML2025 spotlight shows that it can be very private—if you parameterize it right! 📜arxiv.org/abs/2502.02410 #icml Joint work w/ M. Dalirrooyfard, J. Guzelkabaagac, A. Schneider, Y. Nevmyvaka, @guennemann 1/6
2
13
18
1,108
How do LLMs navigate refusal? Our new @ICMLConf paper introduces a gradient-based approach & Representational Independence to map this complex internal geometry. 🚨 New Research Thread! 🚨 The Geometry of Refusal in Large Language Models By @guennemann's lab & @GoogleAI. 🧵👇
1
11
1
29
2,204
📣 Our spicy ICML 2025 position paper: “Graph Learning Will Lose Relevance Due To Poor Benchmarks”. Graph learning is less trendy in the ML world than it was in 2020-2022. We believe the problem is in poor benchmarks that hold the field back - and suggest ways to fix it! 🧵1/10
5
51
12
297
85,066
We will present our work “Multi-Modal and Multi-Attribute Generation of Single Cells with CFGen” at #ICLR2025. Meet us tomorrow at 10 am, Hall 3, poster #17! This is joint work with @TillRichter6 hanyi manuel @AlexanderTong7 @andrea_dittadi @fabian_theis
1
7
1
26
1,898
Happy to share that our paper "Lift Your Molecules: Molecular Graph Generation in Latent Euclidean Space" got accepted to #ICLR2025, and we will be presenting it this week in Singapore! Joint work with @n_gao96, @TomWollschlager, @j_m_sommer, and @guennemann. 🧵 1/
1
18
27
1,839
Congrats to my amazing PhD students: We have 9 papers accepted at #ICLR2025. Reliability, AI4Science, graphs, LLMs, and more (go.tum.de/936150). And if you follow the recent discussions about AI efficiency, you might like our blog and webinars (pruna.ai).
4
14
173
21,174
Lukas Gosch retweeted
📣Announcing VerifAI: AI Verification in the Wild, a workshop at #ICLR2025 VerifAI will gather researchers to explore topics at the intersection of genAI/trustworthyML and verification: verifai-workshop.github.io/ @celine_ylee @theo_olausson @ameeshsh @wellecks @taoyds
21
3
82
30,360
Super happy & honored that our work on certifying NNs against poisoning won the Best Paper Award @AdvMLFrontiers at #NeurIPS2024. Come by our poster 10:40am-12&4-5pm (or talk) tomorrow :) Joint work w/@maha81193, Debarghya Ghoshdastidar & @guennemann L: arxiv.org/pdf/2407.10867
2
14
29
2,444
📢 We are delighted to reveal the Best Paper Awards: 🏆 Can Watermarking Large Language Models Prevent Copyrighted Text Generation and Hide Training Data? and 🏆 Provable Robustness of (Graph) Neural Networks Against Data Poisoning and Backdoor Attacks 👏👏👏
1
2
4
9
7,215
Next week, I'll present our recent paper at NeurIPS 2024 in Vancouver. Many thanks to my amazing collaborators @Bertrand_Charp, @DanielZuegner , and @guennemann!
1
11
19
1,018
Make sure to stop by our #NeurIPS poster on Spatio-Spectral Graph Neural Networks (S²GNNs)! An efficient synergy of spatially & spectrally parametrised graph convolutions. Joint work w/ @geisler_si @dan1elherbst @guennemann. 📎 openreview.net/pdf?id=Cb3kcw… 📆 Dec 13, 11am, #4608
12
22
1,781
Lukas Gosch retweeted
Excited to present our work on Neural Pfaffians at #NeurIPS. 🗣️ Oral: Friday 3:30pm, East Ballroom A, B 📊 Post: Friday 4:30pm - 7:30pm, East Exhibit Hall A-C #3600 📝 Paper: openreview.net/forum?id=HRkn… Happy to chat!
11
1
31
1,160