Diffusion transformers (DiTs) have emerged as a dominant architecture for text-to-image generation, yet their performance drops when generating at resolutions beyond their training range. Existing training-free approaches mitigate this by modifying inference-time attention behavior, often through Rotary Position Embeddings (RoPE) extrapolation combined with attention scaling. However, these strategies apply a uniform and content-agnostic scaling across RoPE components with distinct frequency characteristics, inducing a trade-off between preserving global structure and recovering fine detail. We introduce SEGA, a training-free method that dynamically scales attention across RoPE components according to the latent’s spatial-frequency structure at each denoising step. This adaptive scaling improves both structural coherence and fine-detail fidelity. Experiments show that SEGA consistently improves high-resolution synthesis across multiple target resolutions, outperforming state-of-the-art training-free baselines.
@article{rajabi2026sega,title={{SEGA}: Spectral-Energy Guided Attention for Resolution Extrapolation in Diffusion Transformers},author={Rajabi, Javad and Shaban, Kimia and Roohi, Koorosh and Lindell, David B and Taati, Babak},year={2026},journal={arXiv preprint},}
Preprint
Sizes of witnesses in Covtree
Jette Gutzeit, Kimia Shaban, Karen Yeats, and 1 more author
We examine posets that serve as witnesses to collections of downsets, establishing that there is no linear upper bound of the form n+k+c for any constant c, and introduce the exchange graph of downsets as an analytical tool. We show minimal witnesses satisfy |Q|≤nk-n with improved bounds for specific cases.
@article{gutzeit2026covtree,title={Sizes of witnesses in Covtree},author={Gutzeit, Jette and Shaban, Kimia and Yeats, Karen and Zalel, Stav},year={2026},month=may,journal={arXiv preprint},}
Preprint
When Does RL Help Medical VLMs? Disentangling Vision, SFT, and RL Gains
Ahmadreza Jeddi, Kimia Shaban, Negin Baghbanzadeh, and 4 more authors
We study when reinforcement learning provides meaningful gains over supervised fine-tuning for medical vision-language models, disentangling the contributions of vision encoders, SFT, and RL to model performance.
@article{shaban2026medbridgerl,title={When Does RL Help Medical VLMs? Disentangling Vision, SFT, and RL Gains},author={Jeddi, Ahmadreza and Shaban, Kimia and Baghbanzadeh, Negin and Sharan, Natasha and Moturu, Abhishek and Dolatabadi, Elham and Taati, Babak},year={2026},journal={arXiv preprint},}
2025
Preprint
Statistics and asymptotics of subdivergence-free Feynman integrals in φ^4 theory
Paul-Hermann Balduf, Kimia Shaban, and Johannes Thürigen
We analyze subdivergence-free Feynman integrals in φ⁴ theory up to 18 loops. Statistical analysis reveals exponential average growth accurately described by leading asymptotics only above 25 loops, and importance sampling techniques achieve approximately 1000-fold speedup over uniform sampling.
@article{balduf2025subdivergence,title={Statistics and asymptotics of subdivergence-free {F}eynman integrals in $\phi^4$ theory},author={Balduf, Paul-Hermann and Shaban, Kimia and Th{\"u}rigen, Johannes},year={2025},month=dec,journal={arXiv preprint},}
Thesis
Combinatorial Aspects of Feynman Integrals and Causal Set Theory
This thesis explores combinatorial structures underlying Feynman integrals in φ⁴-theory and connections to causal set theory, including machine learning approaches to period estimation.
@mastersthesis{shaban2025thesis,title={Combinatorial Aspects of {F}eynman Integrals and Causal Set Theory},author={Shaban, Kimia},school={University of Waterloo},year={2025},month=may,}
We apply machine learning to predict Feynman periods in φ⁴-theory, providing new computational tools for estimating the perturbative expansion of quantum field theories.
@article{balduf2024feynman,title={Predicting {F}eynman periods in $\phi^4$-theory},author={Balduf, Paul-Hermann and Shaban, Kimia},journal={Journal of High Energy Physics},year={2024},month=nov,publisher={Springer},doi={10.1007/JHEP11(2024)038},}