t = 0.00

t is the flow matching time step. At t = 0 the dots are pure Gaussian noise. At t = 1 they form a unique rendering of my name. For 0 < t < 1, each dot follows the velocity field (the arrows under your cursor), integrated in 12 Euler steps.

Saeed Khorram

Portrait of Saeed Khorram

/sæˈiːd xorˈræm/

Senior Research Scientist, Apple Tech lead, Visual Generation Foundation Models

Research Interests

Visual Generation and World Modeling
Unified Multimodal Models
Diffusion and Autoregressive Modeling
Representation Learning
Long-tail Learning
Interpretability

← Back

Research

I am a Senior Research Scientist at Apple, where I am tech lead for visual generation foundation models for Apple Intelligence. During my time at Apple, I have worked with Alex Schwing and Lu Jiang on video generation foundation models. On the image side, I have contributed to the scaling of the base model behind ADM 3 Cloud (Apple Diffusion Model), which powers Image Playground and Photos editing tools.

I received my Ph.D. and M.Sc. in Computer Science from Oregon State University, advised by Fuxin Li. During my graduate studies, I worked on generative models, long-tail learning, and interpretability, and interned at Scale AI. I earned my B.Sc. in Electrical Engineering, with a minor in Mathematics, from Amirkabir University of Technology.

My research interests are visual generation and world modeling, unified multimodal models, diffusion and autoregressive modeling, representation learning, long-tail learning, and interpretability.

Ph.D. publications

Taming the Tail in Class-Conditional GANs: Knowledge Sharing via Unconditional Training at Lower Resolutions S. Khorram, M. Jiang, M. Shahbazi, M. H. Danesh, L. Fuxin CVPR 2024 arXiv Code
Comparing the Decision-Making Mechanisms by Transformers and CNNs via Explanation Methods M. Jiang, S. Khorram, L. Fuxin CVPR 2024 Oral, Best Student Paper Runner-up arXiv Project
Diverse Explanations for Object Detectors with Nesterov-Accelerated iGOS++ M. Jiang, S. Khorram, L. Fuxin BMVC 2023 PDF Code
Cycle-Consistent Counterfactuals by Latent Transformations S. Khorram, L. Fuxin CVPR 2022 arXiv Code
Re-Understanding Finite-State Representations of Recurrent Policy Networks M. H. Danesh, A. Koul, A. Fern, S. Khorram ICML 2021 PDF Code
Visualizing Deep Networks by Optimizing with Integrated Gradients Z. Qi, S. Khorram, L. Fuxin AAAI 2020 PDF Demo Code

Full list on Google Scholar

Saeed Khorram, 2026