Seminar Machine Learning in Microscopy WS 26/27

Github Code

Studon Link

Pelz Lab website

Course overview

The seminar meets on Wednesdays, 16:00–17:30, in IZNF room 01.140. The first session is on 28.10.2026. There is no session on 21.10.2026, 23.12.2026, 30.12.2026 and 06.01.2027.

Slides, the topic sign-up and the forum are on StudOn.

Prerequisite: the lecture Data Science for Electron Microscopy is highly recommended.

Requirements

• Give one 30-minute seminar talk on a chosen topic. The talk is graded.
• Attend all journal club meetings and at least 4 seminar talk meetings.
• Present in the journal club meetings. These presentations are casual and not graded.
• The seminar has 2 ECTS, which is about 60 hours of work.

Journal club. In 5 journal club meetings, we study and discuss the papers that lead up to the seminar topics. Go through a paper and explain it in your own words, in a casual style (example). The journal club is your practice for the seminar talk.

Grading of the seminar talk:

Criterion	Weight	What counts
Visualization	30%	One message per slide, large text, figures instead of bullet lists
Know your audience	30%	The audience is not expert researchers: explain the underlying concepts
Timing	10%	Stay within ±2 minutes of 30 minutes
Fluency and logic	30%	A clear story from motivation to summary

A talk that is more than 2 minutes too long or too short loses 0.3 grade points; more than 4 minutes loses 0.6.

Schedule

Date	Session	Homework
28.10.2026	Info & topics	Book a topic on StudOn by 03.11.2026
04.11.2026	Journal club 1
11.11.2026	Journal club 2
18.11.2026	Journal club 3
25.11.2026	Journal club 4
02.12.2026	Journal club 5
09.12.2026	Seminar talks 1 + 2
16.12.2026	Seminar talks 3 + 4
23.12.2026 – 06.01.2027	Winter break, no session
13.01.2027	Seminar talks 5 + 6
20.01.2027	Seminar talks 7 + 8
27.01.2027	Seminar talks 9 + 10
03.02.2027	Buffer

Topics

Each topic has one or two papers. Book your topic on StudOn; each topic can be taken once.

1. Convolutional neural networks for segmentation
   • Rangel DaCosta, L. et al. (2024). A robust synthetic data generation framework for machine learning in high-resolution transmission electron microscopy (HRTEM). npj Computational Materials 10, 1–11. doi:10.1038/s41524-024-01336-0
   • Sytwu, K., Groschner, C. & Scott, M. C. (2022). Understanding the influence of receptive field and network complexity in neural network-guided TEM image analysis. Microscopy and Microanalysis 28, 1896–1904. doi:10.1017/S1431927622012466
2. CNNs and rotationally invariant VAEs: deep Bayesian local crystallography
   • Kalinin, S. V. et al. (2021). Deep Bayesian local crystallography. npj Computational Materials 7, 1–12. doi:10.1038/s41524-021-00621-6
3. CNNs in tomographic image reconstruction
   • Wu, Z. et al. (2023). Three-dimensional nanoscale reduced-angle ptycho-tomographic imaging with deep learning (RAPID). eLight 3, 7. doi:10.1186/s43593-022-00037-9
   • Wu, Z. et al. (2020). Deep learning-based low-dose tomography reconstruction with hybrid-dose measurements. IEEE/ACM AI4S Workshop, 88–95. doi:10.1109/MLHPCAI4S51975.2020.00017
4. CNNs for denoising: DivNoising
   • Prakash, M., Krull, A. & Jug, F. (2020). DivNoising: diversity denoising with fully convolutional variational autoencoders. arXiv:2006.06072
5. Untrained neural networks for denoising
   • Heckel, R. & Hand, P. (2019). Deep Decoder: concise image representations from untrained non-convolutional networks. arXiv:1810.03982
   • Heckel, R. & Soltanolkotabi, M. Denoising and regularization via exploiting the structural bias of convolutional generators. ICLR. openreview
6. Untrained and physics-based neural networks
   • Bostan, E. et al. (2020). Deep phase decoder: self-calibrating phase microscopy with an untrained deep neural network. Optica 7, 559–562. doi:10.1364/OPTICA.389314
   • Monakhova, K. et al. (2021). Untrained networks for compressive lensless photography. Optics Express 29, 20913–20929. doi:10.1364/OE.424075
7. Implicit neural representations
   • Mildenhall, B. et al. (2020). NeRF: representing scenes as neural radiance fields for view synthesis. arXiv:2003.08934
   • Kerbl, B. et al. (2023). 3D Gaussian splatting for real-time radiance field rendering. SIGGRAPH 2023. project page
8. Transformers in image reconstruction
   • Liang, J. et al. (2021). SwinIR: image restoration using Swin Transformer. doi:10.48550/arXiv.2108.10257
   • Liu, Z. et al. (2021). Swin Transformer: hierarchical vision transformer using shifted windows. ICCV 2021, 10012–10022. paper
9. Physics-based neural networks for atomic electron tomography
   • Ren, D., Ophus, C., Chen, M. & Waller, L. (2020). A multiple scattering algorithm for three dimensional phase contrast atomic electron tomography. Ultramicroscopy 208, 112860. doi:10.1016/j.ultramic.2019.112860
10. Diffusion models for image reconstruction
    • Chung, H. et al. (2022). Diffusion posterior sampling for general noisy inverse problems. doi:10.48550/arXiv.2209.14687
    • Liu, J. et al. (2022). DOLCE: a model-based probabilistic diffusion framework for limited-angle CT reconstruction. doi:10.48550/arXiv.2211.12340
11. Atomic-resolution coherent X-ray imaging with physics-based phase retrieval
    • Meziere, J. et al. (2024). Atomic resolution coherent x-ray imaging with physics-based phase retrieval. npj Computational Materials 10, 1–11. doi:10.1038/s41524-024-01340-4
12. Unsupervised deep denoising for 4D-STEM
    • Sadri, A. et al. (2024). Unsupervised deep denoising for four-dimensional scanning transmission electron microscopy. npj Computational Materials 10, 1–13. doi:10.1038/s41524-024-01428-x
13. Few-shot and zero-shot segmentation
    • Akers, S. et al. (2021). Rapid and flexible segmentation of electron microscopy data using few-shot machine learning. npj Computational Materials 7, 1–9. doi:10.1038/s41524-021-00652-z
    • Abebe, W. et al. (2025). SAM-I-Am: semantic boosting for zero-shot atomic-scale electron micrograph segmentation. Computational Materials Science 246, 113400. doi:10.1016/j.commatsci.2024.113400

Archive

The SS 25 version of this seminar had the same format: an intro session, journal clubs, then seminar talks.