Max Zimmer
@maxzimmerberlin
Followers
148
Following
914
Media
28
Statuses
308
ML Research Lead @ IOL Lab @ZuseInstitute | PhD Student at TU Berlin | Interested in Deep Learning, NN Efficiency and Mathematics
Berlin
Joined May 2023
🎉Our paper "Neural Discovery in Mathematics: Do Machines Dream of Colored Planes?" was selected for an ORAL (top 1%) presentation at #ICML25! 🎉. arXiv: blog: 🧵: coming soon
1
2
6
RT @darrenangle: *sniff* *pulls shirt* You know, this is perfect - *gestures wildly* - this is the ultimate perversity of capitalism at its….
0
526
0
RT @kirill_bykov: Personal news: I have defended my PhD thesis “Explaining Representations in Deep Neural Networks” at TU Berlin with summa….
0
2
0
RT @mikelasby: We will be presenting Self-Distilled Sparse Drafters at #ICML this afternoon at the Efficient Systems for Foundation Models….
0
4
0
Generate videos in just a few seconds. Try Grok Imagine, free for a limited time.
297
106
2K
3 posters at #ICML2025 tomorrow 11:00-13:30 🌞. 1⃣ E-2303: Approximating Latent Manifolds in NNs via Vanishing Ideals.2⃣ W-220: Capturing Temporal Dynamics in Large-Scale Canopy Tree Height Estimation.3⃣ W-110: DUNIA: Pixel-Sized Embeddings via Cross-Modal Alignment for EO
2
0
4
RT @lorenz_richter: Excited for #ICML2025 in Vancouver! On Thursday morning, I'm presenting our paper ( on a critic….
0
15
0
Will be at #ICML2025 in Vancouver, feel free to reach out!.
🌟 Join our Team in Berlin 🌟. We are seeking highly motivated Deep Learning PhD students to work on Efficiency, AgenticAI, AI4Math, etc, preferably with strong math/CS background and PyTorch experience. Happy to answer questions here, via DM or at #ICLR2025!
2
0
8
RT @mikelasby: Want faster #LLM inference without sacrificing accuracy? 🚀. Introducing SD², a novel method using Self-Distilled Sparse Draf….
0
9
0
Thread:
Our #AI4Science paper "Capturing Temporal Dynamics in Canopy Tree Height" is accepted at #ICML2025! We present the first 10m temporal map for Europe (2019-22), improving tall tree estimation & tracking change. Blog: Paper: 🧵 1/7
0
0
0
In summary:.1️⃣ SotA 2020 Europe map.2️⃣ Better tall tree estimation.3️⃣ Temporal tracking (2019-2022). Joint work w/ @jan_pauls_, @BerkantTuran_, S. Saatchi, @ciais_philippe, @spokutta & F. Gieseke.
0
0
0
3⃣ Applying the model across 2019-2022 enables tracking changes like deforestation and regrowth, offering valuable insights into temporal forest dynamics. 🧵 6/7
1
0
0
2⃣ We significantly improve predictions for TALL trees (>30m), often underestimated previously. More accurate heights lead to better carbon stock estimates & help identify old-growth forests. 🧵 5/7
1
0
0
The results: .1⃣ SotA performance for Europe in 2020. Our model outperforms previous ones (MAE ↓13%, MSE ↓11%), yields more precise forest structure & edge alignment. Our map shows enhanced detail and better differentiation between forest patches of varying heights. 🧵 4/7
1
0
0
Instead of using annual composites, we process monthly Sentinel-2 time series with a 3D U-Net. This leverages satellite position shifts for richer structural info and improved accuracy. 🧵 3/7.
1
0
0
While our ICML24 work provided a global 2020 map, static snapshots miss crucial forest dynamics. Further, we were still struggling with tall tree estimation. Read Part 1 of the blog post series for more information: 🧵 2/7.
1
0
0