Felix Petersen

Machine learning researcher, postdoc. Investigating differentiable algorithms etc. @Stanford @StanfordAILab.

[Felix Petersen @FHKPetersen]

Excited to share our NeurIPS 2024 Oral, Convolutional Differentiable Logic Gate Networks, leading to a range of inference efficiency records, including inference in only 4 nanoseconds 🏎️. We reduce model sizes by factors of 29x-61x over the SOTA.
Paper: https://t.co/c4xFC0SAid

[Felix Petersen @FHKPetersen]

Join us today for Ubiquitous Ultra-Efficient Mobile Video Super-resolution + Denoising with Logic Gates at #ICCV2025 booth 247. 🌺

[Felix Petersen @FHKPetersen]

I'm excited to share that our work on convolutional differentiable logic gate networks was covered by MIT Technology Review. 🎉

https://t.co/Vm6Fdjkwxm
@StefanoErmon @HildeKuehne

[MIT Technology Review @techreview]

The next generation of neural networks could live in hardware https://t.co/cOWLQpa61O

[Alfredo Canziani @alfcnz]

Convolutional Differentiable Logic Gate Networks @FHKPetersen

[Felix Petersen @FHKPetersen]

Join us today for our Oral presentation at 10:20am in East Ballroom A, B and our poster at 11am-2pm in East Exhibit Hall A-C #2411.

[Felix Petersen @FHKPetersen]

Excited to share our NeurIPS 2024 Oral, Convolutional Differentiable Logic Gate Networks, leading to a range of inference efficiency records, including inference in only 4 nanoseconds 🏎️. We reduce model sizes by factors of 29x-61x over the SOTA.
Paper: https://t.co/c4xFC0SAid

[Felix Petersen @FHKPetersen]

Join us now at our poster #1502 in East Exhibit Hall A-C 1502.

[Felix Petersen @FHKPetersen]

Ever wondered how training dynamics differ between LLMs 🖋️ and Vision 👁️ models? We explore this and close the gap between VMs and LLMs in our #NeurIPS2024 paper "TrAct: Making First-layer Pre-Activations Trainable".
Paper📜 https://t.co/6yIeDfTBq8
Video🎥 https://t.co/CyVVHQTWnp

[Felix Petersen @FHKPetersen]

Join us on Wednesday, 11am-2pm for our poster on Newton Losses in West Ballroom A-D #6207. https://t.co/vv6FMwImUQ

[Felix Petersen @FHKPetersen]

I'm excited to share our NeurIPS 2024 paper "Newton Losses: Using Curvature Information for Learning with Differentiable Algorithms" 🤖.
Paper link 📜: https://t.co/rNAF9HQUST
Video link 🎥: https://t.co/fzcWAAYgn4
@StanfordAILab
1/6 🧵

[Felix Petersen @FHKPetersen]

Learn more in our paper 📑 (https://t.co/6yIeDfTBq8) and check out our paper video directly here on 𝕏!

---
Thanks to my co-authors Christian Borgelt and @StefanoErmon.
@StanfordAILab

[Felix Petersen @FHKPetersen]

Ever wondered how training dynamics differ between LLMs 🖋️ and Vision 👁️ models? We explore this and close the gap between VMs and LLMs in our #NeurIPS2024 paper "TrAct: Making First-layer Pre-Activations Trainable".
Paper📜 https://t.co/6yIeDfTBq8
Video🎥 https://t.co/CyVVHQTWnp

[Felix Petersen @FHKPetersen]

...and it speeds up overall training by factors ranging from 1.25x (for large ViT pre-training) to 4x (for ConvNets).
We benchmark TrAct on a suite of 50 experimental settings.

[Felix Petersen @FHKPetersen]

Check out our 5 minute paper video directly here on 𝕏!
6/6 🧵

[Felix Petersen @FHKPetersen]

I'm excited to share our NeurIPS 2024 paper "Newton Losses: Using Curvature Information for Learning with Differentiable Algorithms" 🤖.
Paper link 📜: https://t.co/rNAF9HQUST
Video link 🎥: https://t.co/fzcWAAYgn4
@StanfordAILab
1/6 🧵

[Felix Petersen @FHKPetersen]

Also a shout-out to the methods we build on:
Perturbed Optimizers @qberthet @mblondel_ml @BachFrancis
NeuralSort @adityagrover_ @ecjwg
SoftSort @eisenjulian
5/6 🧵