Profiles2020-2025
Daniel Soudry
Department of Electrical and Computer Engineering -Technion
20Papers
15Last author
46Unique co-authors
0First author
0Solo
First and last author come from byline position, which some venues assign alphabetically.
Recent papers20
- FP4 All the Way: Fully Quantized Training of Large Language ModelsAuthor 4 of 5NeurIPS 2025 · 9 citations
- Tensor-Parallelism with Partially Synchronized ActivationsAuthor 5 of 6NeurIPS 2025 · 6 citations
- Alias-Free ViT: Fractional Shift Invariance via Linear AttentionAuthor 2 of 2NeurIPS 2025 · 2 citations
- Scaling FP8 training to trillion-token LLMsAuthor 4 of 5ICLR 2025 · 1 citation
- Stable Minima Cannot Overfit in Univariate ReLU Networks: Generalization by Large Step SizesAuthor 4 of 5NeurIPS 2024 · 15 citations
- The Implicit Bias of Gradient Descent on Separable Multiclass DataAuthor 3 of 4NeurIPS 2024 · 14 citations
- Towards Cheaper Inference in Deep Networks with Lower Bit-Width AccumulatorsAuthor 2 of 2ICLR 2024 · 5 citations
- How do Minimum-Norm Shallow Denoisers Look in Function Space?Author 4 of 4NeurIPS 2023 · 11 citations
- Accurate Neural Training with 4-bit Matrix Multiplications at Standard FormatsAuthor 6 of 6ICLR 2023 · 6 citations
- DropCompute: simple and more robust distributed synchronous training via compute variance reductionAuthor 9 of 9NeurIPS 2023 · 4 citations
- Alias-Free Convnets: Fractional Shift Invariance via Polynomial ActivationsAuthor 3 of 3CVPR 2023
- Minimum Variance Unbiased N: M Sparsity for the Neural GradientsAuthor 4 of 4ICLR 2023
- A Statistical Framework for Efficient Out of Distribution Detection in Deep Neural NetworksAuthor 4 of 4ICLR 2022 · 40 citations
- Accelerated Sparse Neural Training: A Provable and Efficient Method to Find N: M Transposable MasksAuthor 6 of 6NeurIPS 2021 · 148 citations
- The Implicit Bias of Minima Stability: A View from Function SpaceAuthor 3 of 3NeurIPS 2021 · 65 citations
- Physics-Aware Downsampling with Deep Learning for Scalable Flood ModelingAuthor 5 of 5NeurIPS 2021 · 12 citations
- Neural gradients are near-lognormal: improved quantized and sparse trainingAuthor 6 of 6ICLR 2021 · 5 citations
- Implicit Bias in Deep Linear Classification: Initialization Scale vs Training AccuracyAuthor 6 of 6NeurIPS 2020 · 98 citations
- At Stability's Edge: How to Adjust Hyperparameters to Preserve Minima Selection in Asynchronous Training of Neural Networks?Author 4 of 4ICLR 2020 · 25 citations
- Augment Your Batch: Improving Generalization Through Instance RepetitionAuthor 6 of 6CVPR 2020
