Research map / Probabilistic and scientific modeling / Dynamical systems
Log-Normal Multiplicative Dynamics for Stable Low-Precision Deep Learning
Keigo Nishida, Eren Mehmet KIRAL, Kenichi Bannai, Mohammad Emtiyaz Khan, Thomas Möllenhoff
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace AdamW with LMD | LMD · ViT ImageNet, from scratch | test top-1 accuracy: 68.11 -> 77.06 (bfloat16); 67.98 -> 77.15 (MXFP6) | Improved |
| replace AdamW with LMD, sequence length 4096 | LMD · GPT-2 OpenWebText | validation loss: 2.980 -> 2.927 (MXFP6) | Improved |
| replace AdamW with LMD, sequence length 1024 | LMD · GPT-2 OpenWebText | validation loss: 2.937 -> 2.961 (bfloat16) | Hurt |
| MXFP6 forward pass instead of bfloat16 for LMD | LMD · ViT ImageNet | test top-1 accuracy: 77.06 -> 77.15 | No clear effect |
| MXFP6 forward pass instead of bfloat16 for AdamW | AdamW · GPT-2 seq len 1024 | validation loss: 2.937 -> 3.015 | Hurt |
| MXFP6 forward pass for Madam | Madam · ViT ImageNet | test top-1 accuracy: 60.14 -> 13.47 | Hurt |
| use multiplicative instead of additive weight decay (ablation) | LMD · ViT and GPT-2 | weight norm / stability | Improved |
| scale gradient by weight magnitude (ablation) | LMD · ViT and GPT-2 | weight norm fluctuation | Improved |
| sample weights (multiplicative noise) instead of using the mean in training (ablation) | LMD · ViT ImageNet | accuracy | Improved |
| sample weights instead of mean with MXFP6 forward (ablation) | LMD · GPT-2 | final validation performance | No clear effect |
Closest papers by meaning
- Unconditional stability of a recurrent neural circuit implementing divisive normalization (NeurIPS 2024 · similarity 0.85)
- Improving Deep Learning Speed and Performance Through Synaptic Neural Balance (AAAI 2025 · similarity 0.85)
- Inferring stochastic low-rank recurrent neural networks from neural data (NeurIPS 2024 · similarity 0.84)
- Brain-like Variational Inference (NeurIPS 2025 · similarity 0.84)
- A Unified Stability Analysis of SAM vs SGD: Role of Data Coherence and Emergence of Simplicity Bias (NeurIPS 2025 · similarity 0.84)
- Deep Linear Network Training Dynamics from Random Initialization: Data, Width, Depth, and Hyperparameter Transfer (ICML 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
