# NOBLE: Accelerating Transformers with Nonlinear Low-Rank Branches

arXiv: https://arxiv.org/abs/2603.06492 · HTML: https://mlsanity.com/paper/2603.06492/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 27 of its 34 references that are in the map.
- First posted on arXiv 6 Mar 2026.

## Builds on (most cited first, 27 of 27)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Rethinking the Inception Architecture for Computer Vision](https://mlsanity.com/paper/1512.00567/) (2015)
- [mixup: Beyond Empirical Risk Minimization](https://mlsanity.com/paper/1710.09412/) (2017)
- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [Parameter-Efficient Transfer Learning for NLP](https://mlsanity.com/paper/1902.00751/) (2019)
- [Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer](https://mlsanity.com/paper/1701.06538/) (2017)
- [QLoRA: Efficient Finetuning of Quantized LLMs](https://mlsanity.com/paper/2305.14314/) (2023)
- [Taming Transformers for High-Resolution Image Synthesis](https://mlsanity.com/paper/2012.09841/) (2020)
- [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://mlsanity.com/paper/2101.03961/) (2021)
- [Implicit Neural Representations with Periodic Activation Functions](https://mlsanity.com/paper/2006.09661/) (2020)
- [Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention](https://mlsanity.com/paper/2006.16236/) (2020)
- [GLU Variants Improve Transformer](https://mlsanity.com/paper/2002.05202/) (2020)
- [Spectral Normalization for Generative Adversarial Networks](https://mlsanity.com/paper/1802.05957/) (2018)
- [Rethinking Attention with Performers](https://mlsanity.com/paper/2009.14794/) (2020)
- [Root Mean Square Layer Normalization](https://mlsanity.com/paper/1910.07467/) (2019)
- [On the Spectral Bias of Neural Networks](https://mlsanity.com/paper/1806.08734/) (2018)
- [DoRA: Weight-Decomposed Low-Rank Adaptation](https://mlsanity.com/paper/2402.09353/) (2024)
- [LoRA+: Efficient Low Rank Adaptation of Large Models](https://mlsanity.com/paper/2402.12354/) (2024)
- [AdaLoRA: Adaptive Budget Allocation for Parameter-Efficient Fine-Tuning](https://mlsanity.com/paper/2303.10512/) (2023)
- [Tensor Programs V: Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer](https://mlsanity.com/paper/2203.03466/) (2022)
- [Primer: Searching for Efficient Transformers for Language Modeling](https://mlsanity.com/paper/2109.08668/) (2021)
- [Mixture-of-Subspaces in Low-Rank Adaptation](https://mlsanity.com/paper/2406.11909/) (2024)
- [Explicit Regularisation in Gaussian Noise Injections](https://mlsanity.com/paper/2007.07368/) (2020)
- [AuroRA: Breaking Low-Rank Bottleneck of LoRA with Nonlinear Mapping](https://mlsanity.com/paper/2505.18738/) (2025)

## Abstract (arXiv)

We introduce NOBLE (Nonlinear lOw-rank Branch for Linear Enhancement), an architectural augmentation that adds nonlinear low-rank branches to transformer linear layers. Unlike LoRA and other parameter-efficient fine-tuning (PEFT) methods, NOBLE is designed for pretraining from scratch. The branch is a permanent part of the architecture as opposed to an adapter for finetuning on top of frozen weights. The branch computes σ(xWdown)Wup where σ is a learnable nonlinearity. We evaluate several activation functions and find that CosNet, a two-layer cosine nonlinearity with learnable frequency and phase with a linear projection in between them in the bottleneck space, performs best. NOBLE achieves substantial improvements with minimal overhead: up to 1.47x step speedup to reach baseline eval loss (up to 32% fewer training steps), with as low as 4% additional parameters and 7% step time overhead, resulting in up to 1.22x net wallclock speedup. Experiments on LLMs (250M and 1.5B parameters), BERT, VQGAN, and ViT consistently show improved training efficiency. We identify one caveat: Mixup/CutMix augmentation interferes with NOBLE's benefits in Imagenet classification along with other stochastic augmentations, but when disabled, ViT also improves. This discrepancy is possibly explained by regularization techniques that encourage smoother fits to the target function while NOBLE may specialize more in sharper aspects of the target function.