When Will Gradient Regularization Be Harmful?

2024-06-14 · Yang Zhao, Hao Zhang, Xiuyuan Hu

Gradient regularization (GR), which aims to penalize the gradient norm atop the loss function, has shown promising results in training modern over-parameterized deep neural networks. However, can we trust this powerful technique? This paper reveals that GR can cause performance degeneration in adaptive optimization scenarios, particularly with learning rate warmup. Our empirical and theoretical analyses suggest this is due to GR inducing instability and divergence in gradient statistics of adaptive optimizers at the initial training stage. Inspired by the warmup heuristic, we propose three GR warmup strategies, each relaxing the regularization effect to a certain extent during the warmup course to ensure the accurate and stable accumulation of gradients. With experiments on Vision Transformer family, we confirm the three GR warmup strategies can effectively circumvent these issues, thereby largely improving the model performance. Meanwhile, we note that scalable models tend to rely more on the GR warmup, where the performance can be improved by up to 3\% on Cifar10 compared to baseline GR. Code is available at \href{https://github.com/zhaoyang-0204/gnp}{https://github.com/zhaoyang-0204/gnp}.

📄 PDF Abstract BibTeX arXiv:2406.09723

Code (1)

zhaoyang-0204/gnp 공식 구현 jax

Methods 이 논문이 사용한 방법론

Attention 설명 없음

Residual Connection 설명 없음

Softmax The Softmax output function transforms a previous layer's output into a vector of probabilities. It is commonly used for multiclass classification. Given an input vector $x$…

Layer Normalization Unlike batch normalization, Layer Normalization directly estimates the normalization statistics from the summed inputs…

BPE Byte Pair Encoding, or BPE, is a subword segmentation algorithm that encodes rare and unknown words as sequences of subword units. The intuition is that various word…

Label Smoothing Label Smoothing is a regularization technique that introduces noise for the labels. This accounts for the fact that datasets may have mistakes in them, so maximizing the…

Adam 설명 없음

Linear Layer A Linear Layer is a projection $\mathbf{XW + b}$.

Defending Against Harmful Supervision Hidden in Benign Samples

2026-06-29 · Bang An, Yibo Yang, Dandan Guo, Ebtisam Alshehri 외 arxiv

Existing defenses are effective when harmful content is explicitly mixed into downstream fine-tuning data, but crafted samples can instead hide harmful supervision inside benign tasks. We propose Embedded Attack, where h…

Safety Anchor: Defending Harmful Fine-tuning via Geometric Bottlenecks

2026-05-07 · Guoxin Lu, Letian Sha, Qing Wang, Peijie Sun 외 arxiv

The safety alignment of Large Language Models (LLMs) remains vulnerable to Harmful Fine-tuning (HFT). While existing defenses impose constraints on parameters, gradients, or internal representations, we observe that they…

FlexiGrad: Adaptive Gradient Modulation for Hierarchical Fine-Grained Classification

2026-07-20 · Zilu Zhou, Dongliang Chang, Junhan Chen, Zhanyu Ma arxiv

Many fine-grained recognition tasks contain hierarchical labels such as order, family and species. Although this supervision should be beneficial, jointly optimising all levels often leads to unstable training because co…

Antibody: Strengthening Defense Against Harmful Fine-Tuning for Large Language Models via Attenuating Harmful Gradient Influence

2026-02-28 · Quoc Minh Nguyen, Trung Le, Jing Wu, Anh Tuan Bui 외 arxiv

Fine-tuning-as-a-service introduces a threat to Large Language Models' safety when service providers fine-tune their models on poisoned user-submitted datasets, a process known as harmful fine-tuning attacks. In this wor…

DISTRIBUTIONAL CONCAVITY REGULARIZATION FOR GANS

2019-05-01 · ICLR 2019 5 · Shoichiro Yamaguchi, Masanori Koyama

We propose Distributional Concavity (DC) regularization for Generative Adversarial Networks (GANs), a functional gradient-based method that promotes the entropy of the generator distribution and works against mode collap…