Full search page →

Gradient accumulation when the loss is not a plain mean

Optional · supplementary note 2 of 4 for DLCV 01A · Training Deep Vision Models in Practice. Nothing in the main text depends on it; citation numbers refer to the reference list at the bottom of this page.

Navigate: this page and this track: Navigate: this page and this track

On this page

Supplementary notes · 4

  1. Cosine annealing with warm restarts
  2. Gradient accumulation when the loss is not a plain mean
  3. EMA of weights: choosing the decay
  4. Discriminative learning rates and gradual unfreezing

Paper guides · 2

  1. Accurate, Large Minibatch SGD (Goyal et al.)
  2. Fine-Tuning can Distort Pretrained Features (Kumar et al.)

Deep Learning for Computer Vision

  1. 02 · Deep Image Classification
View the whole course tree →

This note gives the correct weighting when micro-batches contribute different numbers of items to the loss, and lists what else must follow the optimizer step instead of the micro-batch.

Suppose micro-batch jj contains njn_j loss items (valid pixels, matched boxes or tokens) with summed loss Sj=∑i=1njℓijS_j = \sum_{i=1}^{n_j} \ell_{ij}, where ℓij\ell_{ij} is the loss of item ii in micro-batch jj. The loss of the full batch is the mean over all items,

L=∑j=1kSj∑j=1knj,∇θL=∑j=1k∇θSjN,N=∑j=1knj,L = \frac{\sum_{j=1}^{k} S_j}{\sum_{j=1}^{k} n_j}, \qquad \nabla_\theta L = \frac{\sum_{j=1}^{k} \nabla_\theta S_j}{N}, \quad N = \sum_{j=1}^{k} n_j ,

where kk is the number of micro-batches and NN the total item count. Averaging the micro-batch means, 1k∑jSj/nj\tfrac{1}{k}\sum_j S_j / n_j, gives each micro-batch the same weight whatever its size, which is a different objective whenever the njn_j differ (Exercise 4).

Steps. (1) For each micro-batch, compute the summed loss SjS_j and the count njn_j. (2) Backpropagate SjS_j and add the gradients. (3) Before the optimizer step, divide the accumulated gradient by NN. If NN is only known after all micro-batches (it usually is), keep the gradients unscaled until then. Require N>0N>0; an accumulation window with no valid loss items cannot define this mean.

Across devices, check the reduction convention. Let RR be the number of participating processes, NN the global valid-item count and GrG_r the local summed-loss gradient on process rr. A sum reduction gives ∑rGr\sum_r G_r, which must be divided by NN. Default PyTorch DistributedDataParallel averages gradients [48], giving 1R∑rGr\frac{1}{R}\sum_r G_r; multiply that result by R/NR/N to recover the global item mean. Dividing this already averaged gradient by NN would shrink it by an extra factor of RR. Custom communication hooks or uneven-input policies can change the reduction, so verify the convention before applying this factor.

Counters that follow the optimizer step. The learning-rate schedule, weight decay, gradient clipping and the EMA update should all advance once per optimizer step, not once per micro-batch. Clipping must be applied to the accumulated gradient, after the division by NN; clipping each micro-batch separately clips a different quantity.

What accumulation cannot change. Batch normalization still sees only the micro-batch; so do in-batch negatives in contrastive losses, where the number of negatives per example is set by the micro-batch size. Dropout masks differ per micro-batch, which is harmless. With mixed precision and a loss scaler, the PyTorch documentation specifies that the gradients stay scaled and the scale factor stays constant during one accumulation cycle, and that the inf/NaN check, step skipping and scale update happen once per effective batch [47].

References

  1. T. He, Z. Zhang, H. Zhang, Z. Zhang, J. Xie and M. Li, “Bag of tricks for image classification with convolutional neural networks,” in Proc. IEEE/CVF CVPR, 2019. doi
  2. A. Karpathy, “A recipe for training neural networks,” blog post, 2019. link
  3. scikit-learn developers, “load_digits,” scikit-learn documentation, version 1.9.1. link
  4. M. Buda, A. Maki and M. A. Mazurowski, “A systematic study of the class imbalance problem in convolutional neural networks,” Neural Networks, vol. 106, 2018. doi
  5. T.-Y. Lin, P. Goyal, R. Girshick, K. He and P. Dollár, “Focal loss for dense object detection,” in Proc. IEEE ICCV, 2017. doi
  6. A. Buslaev, V. I. Iglovikov, E. Khvedchenya, A. Parinov, M. Druzhinin and A. A. Kalinin, “Albumentations: Fast and flexible image augmentations,” Information, vol. 11, no. 2, 2020. doi
  7. C. Shorten and T. M. Khoshgoftaar, “A survey on image data augmentation for deep learning,” Journal of Big Data, vol. 6, 2019. doi
  8. K. He, X. Zhang, S. Ren and J. Sun, “Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification,” in Proc. IEEE ICCV, 2015. doi
  9. P. Goyal, P. Dollár, R. Girshick, P. Noordhuis, L. Wesolowski, A. Kyrola, A. Tulloch, Y. Jia and K. He, “Accurate, large minibatch SGD: Training ImageNet in 1 hour,” arXiv:1706.02677 (technical report), 2017. arXiv
  10. S. Ioffe and C. Szegedy, “Batch normalization: Accelerating deep network training by reducing internal covariate shift,” in Proc. ICML, PMLR 37, 2015. link
  11. Y. Wu and K. He, “Group normalization,” in Proc. ECCV, 2018. doi
  12. Y. Bengio, “Practical recommendations for gradient-based training of deep architectures,” in Neural Networks: Tricks of the Trade, 2nd ed., LNCS 7700, Springer, 2012. doi
  13. L. N. Smith, “Cyclical learning rates for training neural networks,” in Proc. IEEE WACV, 2017. doi
  14. L. N. Smith, “A disciplined approach to neural network hyper-parameters: Part 1 — learning rate, batch size, momentum, and weight decay,” US Naval Research Laboratory Technical Report 5510-026, arXiv:1803.09820, 2018. arXiv
  15. I. Loshchilov and F. Hutter, “SGDR: Stochastic gradient descent with warm restarts,” in Proc. ICLR, 2017. arXiv
  16. L. N. Smith and N. Topin, “Super-convergence: Very fast training of neural networks using large learning rates,” in Artificial Intelligence and Machine Learning for Multi-Domain Operations Applications, Proc. SPIE 11006, 2019. doi
  17. PyTorch contributors, “torch.optim.lr_scheduler.OneCycleLR,” PyTorch 2.15 documentation, 2026. link
  18. K. He, X. Zhang, S. Ren and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE CVPR, 2016. doi
  19. J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li and L. Fei-Fei, “ImageNet: A large-scale hierarchical image database,” in Proc. IEEE CVPR, 2009. doi
  20. D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” in Proc. ICLR, 2015. arXiv
  21. I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” in Proc. ICLR, 2019. arXiv
  22. PyTorch contributors, “torch.optim.Adam,” PyTorch 2.15 documentation, 2026. link
  23. N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever and R. Salakhutdinov, “Dropout: A simple way to prevent neural networks from overfitting,” Journal of Machine Learning Research, vol. 15, 2014. link
  24. G. Huang, Y. Sun, Z. Liu, D. Sedra and K. Q. Weinberger, “Deep networks with stochastic depth,” in Proc. ECCV, 2016. doi
  25. L. Prechelt, “Early stopping — but when?,” in Neural Networks: Tricks of the Trade, 2nd ed., LNCS 7700, Springer, 2012. doi
  26. V. Godbole, G. E. Dahl, J. Gilmer, C. J. Shallue and Z. Nado, “Deep learning tuning playbook,” Google Research, GitHub repository, 2023. link
  27. B. T. Polyak and A. B. Juditsky, “Acceleration of stochastic approximation by averaging,” SIAM Journal on Control and Optimization, vol. 30, no. 4, 1992. doi
  28. P. Izmailov, D. Podoprikhin, T. Garipov, D. Vetrov and A. G. Wilson, “Averaging weights leads to wider optima and better generalization,” in Proc. UAI, 2018. arXiv
  29. D. Morales-Brotons, T. Vogels and H. Hendrikx, “Exponential moving average of weights in deep learning: Dynamics and benefits,” Transactions on Machine Learning Research, 2024. arXiv
  30. P. Micikevicius, S. Narang, J. Alben, G. Diamos, E. Elsen, D. Garcia, B. Ginsburg, M. Houston, O. Kuchaiev, G. Venkatesh and H. Wu, “Mixed precision training,” in Proc. ICLR, 2018. arXiv
  31. PyTorch contributors, “Automatic mixed precision package — torch.amp,” PyTorch 2.15 documentation, 2026. link
  32. J. Yosinski, J. Clune, Y. Bengio and H. Lipson, “How transferable are features in deep neural networks?,” in Advances in Neural Information Processing Systems 27, 2014. link
  33. E. J. Hu, Y. Shen, P. Wallis, Z. Allen-Zhu, Y. Li, S. Wang, L. Wang and W. Chen, “LoRA: Low-rank adaptation of large language models,” in Proc. ICLR, 2022. arXiv
  34. J. Howard and S. Ruder, “Universal language model fine-tuning for text classification,” in Proc. ACL, 2018. doi
  35. A. Kumar, A. Raghunathan, R. Jones, T. Ma and P. Liang, “Fine-tuning can distort pretrained features and underperform out-of-distribution,” in Proc. ICLR, 2022. arXiv
  36. C. Zhang, S. Bengio, M. Hardt, B. Recht and O. Vinyals, “Understanding deep learning requires rethinking generalization,” in Proc. ICLR, 2017. arXiv
  37. R. Geirhos, J.-H. Jacobsen, C. Michaelis, R. Zemel, W. Brendel, M. Bethge and F. A. Wichmann, “Shortcut learning in deep neural networks,” Nature Machine Intelligence, vol. 2, 2020. doi
  38. PyTorch contributors, “Reproducibility,” PyTorch 2.15 documentation, 2026. link
  39. Y. Xin, J. Yang, S. Luo, Y. Du, Q. Qin, K. Cen, Y. He, Z. Zhang, B. Fu, X. Yang, G. Zhai, M.-H. Yang and X. Liu, “Parameter-efficient fine-tuning for pre-trained vision models: A survey and benchmark,” arXiv:2402.02242, 2024. arXiv
  40. G. E. Dahl, F. Schneider, Z. Nado, N. Agarwal, C. S. Sastry, P. Hennig, S. Medapati, R. Eschenhagen, P. Kasimbeg, D. Suo et al., “Benchmarking neural network training algorithms,” arXiv:2306.07179, 2023. arXiv
  41. P. Kasimbeg, F. Schneider, R. Eschenhagen, J. Bae, C. S. Sastry, M. Saroufim, B. Feng, L. Wright, E. Z. Yang, Z. Nado, S. Medapati, P. Hennig, M. Rabbat and G. E. Dahl, “Accelerating neural network training: An analysis of the AlgoPerf competition,” in Proc. ICLR, 2025. arXiv
  42. A. Defazio, X. A. Yang, H. Mehta, K. Mishchenko, A. Khaled and A. Cutkosky, “The road less scheduled,” in Advances in Neural Information Processing Systems 37, 2024. link
  43. S. Hu, Y. Tu, X. Han, C. He, G. Cui, X. Long, Z. Zheng, Y. Fang, Y. Huang, W. Zhao et al., “MiniCPM: Unveiling the potential of small language models with scalable training strategies,” in Proc. Conference on Language Modeling (COLM), 2024. arXiv
  44. A. Hägele, E. Bakouch, A. Kosson, L. Ben Allal, L. Von Werra and M. Jaggi, “Scaling laws and compute-optimal training beyond fixed training durations,” in Advances in Neural Information Processing Systems 37, 2024. link
  45. M. Wortsman, G. Ilharco, J. W. Kim, M. Li, S. Kornblith, R. Roelofs, R. Gontijo Lopes, H. Hajishirzi, A. Farhadi, H. Namkoong and L. Schmidt, “Robust fine-tuning of zero-shot models,” in Proc. IEEE/CVF CVPR, 2022. doi
  46. M. Wortsman, G. Ilharco, S. Y. Gadre, R. Roelofs, R. Gontijo-Lopes, A. S. Morcos, H. Namkoong, A. Farhadi, Y. Carmon, S. Kornblith and L. Schmidt, “Model soups: Averaging weights of multiple fine-tuned models improves accuracy without increasing inference time,” in Proc. ICML, PMLR 162, 2022. link
  47. PyTorch contributors, “Automatic mixed precision examples,” PyTorch 2.15 documentation, 2026. link
  48. PyTorch contributors, “DistributedDataParallel,” PyTorch 2.15.0 source documentation, 2026. Official source.