pith. sign in

arxiv: 2606.09928 · v1 · pith:KYNCIZZYnew · submitted 2026-06-07 · 💻 cs.LG · cs.AI

Forward-Only Convolutional Neural Networks with Learnable Channel-Class Assignment

Mohammadnavid Ghader , Saeed Reza Kheradpisheh , Bahar Farahani , Mahmood Fazlali This is my paper

Pith reviewed 2026-06-27 18:42 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords forward-forward algorithmconvolutional neural networkschannel-class assignmentforward-only learningresidual networksimage classificationlocal learning
0
0 comments X

The pith

Learnable channel-class assignment improves forward-forward CNNs on CIFAR-10, CIFAR-100 and Tiny-ImageNet.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper replaces the static channel-to-class partitions used in earlier forward-forward convolutional networks with an assignment that is learned directly from data. Entropy and orthogonality regularization encourage the channels to specialize in distinct ways, while a separate loss-aware weighting step adjusts how much each layer contributes to the final prediction according to its validation accuracy. These two additions are integrated into residual architectures and produce higher classification accuracy than previous forward-only CNNs on the three datasets. The approach also reduces the remaining performance difference relative to networks trained with backpropagation.

Core claim

The introduction of a learnable channel-class assignment mechanism, supported by entropy and orthogonality regularization, together with a loss-aware layer contribution strategy, allows residual forward-forward CNNs to achieve new state-of-the-art results among forward-forward models on CIFAR-10, CIFAR-100, and Tiny-ImageNet.

What carries the argument

Learnable channel-class assignment mechanism that enables adaptive, data-driven specialization of convolutional channels.

If this is right

  • Consistently superior performance across CIFAR-10, CIFAR-100, and Tiny-ImageNet compared to existing forward-only methods.
  • New state-of-the-art performance among FF-based models.
  • Substantial narrowing of the gap with backpropagation-trained models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Adaptive channel specialization may prove necessary for scaling forward-only algorithms to deeper vision networks.
  • The loss-aware weighting idea could transfer to other local learning rules that combine multiple layer predictions.
  • Applying the same learnable assignment to non-residual or non-convolutional forward-only models would test whether the benefit is architecture-specific.

Load-bearing premise

The reported performance gains arise specifically from the learnable channel-class assignment and loss-aware weighting rather than from differences in architecture details, hyperparameters, or training protocol.

What would settle it

Training identical residual CNN architectures on the same datasets with the same protocol but using fixed static channel-class partitions instead of the learnable assignment, then checking whether the accuracy advantage disappears.

Figures

Figures reproduced from arXiv: 2606.09928 by Bahar Farahani, Mahmood Fazlali, Mohammadnavid Ghader, Saeed Reza Kheradpisheh.

Figure 1
Figure 1. Figure 1: Overview of layers and training scheme of DeeperForward [37]. (a) Structure of the CW-Conv block, illustrating grouped channel processing. (b) Training flow of stacked CW-Conv layers, and the local weight updating using layer-wise loss. As discussed in the 2.2, the static assignment of channel groups to different classes, along with their fixed associations throughout the training process, may lead to unde… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed forward-only learning framework for CNNs. (a) Overall architecture of the model, showing stacked CAW-Conv blocks, and the layer contribution strategy used for final decision making. (b) Internal structure of the CAW-Conv block. (c) Training flow of the network, illustrating forward propagation and layer-wise cross-entropy loss optimization. as follows: 𝐺 = 𝐴pooled ⋅ 𝑀, (7) where 𝐴p… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of the effects of entropy (Ent) and orthogonality (Ortho) regularizations (Reg) applied to the learnable channel–class matrix in each CAW-Conv layer, and the influence of the layer contribution strategy (LCS) on final model accuracy on the CIFAR-10 dataset. Compared with other deep FF-based approaches, including DF-R [43], Trifecta [10], and CwComp [33], the proposed model consistently achieves … view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of the rate of accuracy improvement between DeeperForward and our proposed method on test datasets of CIFAR-10, CIFAR-100, and Tiny-ImageNet. Proposed method’s status at the global max accuracy of the DeeperForward method is indicated as a point. CIFAR-100, where DeeperForward exhibits large oscillations and irregular fluctuations. In contrast, proposed method maintains a smoother, more monotoni… view at source ↗
read the original abstract

The Forward-Forward (FF) algorithm offers a biologically inspired alternative to backpropagation by replacing gradient-based credit assignment with local, forward-only objectives. While recent extensions have adapted FF to convolutional neural networks (CNNs), existing formulations rely on static channel-class partitions and struggle to perform effectively in complex tasks. In this work, we introduce a learnable channel-class assignment mechanism that enables adaptive, data-driven specialization of convolutional channels, supported by entropy and orthogonality regularization to promote learning performance. We further propose a loss-aware layer contribution strategy that adaptively weights intermediate-layer predictions based on their validation performance, enhancing the effectiveness of forward-only inference. Integrated into residual CNNs, the proposed method achieves consistently superior performance across CIFAR-10, CIFAR-100, and Tiny-ImageNet compared to existing similar forward-only methods. Notably, it establishes new state-of-the-art performance among FF-based models, substantially narrowing the gap with backpropagation. These findings demonstrate that introducing learnable channel specialization and layer contribution weighting significantly enhances the representational capacity of forward-only learning in deep CNNs.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces a learnable channel-class assignment mechanism for Forward-Forward (FF) convolutional networks, augmented by entropy/orthogonality regularization and a loss-aware weighting of layer contributions. It integrates these into residual CNNs and reports consistent gains over prior FF-CNN baselines on CIFAR-10, CIFAR-100, and Tiny-ImageNet, establishing new state-of-the-art results among FF-based models while narrowing the gap to backpropagation.

Significance. If the performance improvements can be rigorously attributed to the learnable assignment and weighting rather than protocol differences, the work would meaningfully extend the applicability of local, forward-only objectives to deeper CNN architectures and reduce reliance on static partitions that limit prior FF-CNNs.

major comments (2)
  1. [Experiments] Experiments section: the central claim that gains arise specifically from learnable channel-class assignment (rather than residual integration, hyperparameter choices, or training-protocol differences) is not supported by ablations that freeze the assignment mechanism while retaining all other proposed components; without such controls the attribution to the new mechanisms remains insecure.
  2. [Method] § on loss-aware layer contribution: the validation-based weighting is presented as enhancing forward-only inference, yet no quantitative comparison is shown isolating its effect from the channel-assignment module, leaving unclear whether both innovations are load-bearing for the reported SOTA numbers.
minor comments (1)
  1. [Abstract] Abstract supplies no error bars, dataset splits, or implementation specifics; these should be summarized even at high level to allow readers to assess the strength of the empirical claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments highlight the need for stronger controls to attribute performance gains, and we will revise the manuscript accordingly to address both points.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: the central claim that gains arise specifically from learnable channel-class assignment (rather than residual integration, hyperparameter choices, or training-protocol differences) is not supported by ablations that freeze the assignment mechanism while retaining all other proposed components; without such controls the attribution to the new mechanisms remains insecure.

    Authors: We agree that the current experiments do not include ablations that freeze the learnable channel-class assignment while retaining residuals, regularization, and loss-aware weighting. In the revision we will add these controls (e.g., random or static partitions) to isolate the contribution of the learnable assignment and thereby strengthen the attribution of the reported gains. revision: yes

  2. Referee: [Method] § on loss-aware layer contribution: the validation-based weighting is presented as enhancing forward-only inference, yet no quantitative comparison is shown isolating its effect from the channel-assignment module, leaving unclear whether both innovations are load-bearing for the reported SOTA numbers.

    Authors: We acknowledge that an isolated ablation of the loss-aware weighting is missing. The revised manuscript will include a direct comparison (uniform/fixed weights versus the proposed adaptive weighting) while keeping the learnable assignment fixed, to quantify the individual contribution of each component to the final results. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical proposal of mechanisms with reported performance gains.

full rationale

The paper proposes learnable channel-class assignment with regularization and loss-aware weighting for forward-forward CNNs, then reports superior empirical results on CIFAR-10/100 and Tiny-ImageNet versus prior FF methods. No equations, derivations, fitted parameters renamed as predictions, or self-citation chains appear in the provided text. The central claims rest on experimental comparisons rather than any reduction of outputs to inputs by construction, so the work is self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no equations, hyperparameters, or modeling assumptions are visible, so the ledger cannot be populated with concrete entries.

pith-pipeline@v0.9.1-grok · 5730 in / 1095 out tokens · 21516 ms · 2026-06-27T18:42:50.428424+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references · 15 canonical work pages · 3 internal anchors

  1. [1]

    Deep learning without weight transport, in: Advances in Neural Information Processing Systems (NeurIPS)

    Akrout, M., Wilson, C., Humphreys, P., Lillicrap, T., Tweed, D., 2019. Deep learning without weight transport, in: Advances in Neural Information Processing Systems (NeurIPS)

    2019

  2. [2]

    Assessing the scalability of biologically motivated deep learning algorithms, in: Advances in Neural Information Processing Systems (NeurIPS)

    Bartunov, S., Santoro, A., Richards, B., Marris, L., Hinton, G.E., Lillicrap, T., 2018. Assessing the scalability of biologically motivated deep learning algorithms, in: Advances in Neural Information Processing Systems (NeurIPS)

    2018

  3. [3]

    Decoupled greedy learning of cnns, in: International Conference on Machine Learning (ICML)

    Belilovsky, E., Eickenberg, M., Oyallon, E., 2020. Decoupled greedy learning of cnns, in: International Conference on Machine Learning (ICML)

    2020

  4. [4]

    Bengio,Y.,2014.Howauto-encoderscouldprovidecreditassignmentindeepnetworksviatargetpropagation.arXivpreprintarXiv:1407.7906

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  5. [5]

    Learning long-term dependencies with gradient descent is difficult

    Bengio, Y., Simard, P., Frasconi, P., 1994. Learning long-term dependencies with gradient descent is difficult. IEEE Transactions on Neural Networks 5, 157–166

    1994

  6. [6]

    Self-contrastive forward-forward algorithm

    Chen, X., Liu, D., Laydevant, J., Grollier, J., 2025. Self-contrastive forward-forward algorithm. Nature Communications 16, 5978

    2025

  7. [7]

    Unlockingdeeplearning:Abp-freeapproachfor parallelblock-wisetrainingofneuralnetworks,in:ICASSP2024-IEEEInternationalConferenceonAcoustics,SpeechandSignalProcessing, pp

    Cheng,A.,Ping,H.,Wang,Z.,Xiao,X.,Yin,C.,Nazarian,S.,Cheng,M.,Bogdan,P.,2024. Unlockingdeeplearning:Abp-freeapproachfor parallelblock-wisetrainingofneuralnetworks,in:ICASSP2024-IEEEInternationalConferenceonAcoustics,SpeechandSignalProcessing, pp. 4235–4239

    2024

  8. [8]

    Understanding synthetic gradients and decoupled neural interfaces, in: International Conference on Machine Learning (ICML)

    Czarnecki, W.M., Świrszcz, G., Jaderberg, M., Osindero, S., Vinyals, O., Kavukcuoglu, K., 2017. Understanding synthetic gradients and decoupled neural interfaces, in: International Conference on Machine Learning (ICML)

    2017

  9. [9]

    Error-driven input modulation: Solving the credit assignment problem without a backward pass, in: Proceedings of the 39th International Conference on Machine Learning, PMLR

    Dellaferrera, G., Kreiman, G., 2022. Error-driven input modulation: Solving the credit assignment problem without a backward pass, in: Proceedings of the 39th International Conference on Machine Learning, PMLR. pp. 4937–4955

    2022

  10. [10]

    The trifecta: Three simple techniques for training deeper forward-forward networks

    Dooms, T., Tsang, I.J., Oramas, J., 2023. The trifecta: Three simple techniques for training deeper forward-forward networks. arXiv preprint arXiv:2311.18130

    work page arXiv 2023

  11. [11]

    Towards scaling difference target propagation, in: International Conference on Machine Learning (ICML)

    Ernoult, M., Normandin, F., Moudgil, A., Spinney, S., Belilovsky, E., Rish, I., Richards, B., Bengio, Y., 2022. Towards scaling difference target propagation, in: International Conference on Machine Learning (ICML)

    2022

  12. [12]

    Feed-forwardoptimizationwithdelayedfeedbackforneuralnetwork training, in: Neural Information Processing – ICONIP 2024

    Flügel,K.,Coquelin,D.,Weiel,M.,Debus,C.,Streit,A.,Götz,M.,2025. Feed-forwardoptimizationwithdelayedfeedbackforneuralnetwork training, in: Neural Information Processing – ICONIP 2024. Springer, Singapore. volume 15289, pp. 67–78

    2025

  13. [13]

    Ghader, M., Kheradpisheh, S.R., Farahani, B., Fazlali, M., 2024. Enabling privacy-preserving edge ai: Federated learning enhanced with forward-forward algorithm, in: 2024 IEEE International Conference on Omni-layer Intelligent Systems (COINS), pp. 1–7. doi:10.1109/ COINS61597.2024.10622150

    work page arXiv 2024

  14. [14]

    Backpropagation-free spiking neural networks with the forward–forward algorithm

    Ghader, M., Kheradpisheh, S.R., Farahani, B., Fazlali, M., 2026. Backpropagation-free spiking neural networks with the forward–forward algorithm. Scientific Reports 16, 14294. doi:10.1038/s41598-026-41671-4

    work page doi:10.1038/s41598-026-41671-4 2026

  15. [15]

    Understanding the difficulty of training deep feedforward neural networks, in: Proceedings of the International Conference on Artificial Intelligence and Statistics (AISTATS)

    Glorot, X., Bengio, Y., 2010. Understanding the difficulty of training deep feedforward neural networks, in: Proceedings of the International Conference on Artificial Intelligence and Statistics (AISTATS)

    2010

  16. [16]

    Noise-contrastive estimation, in: Proceedings of the International Conference on Artificial Intelligence and Statistics (AISTATS)

    Gutmann, M., Hyvärinen, A., 2010. Noise-contrastive estimation, in: Proceedings of the International Conference on Artificial Intelligence and Statistics (AISTATS). M. Ghader et al.:Preprint submitted to ElsevierPage 14 of 15 Forward-Only Convolutional Neural Networks with Learnable Channel–Class Assignment

    2010

  17. [17]

    The forward-forward algorithm: Some preliminary investigations

    Hinton, G.E., 2022. The forward-forward algorithm: Some preliminary investigations. arXiv preprint arXiv:2212.13345

    work page arXiv 2022

  18. [18]

    Learning and relearning in boltzmann machines, in: Parallel Distributed Processing

    Hinton, G.E., Sejnowski, T.J., 1986. Learning and relearning in boltzmann machines, in: Parallel Distributed Processing. MIT Press

    1986

  19. [19]

    Decoupled parallel backpropagation with convergence guarantee, in: Advances in Neural Information Processing Systems (NeurIPS)

    Huo, Z., Gu, B., Yang, Q., Huang, H., 2018. Decoupled parallel backpropagation with convergence guarantee, in: Advances in Neural Information Processing Systems (NeurIPS)

    2018

  20. [20]

    Decoupled neural interfaces using synthetic gradients, in: International Conference on Machine Learning (ICML)

    Jaderberg, M., Czarnecki, W.M., Osindero, S., Vinyals, O., Graves, A., Silver, D., Kavukcuoglu, K., 2017. Decoupled neural interfaces using synthetic gradients, in: International Conference on Machine Learning (ICML)

    2017

  21. [21]

    Hebbian deep learning without feedback, in: International Conference on Learning Representations (ICLR)

    Journe, A., Rodriguez, H.G., Guo, Q., Moraitis, T., 2023. Hebbian deep learning without feedback, in: International Conference on Learning Representations (ICLR)

    2023

  22. [22]

    Learning multiple layers of features from tiny images

    Krizhevsky, A., 2009. Learning multiple layers of features from tiny images. URL:https://api.semanticscholar.org/CorpusID: 18268744

    2009

  23. [23]

    Imagenet classification with deep convolutional neural networks, in: Advances in Neural Information Processing Systems (NeurIPS)

    Krizhevsky, A., Sutskever, I., Hinton, G.E., 2012. Imagenet classification with deep convolutional neural networks, in: Advances in Neural Information Processing Systems (NeurIPS)

    2012

  24. [24]

    Tinyimagenetvisualrecognitionchallenge

    Le,Y.,Yang,X.S.,2015. Tinyimagenetvisualrecognitionchallenge. URL:https://api.semanticscholar.org/CorpusID:16664790

    2015

  25. [25]

    The mnist database of handwritten digits

    LeCun, Y., Cortes, C., Burges, C.J.C., 1998. The mnist database of handwritten digits. URL:http://yann.lecun.com/exdb/mnist/

    1998

  26. [26]

    Difference target propagation, in: European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD)

    Lee, D.H., Zhang, S., Fischer, A., Bengio, Y., 2015. Difference target propagation, in: European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD)

    2015

  27. [27]

    Symba: Symmetric backpropagation-free contrastive learning with forward-forward algorithm for optimizing convergence

    Lee, H.C., Song, J., 2023. Symba: Symmetric backpropagation-free contrastive learning with forward-forward algorithm for optimizing convergence. arXiv preprint arXiv:2303.08418

    work page arXiv 2023

  28. [28]

    Random synaptic feedback weights support error backpropagation for deep learning

    Lillicrap, T.P., Cownden, D., Tweed, D.B., Akerman, C.J., 2016. Random synaptic feedback weights support error backpropagation for deep learning. Nature Communications 7, 13276

    2016

  29. [29]

    Direct feedback alignment provides learning in deep neural networks, in: Advances in Neural Information Processing Systems (NeurIPS)

    Nøkland, A., 2016. Direct feedback alignment provides learning in deep neural networks, in: Advances in Neural Information Processing Systems (NeurIPS)

    2016

  30. [30]

    The predictive forward-forward algorithm

    Ororbia, A., Mali, A.A., 2023. The predictive forward-forward algorithm. arXiv preprint arXiv:2301.01452

    work page arXiv 2023

  31. [31]

    Contrastive signal–dependent plasticity: Self-supervised learning in spiking neural circuits

    Ororbia, A.G., 2024. Contrastive signal–dependent plasticity: Self-supervised learning in spiking neural circuits. Science Advances 10, eadn6076

    2024

  32. [32]

    Backpropagation-free deep learning with recursive local representation alignment, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp

    Ororbia, A.G., Mali, A., Kifer, D., Giles, C.L., 2023. Backpropagation-free deep learning with recursive local representation alignment, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 9327–9335

    2023

  33. [33]

    Convolutionalchannel-wisecompetitivelearningfortheforward- forward algorithm, in: AAAI Conference on Artificial Intelligence

    Papachristodoulou,A.,Kyrkou,C.,Timotheou,S.,Theocharides,T.,2023. Convolutionalchannel-wisecompetitivelearningfortheforward- forward algorithm, in: AAAI Conference on Artificial Intelligence

    2023

  34. [34]

    Sedona: Search for decoupled neural networks toward greedy block-wise learning, in: International Conference on Learning Representations (ICLR)

    Pyeon, M., Moon, J., Hahn, T., Kim, G., 2021. Sedona: Search for decoupled neural networks toward greedy block-wise learning, in: International Conference on Learning Representations (ICLR)

    2021

  35. [35]

    Learning representations by back-propagating errors

    Rumelhart, D.E., Hinton, G.E., Williams, R.J., 1986. Learning representations by back-propagating errors. Nature 323, 533–536

    1986

  36. [36]

    Hpff:Hierarchicallocallysupervisedlearningwithpatchfeaturefusion

    Su,J.,He,C.,Zhu,F.,Xu,X.,Guan,D.,Si,C.,2024. Hpff:Hierarchicallocallysupervisedlearningwithpatchfeaturefusion. arXivpreprint arXiv:2407.05638

    work page arXiv 2024

  37. [37]

    Deeperforward: Enhanced forward-forward training for deeper and better performance, in: International Conference on Learning Representations (ICLR)

    Sun, L., Zhang, Y., He, W., Wen, J., Shen, L., Xie, W., 2025. Deeperforward: Enhanced forward-forward training for deeper and better performance, in: International Conference on Learning Representations (ICLR)

    2025

  38. [38]

    Forward-forward learning achieves highly selective latent representations for out-of-distribution detection in fully spiking neural networks

    Terres-Escudero, E.B., Del Ser, J., Martínez-Seras, A., Garcia-Bringas, P., 2025. Forward-forward learning achieves highly selective latent representations for out-of-distribution detection in fully spiking neural networks. arXiv preprint arXiv:2407.14097

    work page arXiv 2025

  39. [39]

    Emerging neohebbian dynamics in forward-forward learning: Implications for neuromorphic computing

    Terres-Escudero, E.B., Ser, J.D., Garcia-Bringas, P., 2024. Emerging neohebbian dynamics in forward-forward learning: Implications for neuromorphic computing. arXiv preprint arXiv:2406.16479

    work page arXiv 2024

  40. [40]

    Instance Normalization: The Missing Ingredient for Fast Stylization

    Ulyanov, D., Vedaldi, A., Lempitsky, V.S., 2016. Instance normalization: The missing ingredient for fast stylization. arXiv preprint arXiv:1607.08022

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  41. [41]

    Attention is all you need, in: Advances in Neural Information Processing Systems (NeurIPS)

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., Polosukhin, I., 2017. Attention is all you need, in: Advances in Neural Information Processing Systems (NeurIPS)

    2017

  42. [42]

    Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms

    Xiao, H., Rasul, K., Vollgraf, R., 2017. Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms. ArXiv abs/1708.07747. URL:https://api.semanticscholar.org/CorpusID:702279

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  43. [43]

    Advancing the forward-forward algorithm towards high-performance deep local learning

    Xu, S., Wu, Y., Wu, J., Deng, L., Xu, M., Wen, Q., Li, G., 2026. Advancing the forward-forward algorithm towards high-performance deep local learning. Neural Networks 200, 108765. doi:10.1016/j.neunet.2026.108765

    work page doi:10.1016/j.neunet.2026.108765 2026

  44. [44]

    A theory for the sparsity emerged in the forward forward algorithm

    Yang, Y., 2023. A theory for the sparsity emerged in the forward forward algorithm. arXiv preprint arXiv:2311.05667

    work page arXiv 2023

  45. [45]

    Thecascadedforwardalgorithmforneuralnetworktraining

    Zhao,G.,Wang,T.,Jin,Y.,Lang,C.,Li,Y.,Ling,H.,2025. Thecascadedforwardalgorithmforneuralnetworktraining. PatternRecognition 161, 111292

    2025

  46. [46]

    Understanding why vit trains badly on small datasets: An intuitive perspective

    Zhu, H., Chen, B., Yang, C., 2023. Understanding why vit trains badly on small datasets: An intuitive perspective. arXiv preprint arXiv:2302.03751

    work page arXiv 2023

  47. [47]

    Deep companion learning: Enhancing generalization through historical consistency, in: European Conference on Computer Vision (ECCV)

    Zhu, R., Saligrama, V., 2024. Deep companion learning: Enhancing generalization through historical consistency, in: European Conference on Computer Vision (ECCV). M. Ghader et al.:Preprint submitted to ElsevierPage 15 of 15

    2024