Pith. sign in

REVIEW 4 major objections 4 minor 94 references

Accelerating Learned Image Compression Through Modeling Neural Training Dynamics

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Training a learned image codec can run in about 62% of the usual wall-clock time, with the same final rate–distortion quality, when most weights are locked as affine copies of a few reference weights.

desk verdict Real training-time savings, but the paper's own mechanism is undercut by a missing SMA-only control at the same epoch budget. read the letter →

arxiv 2505.18107 v1 pith:XE54XNL6 submitted 2025-05-23 eess.IV cs.CV

Yichi Zhang , Zhihao Duan , Yuning Huang , Fengqing Zhu This is my paper · ORCID
classification eess.IVcs.CV MSC 68T0794A08
keywords learnedimagecompressiontrainingaccelerationcorrelationmodedecompositiondynamiclow-dimensionalsubspaceparameterembeddingsampling-then-moving-averagenoisyquadraticmodel
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that the expensive training of learned image compression (LIC) models can be substantially accelerated without sacrificing final rate–distortion performance. The key claim is that during most of training, parameter trajectories are so strongly correlated within a few 'modes' that every non-reference weight can be expressed as an affine function of one reference weight per mode, with coefficients that stabilize after roughly the first 20 epochs. Once the coefficients stabilize, the paper embeds the non-reference parameters — they stop being trained and simply follow the reference parameters. Combined with a moving-average step that smooths SGD noise, this reduces the training space dimension and trainable-parameter count by about half, and the authors report the same or slightly better BD-Rate in roughly 62% of the wall-clock time on three modern codecs (ELIC, TCM-S, FLIC) and on four domain-specific codecs. A theoretical analysis on a noisy quadratic model provides one supporting mechanism: the method's steady-state training variance is provably lower than standard SGD's.

What carries the argument

The load-bearing mechanism is Correlation Mode Decomposition (CMD), which expresses each non-reference weight as an affine function of a mode reference weight, $w_i(t)\approx k_i w_m(t)+d_i$, after clustering parameter trajectories into a few correlated modes. The method first performs CMD after $F=20$ epochs of normal training, then progressively 'true-embeds' the least-sensitive 1% of parameters per epoch once their affine coefficients have stabilized, while a small dummy-embedding step reinitializes a few extra parameters to their embedded values. A companion Sampling-then-Moving-Average (SMA) step interpolates periodically sampled SGD states and copies the moving average back into the optimizer, keeping temporal behavior smooth enough for the affine model to remain valid. Together these mechanisms shrink the training-space dimension toward the number of modes, which is as few as 50–120 in practice.

What would settle it

Pick a new LIC architecture not reported in the paper, run the method with F=20, and record the maximum relative drift of the fixed coefficients (k_i, d_i) between epoch 20 and the final epoch alongside the final BD-Rate against an SGD-trained anchor. If a large fraction of coefficients drift by more than a few percent after embedding, or if freezing them at epoch 20 costs more than roughly a 1% BD-Rate penalty, the central assumption would be contradicted.

Watch

Extended reading notes

Core claim

The central discovery is that LIC training dynamics are highly compressible: after a short warm-up, the trajectory of each weight $w_i(t)$ is well approximated by $w_i(t) \approx k_i w_m(t) + d_i$, where $w_m(t)$ is a reference trajectory from the same correlation mode. This linear representation, obtained by Correlation Mode Decomposition, lets the authors freeze the affine coefficients $(k_i, d_i)$ and 'embed' the corresponding weights so they are no longer optimized directly; they evolve through the reference parameter instead. The paper further shows that only about 25% of parameters are sensitive to being frozen this way, and that a Sampling-then-Moving-Average (SMA) schedule is needed to keep trajectories smooth enough for the approximation to hold. In experiments, the method reaches final R-D performance comparable to SGD in roughly 62% of the training time, and on a noisy quadratic model it provably attains lower steady-state variance than SGD.

Load-bearing premise

After roughly the first 20 epochs, each non-reference weight can be represented by a fixed affine function of its mode's reference weight whose coefficients do not drift meaningfully for the rest of training.

Editorial extensions

If this is right

  • Across the three benchmark codecs ELIC, TCM-S, and FLIC, the method reaches the final R-D point in roughly 62% of the SGD wall-clock time, with slightly negative BD-Rate on Kodak (about -0.7%).
  • The number of trainable parameters at the final epoch is roughly halved (e.g., FLIC drops from 70.96M to 41.39M), and the total parameter-epochs across all rate points fall to about 51% of the SGD total.
  • The same affine-embedding recipe transfers to stereo, remote-sensing, screen-content, and raw-image codecs, each trained in about 60–64% of the corresponding SGD time with comparable or better BD-Rate.
  • On a noisy quadratic model the proposed update rule provably has lower steady-state training variance than standard SGD at the same learning rate, which the paper identifies as the stabilizing mechanism for the embedding.
  • The method beats the low-dimensional baselines P-SGD, P-BFGS, and TWA (which diverge on LIC) and the sparsity baselines RigL and SRigL (which do not reduce training time), so the combination of mode embedding with moving-average smoothing is claimed to be the key.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the affine-coefficient stability measured at epoch 20 fails on architectures with very different layer dynamics (for instance, long-range transformer blocks), periodic re-estimation of the coefficients, or a slower embedding schedule, would likely be needed to preserve the time savings outside the paper's model set.
  • Combining the frozen parameters with sparse backpropagation that skips gradient computation for embedded weights could cut per-epoch cost as well as epoch count; the paper itself flags this as a non-trivial open implementation step.
  • The same trajectory-compression idea could be applied to the intra-frame coder of learned video compression, since that coder is itself an LIC; the motion-estimation modules would require separate treatment.
  • The strong correlation between 'CMD instant performance' and final BD-Rate suggests a cheap single-forward-pass diagnostic for how many effective dimensions any codec architecture needs, which may generalize beyond compression.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes two mechanisms to accelerate the training of learned image compression (LIC) models: STDET, which clusters parameter trajectories into modes, expresses non-reference parameters as affine functions of reference parameters, and progressively freezes ('embeds') parameters with stable affine coefficients; and SMA, which periodically averages sampled SGD iterates and synchronizes them back into training. The authors report that their method reaches the same or slightly better rate-distortion performance as standard SGD in roughly 62% of the wall-clock training time across ELIC, TCM-S, and FLIC, with additional experiments on stereo, remote sensing, screen content, and raw-image codecs. They also provide a noisy-quadratic-model analysis intended to show that the method reduces steady-state training variance relative to SGD.

Significance. Speeding up LIC training is a practically important goal, and the paper demonstrates a large reduction in wall-clock time on several modern codecs with essentially unchanged rate-distortion performance. The experimental scope is broad, the implementations of the compared codecs are publicly available, and the paper reports detailed ablations of the main hyperparameters. The variance analysis is derived from the algorithm's update rules rather than fitted to the measurements, and the paper includes a useful comparison with low-dimensional and sparsity-based training methods. The main weakness is that the reported wall-clock acceleration is not cleanly attributed to the proposed parameter-embedding mechanism, because the SMA-only control at the reduced epoch budget is missing and the paper admits that embedding does not reduce per-epoch gradient time in current frameworks.

major comments (4)
  1. [Appendix A.3 / Fig. 15] The decisive control experiment is missing. Appendix A.3 reports that 'ELIC + SMA' reaches the same final R-D loss (0.3442) as the full method, but the text states that this SMA-only run uses the same 120-epoch budget as standard SGD. Since Fig. 15 shows the full proposed method reaching the same loss by epoch 70, and since Appendix A.5 states that embedded parameters do not reduce per-epoch gradient time on current frameworks, the entire reported wall-clock saving comes from training for fewer epochs. The paper must report an SMA-only run at the proposed method's reduced budget (70 epochs for lambda=0.0018, 50 epochs for fine-tuning). If that run also reaches approximately 0.3442, the acceleration is attributable to the moving-average mechanism rather than to the dimension/parameter reduction that the title and abstract emphasize; if it does not, the result would support the claimed role of STDET.
  2. [Section 4.3 / Appendix A.5] The complexity analysis does not establish that reducing the number of trainable parameters reduces training time. Table 1 reports 'Total trainable params' and 'Final trainable params', but Section A.5 explicitly acknowledges that setting requires_grad=False does not prevent the embedded parameters from being computed in the chain rule, so per-epoch wall-clock time is essentially unchanged. Under this admission, the total-parameter reduction is an accounting metric, not a computational saving. The paper should report per-epoch wall-clock time as a function of the fraction of embedded parameters, or implement sparse backward computation that actually skips embedded parameters, before claiming that dimension reduction accelerates convergence.
  3. [Section 4.4, Tables 3-5] The hyperparameters F, M, S, P, L, alpha, and l appear to be selected using final BD-Rate on the same evaluation benchmarks that are later used to report the headline results. The ablation tables in Section 4.4 report BD-Rate without specifying a separate validation set, and the chosen configuration in each column ('0%' anchor) is the one used in Table 1. To avoid selection-on-test circularity, the hyperparameters should be chosen on a held-out validation split or on a subset of the training data, and the test-set results in Table 1 should be reported only for the final selected configuration.
  4. [Appendix A.6, Eq. (14)] The theoretical variance comparison rests on assumptions that are not verified in the paper. The inequality V*_Proposed <= V*_SGD requires both alpha in (0,1) and E[k_i^2] <= 1, and the proof also uses E[d_i] -> 0. The paper states only that E[d_i] -> 0 'empirically' and provides no evidence for E[k_i^2] <= 1. The authors should report the empirical distribution of k_i (and d_i) from the models in Section 4; if E[k_i^2] exceeds 1 for some modes, the stated inequality no longer follows. In addition, the variance of embedded parameters is written as E[k_i^2] V*_SMA, which omits a possible contribution from Var(d_i) unless d_i is treated as a deterministic offset.
minor comments (4)
  1. [Appendix A.2] The sentence 'following standard practie' contains a typo: 'practie' should be 'practice'.
  2. [Appendix A.4 / Fig. 16] Two subfigures are both labeled '(c) Proposed SMA', one for l=1 and one for l=2; the labels should be (c) and (d).
  3. [Table 1] The heading 'Total trainable params' is ambiguous because the reported values are parameter-epoch products (e.g., 35.42M x 520 epochs = 18,418M), not the number of parameters at any single time. The caption or column header should state this explicitly.
  4. [Section 3.3.2] The connection of SMA to the Lookahead optimizer is stated only briefly; a short formal comparison or a citation to the Lookahead update when l>1 would help readers understand the exact relationship.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the acceleration claim is an empirical comparison, though the paper's own A.3/A.5 expose an attribution gap that is a soundness concern rather than a circularity.

full rationale

The paper's derivation chain is constructive and self-contained: CMD affine embedding (Eq. 3), coefficient updates (Eq. 6), stability-based embedding (Eq. 7), and SMA interpolation (Eqs. 8-9) define an algorithm whose outputs (final R-D loss, BD-Rate, training time) are measured against an independent SGD baseline rather than read off from its inputs. The variance analysis in Appendix A.6 is derived from the algorithm's own update rules and standard noisy-quadratic recurrences, not fitted to the experimental results; the unproven condition E[k_i^2] ≤ 1 in Eq. 14 is an unsupported-assumption/correctness risk, not a presupposition of the conclusion. Hyperparameters F, M, S, P, L, α, l are selected through ablations scored on the same benchmark sets used for the final BD-Rate, which is test-set selection (an overfitting/soundness concern), not a quantity equivalent by construction to the selection criterion. Self-citations (e.g., Zhang et al. 2024b,c; 2025) are background LIC references and are not load-bearing for the acceleration claim; CMD is imported from Brokman et al. (2024) with its affine assumption stated explicitly rather than smuggled. The genuine weakness is empirical attribution: A.5 admits embedded parameters do not reduce per-epoch gradient time in current frameworks, and A.3 shows SMA alone reaches the same final loss at the SGD epoch budget while no SMA-only run at the proposed 70-epoch budget is reported, so the conclusion that dimension reduction causes the speedup is not established. That is a missing-control problem, not a circular definition or a fitted-input-called-prediction reduction.

Assumptions & free parameters 10 free parameters · 5 assumptions · 0 invented entities

The central method rests on several modeling assumptions and hand-picked hyperparameters. The mode decomposition assumes parameters in each mode are affine functions of a reference (Eq. 3), and that these coefficients stabilize after an initial head-stage. The sensitivity analysis assumes roughly 25% of parameters are sensitive, based on pruning literature, and uses a heuristic layer-wise/parameter-wise split. The theoretical analysis assumes E[k_i^2] <= 1 and E[d_i] -> 0 without proof. Hyperparameters F, M, S, P, L, alpha, l are all chosen via ablation on the evaluation datasets, making them fitted values rather than derived constants.

free parameters (10)
  • F (head-stage epochs) = 20 for lambda=0.0018, 10 for fine-tuning
    Selected from ablation (F=10 too early, F=30 marginal gain); controls when CMD and embedding start.
  • M (number of modes) = 50 (ELIC), 70 (TCM-S), 100 (FLIC)
    Selected per model via CMD instant performance traversal (Sec. 4.4.4).
  • S (sampled trajectories) = 200 x M (e.g., 10k, 14k, 20k)
    Chosen with M; used to approximate the correlation matrix without computing all N^2 entries.
  • P (embedding percentage per epoch) = 1%
    Chosen from ablation; smaller frequent embedding performed best.
  • L (embedding period) = 1
    Chosen from ablation; embedding every epoch was best.
  • alpha (SMA moving average factor) = 0.8
    Selected from ablation (alpha=0.8, l=5 was the best configuration).
  • l (SMA sampling interval) = 5
    Selected from ablation; values 3, 5, 10 were similar, 5 chosen as default.
  • Sensitivity threshold and layer split = 25% sensitive, top 50% layers, half-half split
    Heuristic based on pruning literature (Novak et al. 2018, Yang et al. 2023) and ad hoc layer division.
  • Dummy embedding schedule = min(tP/2%, 25%)
    Ad hoc schedule for reinitializing additional parameters to their embedded versions each epoch.
  • Perturbation scale sigma = fraction of max parameter value in layer
    Used in Gaussian perturbation for layer sensitivity; exact value not specified in the paper.
assumptions (5)
  • domain assumption LIC parameter trajectories can be approximated as affine transformations of a reference within a mode (Eq. 3)
    Core CMD assumption; supported only by Fig. 1 for ELIC at one lambda.
  • domain assumption Affine coefficients stabilize after the head-stage (F epochs)
    Used in Step 4 to justify embedding; evidence is Fig. 1 for ELIC.
  • domain assumption About 75% of parameters are insensitive to being fixed/pruned
    Based on Novak et al. 2018 and Yang et al. 2023; used to set the 25% sensitive threshold.
  • ad hoc to paper E[k_i^2] <= 1 and E[d_i] -> 0 in the noisy quadratic analysis
    Needed to make the variance-comparison factor <= I in Eq. 14; not proven empirically or theoretically.
  • standard math Standard contraction mapping / Banach fixed point theorem for the expectation recursion
    Used in Appendix A.6 to show convergence of expected parameters.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Accelerating Learned Image Compression Through Modeling Neural Training Dynamics." pith.science (2026). https://pith.science/paper/XE54XNL6

@misc{pith2026250518107,
  author       = {Pith},
  title        = {Pith review of: Accelerating Learned Image Compression Through Modeling Neural Training Dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XE54XNL6}},
  note         = {Machine review of arXiv:2505.18107}
}
read the original abstract

As learned image compression (LIC) methods become increasingly computationally demanding, enhancing their training efficiency is crucial. This paper takes a step forward in accelerating the training of LIC methods by modeling the neural training dynamics. We first propose a Sensitivity-aware True and Dummy Embedding Training mechanism (STDET) that clusters LIC model parameters into few separate modes where parameters are expressed as affine transformations of reference parameters within the same mode. By further utilizing the stable intra-mode correlations throughout training and parameter sensitivities, we gradually embed non-reference parameters, reducing the number of trainable parameters. Additionally, we incorporate a Sampling-then-Moving Average (SMA) technique, interpolating sampled weights from stochastic gradient descent (SGD) training to obtain the moving average weights, ensuring smooth temporal behavior and minimizing training state variances. Overall, our method significantly reduces training space dimensions and the number of trainable parameters without sacrificing model performance, thus accelerating model convergence. We also provide a theoretical analysis on the Noisy quadratic model, showing that the proposed method achieves a lower training variance than standard SGD. Our approach offers valuable insights for further developing efficient training methods for LICs.

Figures

Figures reproduced from arXiv: 2505.18107 by the authors.

Figure 1
Figure 1. ELIC (He et al., 2022) model, λ = 0.0018. (a) Clustered correlation matrix of sampled 10k parameter trajectories trained on COCO2017 dataset (Lin et al., 2014), decomposed to 10 modes. The diagonal block structure indicates high correlations of the parameters within each mode, which shows the accurate representation of the proposed method. (b) The table shows the percentage of affine coefficients {ki} N i=1 relative… view at source ↗
Figure 2
Figure 2. Testing loss comparison of various methods. Please zoom in for more details. The proposed method clearly converges much faster than standard SGD on various LICs. Additionally, as shown in the upper right corner, our method achieves a similar final convergence compared to SGD. λ = 0.0018, Testing R-D loss = λ · 2552 · MSE + Bpp. Evaluated on the Kodak dataset. 3 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. A typical pipeline in learned image compression. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: R-D curves of various methods. Please zoom in for more details. We also compare our approach with other efficient low-dimensional training methods, including P-SGD (Li et al., 2022b), P-BFGS (Li et al., 2022b), TWA (Li et al., 2023b), and sparsity-based training method…
Figure 5
Figure 5. Figure 5: Ablation experiments on proposed methods. ELIC model. 4.4.1 What is the impact of the factors in the CMD calculation procedure? In our method, three key factors influence the Step 1 CMD calculation. The first key factor is the predefined epoch F, which determines when …
Figure 6
Figure 6. Figure 6: Ablation experiments on proposed methods. TCM-S model. 0.1 0.2 0.3 bpp 30 32 PSNR (dB) F=10 F=30 M=10 M=100 S=1k S=5k S=20k F=20, M=50, S=10k (a) Mode decomposition 0.1 0.2 0.3 bpp 30 32 PSNR (dB) P=1% & L=2 P=1.5% & L=1 P=2% & L=1 P=2% & L=2 P=5% & L=5 P=10% & L=10 P:…
Figure 7
Figure 7. Figure 7: Ablation experiments on proposed methods. FLIC model [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: Ablation experiments on modes M and sampled trajectories S. Deeper color represents better performance. During our experiment, we observed an intriguing phenomenon: the CMD instant performance—i.e., the performance of the CMD-decomposed model (Step 1) evaluated immedia…
Figure 9
Figure 9. Figure 9: The instant CMD performance of different modes M and sampled trajectories S. Deeper color represents better performance. 5 Conclusion In this paper, we accelerate the training of LICs by modeling the neural training dynamics, considering the linear representations of L…
Figure 10
Figure 10. Figure 10: Rate-distortion curves in terms of PSNR on the Cityscapes and InStereo2K datasets. A.1.2 Remote Sensing Image Compression For remote sensing image compression, we evaluate our method on HL-RSCompNet (Xiang & Liang, 2024) using the following datasets: • GEset: This dat…
Figure 11
Figure 11. Figure 11: Rate-distortion curves with different methods on all the datasets [PITH_FULL_IMAGE:figures/full_fig_p027_11.png]
Figure 12
Figure 12. Figure 12: Rate-distortion curves with different compression methods on all the datasets [PITH_FULL_IMAGE:figures/full_fig_p028_12.png]
Figure 13
Figure 13. Figure 13: Rate-distortion curves with different compression methods on all the datasets. As has been demonstrated in the extensive experiments on stereo images, remote sensing images, screen content images, and raw images, our method consistently performs effectively across div…
Figure 14
Figure 14. Figure 14: Testing loss behavior across five independent runs for different methods. The evaluation is conducted on the Kodak dataset with λ = 0.0018. The testing R-D loss is computed as λ· 2552 ·MSE+Bpp. the main experiments. We employed the ELIC model (He et al., 2022) with λ …
Figure 15
Figure 15. Figure 15: Testing loss comparison of various techniques. The upper right corner shows the final convergence [PITH_FULL_IMAGE:figures/full_fig_p032_15.png]
Figure 16
Figure 16. Figure 16: Illustrative comparison of EMA (Szegedy et al., 2016; Polyak & Juditsky, 1992), explicit align￾ment (Song et al., 2023), and SMA (l = 1 & l = 2) over four training iterations. 32 [PITH_FULL_IMAGE:figures/full_fig_p032_16.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

94 extracted references · 74 canonical work pages

  1. [1]

    Intrinsic dimensionality explains the effectiveness of language model fine-tuning

    Armen Aghajanyan, Sonal Gupta, and Luke Zettlemoyer. Intrinsic dimensionality explains the effectiveness of language model fine-tuning. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pp.\ 7319--7328, August 2021

    2021

  2. [2]

    Towards efficient image compression without autoregressive models

    Muhammad Salman Ali, Yeongwoong Kim, Maryam Qamar, Sung-Chang Lim, Donghyun Kim, Chaoning Zhang, Sung-Ho Bae, and Hui Yong Kim. Towards efficient image compression without autoregressive models. Advances in Neural Information Processing Systems, 36: 0 7392--7404, 2023

    2023

  3. [3]

    Nonlinear transform coding

    Johannes Ball \'e , Philip A Chou, David Minnen, Saurabh Singh, Nick Johnston, Eirikur Agustsson, Sung Jin Hwang, and George Toderici. Nonlinear transform coding. IEEE Journal of Selected Topics in Signal Processing, 15 0 (2): 0 339--353, 2020

    2020

  4. [4]

    Variational image compression with a scale hyperprior

    Johannes Ballé, David Minnen, Saurabh Singh, Sung Jin Hwang, and Nick Johnston. Variational image compression with a scale hyperprior. International Conference on Learning Representations, 2018

    2018

  5. [5]

    Instereo2k: a large real dataset for stereo matching in indoor scenes

    Wei Bao, Wei Wang, Yuhua Xu, Yulan Guo, Siyu Hong, and Xiaohu Zhang. Instereo2k: a large real dataset for stereo matching in indoor scenes. Science China Information Sciences, 63: 0 1--11, 2020

    2020

  6. [6]

    Image reconstruction via deep image prior subspaces

    Riccardo Barbano, Javier Antoran, Johannes Leuschner, Jos \'e Miguel Hern \'a ndez-Lobato, Bangti Jin, and Zeljko Kereta. Image reconstruction via deep image prior subspaces. Transactions on Machine Learning Research, 2024

    2024

  7. [7]

    Compressai: a pytorch library and evaluation platform for end-to-end compression research, 2020

    Jean B \'e gaint, Fabien Racap \'e , Simon Feltman, and Akshay Pushparaja. Compressai: a pytorch library and evaluation platform for end-to-end compression research, 2020

    2020

  8. [8]

    Bj ntegaard

    G. Bj ntegaard. Calculation of average PSNR differences between rd-curves. ITU-T SG 16/Q6, 13th VCEG Meeting, April 2001

    2001

Show all 94 references
  1. [9]

    Enhancing neural training via a correlated dynamics model

    Jonathan Brokman, Roy Betser, Rotem Turjeman, Tom Berkov, Ido Cohen, and Guy Gilboa. Enhancing neural training via a correlated dynamics model. The Twelfth International Conference on Learning Representations, 2024

    2024

  2. [10]

    Brunton, Marko Budi s i\' c , Eurika Kaiser, and J

    Steven L. Brunton, Marko Budi s i\' c , Eurika Kaiser, and J. Nathan Kutz. Modern koopman theory for dynamical systems. SIAM Review, 64 0 (2): 0 229--340, 2022

    2022

  3. [11]

    Robust overfitting may be mitigated by properly learned smoothening

    Tianlong Chen, Zhenyu Zhang, Sijia Liu, Shiyu Chang, and Zhangyang Wang. Robust overfitting may be mitigated by properly learned smoothening. International Conference on Learning Representations, 2021

    2021

  4. [12]

    Illuminant estimation for color constancy: why spatial-domain methods work and the role of the color distribution

    Dongliang Cheng, Dilip K Prasad, and Michael S Brown. Illuminant estimation for color constancy: why spatial-domain methods work and the role of the color distribution. Journal of the Optical Society of America A, 31 0 (5): 0 1049--1058, 2014

    2014

  5. [13]

    The cityscapes dataset for semantic urban scene understanding

    Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The cityscapes dataset for semantic urban scene understanding. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Rec...

    2016

  6. [14]

    Parameter-efficient fine-tuning of large-scale pre-trained language models

    Ning Ding, Yujia Qin, Guang Yang, Fuchao Wei, Zonghan Yang, Yusheng Su, Shengding Hu, Yulin Chen, Chi-Min Chan, Weize Chen, et al. Parameter-efficient fine-tuning of large-scale pre-trained language models. Nature Machine Intelligence, 5 0 (3): 0 220--235, 2023

    2023

  7. [15]

    Optimizing neural networks via koopman operator theory

    Akshunna S Dogra and William Redman. Optimizing neural networks via koopman operator theory. Advances in Neural Information Processing Systems, 33: 0 2087--2097, 2020

    2020

  8. [16]

    Qarv: Quantization-aware resnet vae for lossy image compression

    Zhihao Duan, Ming Lu, Jack Ma, Yuning Huang, Zhan Ma, and Fengqing Zhu. Qarv: Quantization-aware resnet vae for lossy image compression. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

    2023

  9. [17]

    Asymmetric numeral systems, 2009

    Jarek Duda. Asymmetric numeral systems, 2009

    2009

  10. [18]

    Rigging the lottery: Making all tickets winners

    Utku Evci, Trevor Gale, Jacob Menick, Pablo Samuel Castro, and Erich Elsen. Rigging the lottery: Making all tickets winners. Proceedings of the International Conference on Machine Learning, pp.\ 2943--2952, 2020

    2020

  11. [19]

    Understanding the difficulty of training deep feedforward neural networks

    Xavier Glorot and Yoshua Bengio. Understanding the difficulty of training deep feedforward neural networks. Proceedings of the International Conference on Artificial Intelligence and Statistics, pp.\ 249--256, 2010

    2010

  12. [20]

    Improving neural network training in low dimensional random bases

    Frithjof Gressmann, Zach Eaton-Rosen, and Carlo Luschi. Improving neural network training in low dimensional random bases. Advances in Neural Information Processing Systems, 33: 0 12140--12150, 2020

    2020

  13. [21]

    EVC : Towards real-time neural image compression with mask decay

    Wang Guo-Hua, Jiahao Li, Bin Li, and Yan Lu. EVC : Towards real-time neural image compression with mask decay. The Eleventh International Conference on Learning Representations, 2023

    2023

  14. [22]

    Checkerboard context model for efficient learned image compression

    Dailan He, Yaoyan Zheng, Baocheng Sun, Yan Wang, and Hongwei Qin. Checkerboard context model for efficient learned image compression. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 14771--14780, 2021

    2021

  15. [23]

    E LIC : Efficient learned image compression with unevenly grouped space-channel contextual adaptive coding

    Dailan He, Ziming Yang, Weikun Peng, Rui Ma, Hongwei Qin, and Yan Wang. E LIC : Efficient learned image compression with unevenly grouped space-channel contextual adaptive coding. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 5718--572...

    2022

  16. [24]

    Asymmetric valleys: Beyond sharp and flat local minima

    Haowei He, Gao Huang, and Yang Yuan. Asymmetric valleys: Beyond sharp and flat local minima. Advances in Neural Information Processing Systems, 32, 2019

    2019

  17. [25]

    Delving deep into rectifiers: Surpassing human-level performance on imagenet classification

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Delving deep into rectifiers: Surpassing human-level performance on imagenet classification. Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 1026--1034, 2015

    2015

  18. [26]

    Momentum contrast for unsupervised visual representation learning

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual representation learning. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 9729--9738, 2020

    2020

  19. [27]

    Lo RA : Low-rank adaptation of large language models

    Edward J Hu, yelong shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lo RA : Low-rank adaptation of large language models. International Conference on Learning Representations, 2022

    2022

  20. [28]

    JPEG AI Common Training & Test Conditions v8.0

    ISO/IEC JTC 1/SC29/WG1 . JPEG AI Common Training & Test Conditions v8.0 . CPM, 100th Meeting, Covilh \ a , Portugal, July 2023. Document number: N100600

    2023

  21. [29]

    Garipov, Dmitry P

    Pavel Izmailov, Dmitrii Podoprikhin, T. Garipov, Dmitry P. Vetrov, and Andrew Gordon Wilson. Averaging weights leads to wider optima and better generalization. Conference on Uncertainty in Artificial Intelligence, 2018

    2018

  22. [30]

    Visual prompt tuning

    Menglin Jia, Luming Tang, Bor-Chun Chen, Claire Cardie, Serge Belongie, Bharath Hariharan, and Ser-Nam Lim. Visual prompt tuning. European Conference on Computer Vision, pp.\ 709--727, 2022

    2022

  23. [31]

    Towards practical real-time neural video compression, 2025

    Zhaoyang Jia, Bin Li, Jiahao Li, Wenxuan Xie, Linfeng Qi, Houqiang Li, and Yan Lu. Towards practical real-time neural video compression, 2025. URL https://arxiv.org/abs/2502.20762

    2025 arXiv

  24. [32]

    Variable-rate learned image compression with multi-objective optimization and quantization-reconstruction offsets

    Fatih Kamisli, Fabien Racap \'e , and Hyomin Choi. Variable-rate learned image compression with multi-objective optimization and quantization-reconstruction offsets. Proceedings of the Data Compression Conference, 2024

    2024

  25. [33]

    Multi-layer random perturbation training for improving model generalization efficiently

    Lis Kanashiro Pereira, Yuki Taya, and Ichiro Kobayashi. Multi-layer random perturbation training for improving model generalization efficiently. Proceedings of the Fourth BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP, pp.\ 303--310, November 2021

    2021

  26. [34]

    A software platform for manipulating the camera imaging pipeline

    Hakki Can Karaimer and Michael S Brown. A software platform for manipulating the camera imaging pipeline. Computer Vision--ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11--14, 2016, Proceedings, Part I 14, pp.\ 429--444, 2016

    2016

  27. [35]

    Understanding black-box predictions via influence functions

    Pang Wei Koh and Percy Liang. Understanding black-box predictions via influence functions. Proceedings of the 34th International Conference on Machine Learning, 70: 0 1885--1894, 2017

    2017

  28. [36]

    Dynamic sparse training with structured sparsity

    Mike Lasby, Anna Golubeva, Utku Evci, Mihai Nica, and Yani Ioannou. Dynamic sparse training with structured sparsity. Proceedings of the International Conference on Learning Representations, 2024

    2024

  29. [37]

    Measuring the intrinsic dimension of objective landscapes

    Chunyuan Li, Heerad Farkhoor, Rosanne Liu, and Jason Yosinski. Measuring the intrinsic dimension of objective landscapes. International Conference on Learning Representations, 2018

    2018

  30. [38]

    Frequency-aware transformer for learned image compression

    Han Li, Shaohui Li, Wenrui Dai, Chenglin Li, Junni Zou, and Hongkai Xiong. Frequency-aware transformer for learned image compression. The Twelfth International Conference on Learning Representations, 2024 a

    2024

  31. [39]

    Deep contextual video compression

    Jiahao Li, Bin Li, and Yan Lu. Deep contextual video compression. Proceedings of the Advances in Neural Information Processing Systems, 34: 0 18114--18125, 2021

    2021

  32. [40]

    Hybrid spatial-temporal entropy modelling for neural video compression

    Jiahao Li, Bin Li, and Yan Lu. Hybrid spatial-temporal entropy modelling for neural video compression. Proceedings of the ACM International Conference on Multimedia, pp.\ 1503--1511, 2022 a

    2022

  33. [41]

    Neural video compression with diverse contexts

    Jiahao Li, Bin Li, and Yan Lu. Neural video compression with diverse contexts. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 22616--22626, 2023 a

    2023

  34. [42]

    Neural video compression with feature modulation

    Jiahao Li, Bin Li, and Yan Lu. Neural video compression with feature modulation. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 26099--26108, 2024 b

    2024

  35. [43]

    Low dimensional trajectory hypothesis is true: Dnns can be trained in tiny subspaces

    Tao Li, Lei Tan, Zhehao Huang, Qinghua Tao, Yipeng Liu, and Xiaolin Huang. Low dimensional trajectory hypothesis is true: Dnns can be trained in tiny subspaces. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45 0 (3): 0 3411--3420, 2022 b

    2022

  36. [44]

    Subspace adversarial training

    Tao Li, Yingwen Wu, Sizhe Chen, Kun Fang, and Xiaolin Huang. Subspace adversarial training. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 13409--13418, 2022 c

    2022

  37. [45]

    Trainable weight averaging: Efficient training by optimizing historical solutions

    Tao Li, Zhehao Huang, Qinghua Tao, Yingwen Wu, and Xiaolin Huang. Trainable weight averaging: Efficient training by optimizing historical solutions. The Eleventh International Conference on Learning Representations, 2023 b

    2023

  38. [46]

    Revisiting random weight perturbation for efficiently improving generalization

    Tao Li, Qinghua Tao, Weihao Yan, Yingwen Wu, Zehao Lei, Kun Fang, Mingzhen He, and Xiaolin Huang. Revisiting random weight perturbation for efficiently improving generalization. Transactions on Machine Learning Research, 2024 c

    2024

  39. [47]

    Deep model fusion: A survey, 2023 c

    Weishi Li, Yong Peng, Miao Zhang, Liang Ding, Han Hu, and Li Shen. Deep model fusion: A survey, 2023 c

    2023

  40. [48]

    Microsoft COCO : Common O bjects in C ontext

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll \'a r, and C Lawrence Zitnick. Microsoft COCO : Common O bjects in C ontext. European Conference on Computer Vision, pp.\ 740--755, 2014

    2014

  41. [49]

    Learned image compression with mixed transformer-cnn architectures

    Jinming Liu, Heming Sun, and Jiro Katto. Learned image compression with mixed transformer-cnn architectures. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 14388--14397, June 2023

    2023

  42. [50]

    Bidirectional stereo image compression with cross-dimensional entropy model

    Zhening Liu, Xinjie Zhang, Jiawei Shao, Zehong Lin, and Jun Zhang. Bidirectional stereo image compression with cross-dimensional entropy model. Proceedings of the European Conference on Computer Vision, pp.\ 480--496, 2024

    2024

  43. [51]

    Deep learning for universal linear embeddings of nonlinear dynamics

    Bethany Lusch, J Nathan Kutz, and Steven L Brunton. Deep learning for universal linear embeddings of nonlinear dynamics. Nature communications, 9 0 (1): 0 4950, 2018

    2018

  44. [52]

    New insights and perspectives on the natural gradient method

    James Martens. New insights and perspectives on the natural gradient method. Journal of Machine Learning Research, 21 0 (146): 0 1--76, 2020

    2020

  45. [53]

    Advancing the rate-distortion-computation frontier for neural image compression

    David Minnen and Nick Johnston. Advancing the rate-distortion-computation frontier for neural image compression. 2023 IEEE International Conference on Image Processing, pp.\ 2940--2944, 2023

    2023

  46. [54]

    Channel-wise autoregressive entropy models for learned image compression

    David Minnen and Saurabh Singh. Channel-wise autoregressive entropy models for learned image compression. 2020 IEEE International Conference on Image Processing, pp.\ 3339--3343, 2020

    2020

  47. [55]

    Importance estimation for neural network pruning

    Pavlo Molchanov, Arun Mallya, Stephen Tyree, Iuri Frosio, and Jan Kautz. Importance estimation for neural network pruning. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 11264--11272, 2019

    2019

  48. [56]

    Exponential moving average of weights in deep learning: Dynamics and benefits

    Daniel Morales-Brotons, Thijs Vogels, and Hadrien Hendrikx. Exponential moving average of weights in deep learning: Dynamics and benefits. Transactions on Machine Learning Research, 2024

    2024

  49. [57]

    Decomposed linear dynamical systems (dlds) for learning the latent components of neural dynamics

    Noga Mudrik, Yenho Chen, Eva Yezerets, Christopher J Rozell, and Adam S Charles. Decomposed linear dynamical systems (dlds) for learning the latent components of neural dynamics. Journal of Machine Learning Research, 25 0 (59): 0 1--44, 2024

    2024

  50. [58]

    Modern hierarchical, agglomerative clustering algorithms, 2011

    Daniel M \"u llner. Modern hierarchical, agglomerative clustering algorithms, 2011

    2011

  51. [59]

    Learning srgb-to-raw-rgb de-rendering with content-aware metadata

    Seonghyeon Nam, Abhijith Punnappurath, Marcus A Brubaker, and Michael S Brown. Learning srgb-to-raw-rgb de-rendering with content-aware metadata. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 17704--17713, 2022

    2022

  52. [60]

    A review on weight initialization strategies for neural networks

    Meenal V Narkhede, Prashant P Bartakke, and Mukul S Sutaone. A review on weight initialization strategies for neural networks. Artificial intelligence review, 55 0 (1): 0 291--322, 2022

    2022

  53. [61]

    Scid: A database for screen content images quality assessment

    Zhangkai Ni, Lin Ma, Huanqiang Zeng, Ying Fu, Lu Xing, and Kai-Kuang Ma. Scid: A database for screen content images quality assessment. Proceedings of International Symposium on Intelligent Signal Processing and Communication Systems, pp.\ 774--779, 2017

    2017

  54. [62]

    Abolafia, Jeffrey Pennington, and Jascha Sohl-Dickstein

    Roman Novak, Yasaman Bahri, Daniel A. Abolafia, Jeffrey Pennington, and Jascha Sohl-Dickstein. Sensitivity and generalization in neural networks: an empirical study. International Conference on Learning Representations, 2018

    2018

  55. [63]

    Banach's fixed point theorem for partial metric spaces, 2004

    Sandra Oltra and Oscar Valero. Banach's fixed point theorem for partial metric spaces, 2004

    2004

  56. [64]

    Acceleration of stochastic approximation by averaging

    Boris T Polyak and Anatoli B Juditsky. Acceleration of stochastic approximation by averaging. SIAM Journal on Control and Optimization, 30 0 (4): 0 838--855, 1992

    1992

  57. [65]

    Your transformer is secretly linear, 2024

    Anton Razzhigaev, Matvey Mikhalchuk, Elizaveta Goncharova, Nikolai Gerasimenko, Ivan Oseledets, Denis Dimitrov, and Andrey Kuznetsov. Your transformer is secretly linear, 2024

    2024

  58. [66]

    Exact solutions to the nonlinear dynamics of learning in deep linear neural networks, 2013

    Andrew M Saxe, James L McClelland, and Surya Ganguli. Exact solutions to the nonlinear dynamics of learning in deep linear neural networks, 2013

    2013

  59. [67]

    No more pesky learning rates

    Tom Schaul, Sixin Zhang, and Yann LeCun. No more pesky learning rates. Proceedings of the 30th International Conference on Machine Learning, 28 0 (3): 0 343--351, 2013

    2013

  60. [68]

    Dynamic mode decomposition of numerical and experimental data

    Peter J Schmid. Dynamic mode decomposition of numerical and experimental data. Journal of Fluid Mechanics, 656: 0 5--28, 2010

    2010

  61. [69]

    Dynamic mode decomposition and its variants

    Peter J Schmid. Dynamic mode decomposition and its variants. Annual Review of Fluid Mechanics, 54 0 (1): 0 225--254, 2022

    2022

  62. [70]

    Temporal context mining for learned video compression

    Xihua Sheng, Jiahao Li, Bin Li, Li Li, Dong Liu, and Yan Lu. Temporal context mining for learned video compression. IEEE Transactions on Multimedia, 25: 0 7311--7322, 2022

    2022

  63. [71]

    Consistency models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. Proceedings of the International Conference on Machine Learning, 202: 0 32211--32252, 2023

    2023

  64. [72]

    Rethinking the inception architecture for computer vision

    Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 2818--2826, 2016

    2016

  65. [73]

    Adanic: Towards practical neural image compression via dynamic transform routing

    Lvfang Tao, Wei Gao, Ge Li, and Chenhao Zhang. Adanic: Towards practical neural image compression via dynamic transform routing. Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 16879--16888, October 2023

    2023

  66. [74]

    Beyond learned metadata-based raw image reconstruction

    Yufei Wang, Yi Yu, Wenhan Yang, Lanqing Guo, Lap-Pui Chau, Alex C Kot, and Bihan Wen. Beyond learned metadata-based raw image reconstruction. International Journal of Computer Vision, 132 0 (12): 0 5514--5533, 2024

    2024

  67. [75]

    Towards certificated model robustness against weight perturbations

    Tsui-Wei Weng, Pu Zhao, Sijia Liu, Pin-Yu Chen, Xue Lin, and Luca Daniel. Towards certificated model robustness against weight perturbations. Proceedings of the AAAI Conference on Artificial Intelligence, 34 0 (04): 0 6356--6363, 2020

    2020

  68. [76]

    Ecsic: Epipolar cross attention for stereo image compression

    Matthias W \"o dlinger, Jan Kotera, Manuel Keglevic, Jan Xu, and Robert Sablatnig. Ecsic: Epipolar cross attention for stereo image compression. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp.\ 3436--3445, 2024

    2024

  69. [77]

    Understanding short-horizon bias in stochastic meta-optimization

    Yuhuai Wu, Mengye Ren, Renjie Liao, and Roger Grosse. Understanding short-horizon bias in stochastic meta-optimization. International Conference on Learning Representations, 2018

    2018

  70. [78]

    Remote sensing image compression based on high-frequency and low-frequency components

    Shao Xiang and Qiaokang Liang. Remote sensing image compression based on high-frequency and low-frequency components. IEEE Transactions on Geoscience and Remote Sensing, 2024

    2024

  71. [79]

    Perceptual quality assessment of screen content images

    Huan Yang, Yuming Fang, and Weisi Lin. Perceptual quality assessment of screen content images. IEEE Transactions on Image Processing, 24 0 (11): 0 4408--4421, 2015

    2015

  72. [80]

    Deep neural network pruning method based on sensitive layers and reinforcement learning

    Wenchuan Yang, Haoran Yu, Baojiang Cui, Runqi Sui, and Tianyu Gu. Deep neural network pruning method based on sensitive layers and reinforcement learning. Artificial Intelligence Review, 56 0 (Suppl 2): 0 1897--1917, 2023

    1917

  73. [81]

    Computationally-efficient neural image compression with shallow decoders

    Yibo Yang and Stephan Mandt. Computationally-efficient neural image compression with shallow decoders. Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 530--540, 2023

    2023

  74. [82]

    Safer: Layer-level sensitivity assessment for efficient and robust neural network inference, 2023

    Edouard Yvinec, Arnaud Dapogny, and Kevin Bailly. Safer: Layer-level sensitivity assessment for efficient and robust neural network inference, 2023

    2023

  75. [83]

    Are all layers created equal? Journal of Machine Learning Research, 23 0 (67): 0 1--28, 2022

    Chiyuan Zhang, Samy Bengio, and Yoram Singer. Are all layers created equal? Journal of Machine Learning Research, 23 0 (67): 0 1--28, 2022

    2022

  76. [84]

    Which algorithmic choices matter at which batch sizes? insights from a noisy quadratic model

    Guodong Zhang, Lala Li, Zachary Nado, James Martens, Sushant Sachdeva, George E Dahl, Christopher J Shallue, and Roger Grosse. Which algorithmic choices matter at which batch sizes? insights from a noisy quadratic model. Advances in Neural Information Processing Systems, 2019 a

    2019

  77. [85]

    Lookaround optimizer: k steps around, 1 step average

    Jiangtao Zhang, Shunyu Liu, Jie Song, Tongtian Zhu, Zhengqi Xu, and Mingli Song. Lookaround optimizer: k steps around, 1 step average. Thirty-seventh Conference on Neural Information Processing Systems, 2023

    2023

  78. [86]

    Lookahead optimizer: k steps forward, 1 step back

    Michael Zhang, James Lucas, Jimmy Ba, and Geoffrey E Hinton. Lookahead optimizer: k steps forward, 1 step back. Advances in Neural Information Processing Systems, 32, 2019 b

    2019

  79. [87]

    Theoretical bound-guided hierarchical vae for neural image codecs

    Yichi Zhang, Zhihao Duan, Yuning Huang, and Fengqing Zhu. Theoretical bound-guided hierarchical vae for neural image codecs. 2024 IEEE International Conference on Multimedia and Expo (ICME), pp.\ 1--6, 2024 a

    2024

  80. [88]

    Another way to the top: Exploit contextual clustering in learned image coding

    Yichi Zhang, Zhihao Duan, Ming Lu, Dandan Ding, Fengqing Zhu, and Zhan Ma. Another way to the top: Exploit contextual clustering in learned image coding. Proceedings of the AAAI Conference on Artificial Intelligence, 38 0 (8): 0 9377--9386, 2024 b

    2024

  81. [89]

    On efficient neural network architectures for image compression

    Yichi Zhang, Zhihao Duan, and Fengqing Zhu. On efficient neural network architectures for image compression. 2024 IEEE International Conference on Image Processing, 2024 c

    2024

  82. [90]

    Balanced rate-distortion optimization in learned image compression

    Yichi Zhang, Zhihao Duan, Yuning Huang, and Fengqing Zhu. Balanced rate-distortion optimization in learned image compression. arXiv preprint arXiv:2502.20161, 2025

    2025 arXiv

  83. [91]

    Enhanced screen content image compression: A synergistic approach for structural fidelity and text integrity preservation

    Fangtao Zhou, Xiaofeng Huang, Peng Zhang, Meng Wang, Zhao Wang, Yang Zhou, and Haibing Yin. Enhanced screen content image compression: A synergistic approach for structural fidelity and text integrity preservation. Proceedings of the ACM International Conference on Multimedia,...

    2024

  84. [92]

    Towards understanding why lookahead generalizes better than sgd and beyond

    Pan Zhou, Hanshu Yan, Xiaotong Yuan, Jiashi Feng, and Shuicheng Yan. Towards understanding why lookahead generalizes better than sgd and beyond. Advances in Neural Information Processing Systems, 34: 0 27290--27304, 2021 a

    2021

  85. [93]

    Efficient neural network training via forward and backward propagation sparsification

    Xiao Zhou, Weizhong Zhang, Zonghao Chen, Shizhe Diao, and Tong Zhang. Efficient neural network training via forward and backward propagation sparsification. Advances in Neural Information Processing Systems, 34: 0 15216--15229, 2021 b

    2021

  86. [94]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.