Pith. sign in

REVIEW 3 major objections 6 minor 87 references

On-device Content-based Recommendation with Single-shot Embedding Pruning: A Cooperative Game Perspective

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Shaver is a single-shot embedding pruning method that gives every embedding parameter a Shapley value as a fair attribution score, so one well-trained recommender model can be compressed to any parameter budget without retraining.

desk verdict Shaver is a credible single-shot embedding pruner for CTR models, with a theorem that genuinely buys a complexity reduction; just know the theorem and the algorithm only work when every field activates exactly one feature per row, and the paper doesn't say that loudly. read the letter →

arxiv 2411.13052 v2 pith:DI6NZIOC submitted 2024-11-20 cs.IR cs.LG

Hung Vinh Tran , Tong Chen , Guanhua Ye , Quoc Viet Hung Nguyen , Kai Zheng , Hongzhi Yin This is my paper · ORCID
classification cs.IRcs.LG
keywords embeddingpruningShapleyvaluesingle-shotcontent-basedrecommendationclick-throughratepredictionon-devicerecommendersystemsfield-awarecodebookcooperativegametheory
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

Shaver is a method for compressing the embedding tables of content-based recommender systems in a single shot, without retraining for each target size. The paper's central claim is that every embedding parameter can be given a Shapley value as a fair attribution score, and that pruning the lowest-scoring parameters to any budget preserves recommendation accuracy well. A theorem about corresponding players is what makes this tractable: because each feature field contributes exactly one active feature per instance, the parameter-level Shapley game collapses to a field-level game, cutting estimation cost from $O(nd|D|)$ to $O(md|D|)$. A field-aware codebook fills pruned entries with learned per-field placeholder values instead of zeros, limiting the damage to feature-interaction computations. If correct, Shaver lets a single well-trained model be adapted to heterogeneous device capacities, federated settings, or changing memory and battery conditions.

What carries the argument

The load-bearing mechanism is the corresponding-player identity (Definition 4.2 and Theorem 4.3): for a data instance in which feature $i$ of field $j$ is the sole activated feature, the Shapley value of embedding parameter $(i,c)$ under the parameter-level value function equals the Shapley value of field-slot player $(j,c)$ under the field-level value function. This identity is what collapses the combinatorial game over $nd$ parameters to one over $md$ field slots, making the Monte Carlo estimation in Algorithm 1 tractable. Two auxiliary mechanisms support the result: the local-value formulation of Eq. (10), which turns global Shapley values into averages over per-instance permutations, and the field-aware codebook of Eqs. (14)-(15), which imputes pruned parameters with per-field weighted-average embeddings rather than zeros.

What would settle it

Take a small CTR model with a field whose instances can activate two features simultaneously, enumerate all parameter subsets to compute exact Shapley values, and compare them with Shaver's field-level estimates on the same data; any material gap in the attribution of an active parameter would falsify Theorem 4.3's equality. A simpler check is whether the method's validation AUC degrades sharply when a multi-hot field is included in an otherwise one-hot dataset.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that Shapley-value attribution, previously considered too expensive for parameter-level pruning in recommender systems, can be computed efficiently for embedding tables because of the one-hot structure of categorical CTR data. Definition 4.2 pairs each activated embedding parameter $(i,c)$ with a field-level player $(j,c)$, and Theorem 4.3 proves their Shapley values are equal for that instance. Consequently, estimating attribution over the $n$ features reduces to estimating it over the $m$ fields, and Algorithm 1 accumulates unbiased Monte Carlo estimates in $O(md|D|)$ forward passes. The pruned model replaces low-scoring entries with per-field codebook values that solve in closed form, giving a single-shot compressor that the experiments show matches or beats trained-from-scratch baselines on Criteo, Avazu, and KDD across sparsity rates from 20% to 99.9%.

Load-bearing premise

The method's efficiency rests on a single one-hot premise: every field activates exactly one feature per instance, so an inactive embedding parameter is a null player and the field-level game exactly mirrors the parameter-level game.

Editorial extensions

If this is right

  • A trained CTR model can be shipped once and specialized to any device memory budget by ranking parameters and truncating, with no per-budget retraining or mask search.
  • The $O(md|D|)$ cost means attribution can be recomputed on-device or in streaming and federated settings whenever the model or data distribution shifts.
  • Because the codebook stores only $md$ values, replacing zero-padding adds negligible storage while preserving more of the dot-product and element-wise feature interactions that CTR models rely on.
  • The method extends beyond the three reported datasets to any categorical CTR model whose input is a concatenation of one-hot field encodings, including DeepFM and DCN-v2 backbones.
  • Combining Shaver's scores with post-pruning fine-tuning yields further gains in several settings, though the paper notes occasional degradation from validation-set leakage.

Reading between the lines

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

  • Our inference: the one-hot assumption in Eq. (1) is the hinge. If a deployment uses multi-hot fields, user-defined dense features, or feature crosses, the corresponding-player equality no longer holds exactly; a fallback would be to split multi-hot fields into per-feature singleton games or treat the active set as a coalition, at extra cost.
  • Our inference: the codebook's tendency to spend less budget on high-frequency features (Figure 4) suggests a principled pruning policy could be derived from feature frequency and field interactions, not just raw attribution magnitude.
  • Our inference: Shaver's attribution scores are computed from a trained model, so the method should transfer to other interaction-heavy architectures such as factorization machines and cross networks; a natural test is whether attribution scores computed on DeepFM also prune a DCN-v2 model effectively.
  • Our inference: because estimation only needs forward passes, the same scoring procedure could be applied periodically in a streaming setting to adapt an embedding table to drift, something the paper motivates but does not evaluate.
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

3 major / 6 minor

Summary. The paper introduces Shaver, a single-shot embedding-pruning method for content-based recommender systems (CTR models). Shaver assigns each embedding parameter a Shapley value as an attribution score, so that a trained model can be pruned to any parameter budget without retraining. To make Shapley computation tractable, the paper proves Theorem 4.3, which reduces the complexity from O(nd|D|) to O(md|D|) under the assumption that each feature field has exactly one activated feature per instance. The paper also proposes a field-aware codebook to replace zero-padding for pruned parameters, with a closed-form solution. Experiments on Criteo, Avazu, and KDD with DeepFM and DCN-Mix backbones compare Shaver with non-single-shot and single-shot baselines, and an on-device benchmark is reported.

Significance. If the theoretical reduction and the empirical claims hold, Shaver would be a useful contribution: it targets a practical single-shot pruning setting, provides a nontrivial complexity reduction, ships code, and includes a synthetic-data verification that the estimated Shapley values are close to exact ones. The experiments cover three large-scale datasets, two backbones, many sparsity levels, and an edge-device deployment benchmark, which is a real strength. However, the central efficiency theorem depends on a single-activation-per-field assumption that is not stated as a limitation, and the experimental sections report no variance across seeds. Both issues bound the significance of the work as written.

major comments (3)
  1. [§4.3, Definition 4.2, Appendix A] Theorem 4.3 and its proof are internally sound only when every feature field activates exactly one feature per instance, as encoded in Eq. (1). Under multi-hot fields—e.g., movie genres or user interest tags—the 'corresponding player' of Definition 4.2 is not unique, the map g in Eq. (16) is no longer single-valued, Algorithm 1 line 9 is ambiguous, and the counting argument behind the O(md|D|) complexity claim collapses. Multi-hot categorical features are routine in content-based recommendation, so this is a load-bearing restriction on the paper's central claim. The authors should either state this assumption as a clear scope limitation or extend the method and theorem to multi-hot fields (for instance, by decomposing fields into per-feature subgames).
  2. [§5.2–5.4, Tables 2 and 3, Figure 2] No error bars, confidence intervals, or multiple-seed results are reported anywhere. Many of the claimed advantages are based on AUC differences of 0.0001–0.001, which are within the range where seed-to-seed variation matters. Since the headline contribution is 'competitive performance' across budgets, single-run results are not sufficient to support the comparison. Please report means and standard deviations over at least three seeds, or explicitly state that only one seed was used and discuss the implications.
  3. [§4.4, Eq. (14), Appendix A] Equation (14) defines the optimal codebook with respect to a uniformly sampled pruned set Q of size B, but in the actual pruning procedure Q is the set of parameters with the lowest Shapley values. Consequently, the closed-form solution in Eq. (15) is optimal for a different distribution over pruning masks than the one actually used. The paper calls C* 'optimal' and attributes performance gains to it, so this mismatch is a gap between the derivation and the deployed algorithm. Either derive the codebook for the actual selection rule or explicitly present it as a heuristic approximation.
minor comments (6)
  1. [§4.2, Eq. (10)] The notation R(x,y) is used both for a random permutation and as a paired sample; this is confusing because R was previously an ordered set. Please rename one of them.
  2. [§4.2, text after Eq. (10)] There is a typo: 'with a quality, decent-sizedD' should read 'with a quality, decent-sized D'.
  3. [§4.3, first paragraph] The word 'Shapely' in 'efficient Shapely value computation' is misspelled; it should be 'Shapley'.
  4. [Table 1 and Appendix C.2] The dataset preprocessing description does not state whether any field in Criteo, Avazu, or KDD is multi-hot. Since the theorem assumes single activation per field, please make explicit whether the preprocessed datasets satisfy this condition.
  5. [Figure 4] The x-axis label 'Frequency Bins' is not self-explanatory; please clarify whether bins are equal-frequency or equal-width, and whether the three bins contain equal numbers of features.
  6. [§5.4, closing paragraph] The phrase 'can cause data leakage' is inaccurate if 'leakage' refers to test-set contamination: Shapley values are computed from training and validation data, and the test set is held out. Please rephrase to describe the selection effect on fine-tuning more precisely.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Shaver's Shapley-value derivation is proven from stated one-hot assumptions, and its results are evaluated on held-out test data.

full rationale

Shaver's derivation chain is self-contained. The Shapley-value definition (Eqs. 5-8), the local/global decomposition (Eqs. 9-10), and Theorem 4.3 form a mathematical argument with explicitly stated assumptions: single activation per feature field (Eq. 1) and the corresponding-player mapping in Definition 4.2. Theorem 4.3 is proved in Appendix A via explicit binomial and Chu-Vandermonde identities rather than assumed, and the O(md|D|) complexity claim follows from that proof together with Algorithm 1, which updates exactly the activated parameter's Shapley accumulator. The codebook is a closed-form minimizer of a stated expected-distance objective (Eqs. 14-15), using feature frequencies without encoding the target test results. The only self-citations ([7], [55], and related on-device recommendation works) are contextual or provide public data splits and preprocessing; no load-bearing claim rests on an unverified self-cited theorem. The validation-set usage for attribution scores is explicitly disclosed in Section 5.4 as a potential source of data leakage during fine-tuning, but this is analogous to model selection on validation data and does not make the test-set evaluation circular. The multi-hot field limitation raised by a skeptical reading is a domain/assumption risk: if a field activates multiple features, the corresponding-player map in Definition 4.2 is not unique and Theorem 4.3's premise fails. That is a correctness/robustness concern about the stated assumptions, not a circularity, because the paper's own equations do not presuppose the theorem's conclusion.

Assumptions & free parameters 0 free parameters · 3 assumptions · 1 invented entities

The central method does not fit any constant to the target result: Shapley scores are computed from the trained model, and the codebook has a closed-form weighted-mean solution. The main assumptions are the one-hot feature structure, the placeholder-based definition of removal, and standard Shapley sampling math.

assumptions (3)
  • domain assumption Each categorical feature field contributes exactly one active feature per data instance, so all other features in the field are inactive for that instance.
    Invoked by Eq. (1) and Definition 4.2; this is what makes inactive embedding parameters null players and enables Theorem 4.3's reduction from nd players to md players.
  • domain assumption Pruning can be modeled as replacing a parameter with a placeholder value, and the loss difference under this replacement is a meaningful attribution signal.
    The Shapley value function in Eq. (8) and the codebook objective in Eq. (14) both assume removal is value substitution; the final pruning uses the codebook while Shapley values are computed in a zero-out flavor, and this mismatch is not analyzed.
  • standard math The Chu-Vandermonde identity and the linearity of Shapley values over data samples justify the closed-form solution and unbiasedness claims.
    Used in Appendix A for Theorem 4.3 and in Appendix B.2 for the convergence and unbiasedness analysis.
invented entities (1)
  • Field-aware codebook C in R^{m x d} independent evidence
    purpose: Provides per-field placeholder values to replace zero-padding on pruned embedding parameters during inference.
    The codebook is a concrete model artifact with a closed-form frequency-weighted mean solution, not an unobservable postulated entity. Its empirical benefit is tested in Tables 2 and 3 and Figures 2 and 4.

how reviews work

0 comments
Cite this review

Pith. "Pith review of On-device Content-based Recommendation with Single-shot Embedding Pruning: A Cooperative Game Perspective." pith.science (2026). https://pith.science/paper/DI6NZIOC

@misc{pith2026241113052,
  author       = {Pith},
  title        = {Pith review of: On-device Content-based Recommendation with Single-shot Embedding Pruning: A Cooperative Game Perspective},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DI6NZIOC}},
  note         = {Machine review of arXiv:2411.13052}
}
read the original abstract

Content-based Recommender Systems (CRSs) play a crucial role in shaping user experiences in e-commerce, online advertising, and personalized recommendations. However, due to the vast amount of categorical features, the embedding tables used in CRS models pose a significant storage bottleneck for real-world deployment, especially on resource-constrained devices. To address this problem, various embedding pruning methods have been proposed, but most existing ones require expensive retraining steps for each target parameter budget, leading to enormous computation costs. In reality, this computation cost is a major hurdle in real-world applications with diverse storage requirements, such as federated learning and streaming settings. In this paper, we propose Shapley Value-guided Embedding Reduction (Shaver) as our response. With Shaver, we view the problem from a cooperative game perspective, and quantify each embedding parameter's contribution with Shapley values to facilitate contribution-based parameter pruning. To address the inherently high computation costs of Shapley values, we propose an efficient and unbiased method to estimate Shapley values of a CRS's embedding parameters. Moreover, in the pruning stage, we put forward a field-aware codebook to mitigate the information loss in the traditional zero-out treatment. Through extensive experiments on three real-world datasets, Shaver has demonstrated competitive performance with lightweight recommendation models across various parameter budgets. The source code is available at https://github.com/chenxing1999/shaver

Figures

Figures reproduced from arXiv: 2411.13052 by the authors.

Figure 1
Figure 1. The overview of Shaver. We calculate placeholder values (codebook C), and then compute Shapley values 𝜙𝑣 from the provided dataset D. On any required memory budget 𝐵, we replace embedding parameters with the lowest attribution scores by placeholder values. 4.2 Local and Global Values If straightforward Monte Carlo approximation of 𝜙𝑣 (𝑖, 𝑐) is in use, for all |N | = 𝑛𝑑 embedding parameters, |N ||D | = 𝑛𝑑|D | forward… view at source ↗
Figure 2
Figure 2. Comparative results with single-shot baselines, where DCN-Mix is used as the backbone. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 4
Figure 4. Comparison of how different Shaver variants prune embeddings for fea￾tures in different frequency buckets [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: A graphic explanation on the notion of correspond [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Comparative results with single-shot baselines for DeepFM backbone. [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

87 extracted references · 67 canonical work pages

  1. [1]

    Yi Wang Aden. 2012. KDD Cup 2012, Track 2. https://kaggle.com/competitions/ kddcup2012-track2

    2012

  2. [2]

    Marco Ancona, Cengiz Öztireli, and Markus Gross. 2020. Shapley value as principled metric for structured network pruning.arXiv preprint arXiv:2006.01795 (2020)

    work page Pith review arXiv 2020

  3. [3]

    Omer Ben-Porat and Moshe Tennenholtz. 2018. A game-theoretic approach to recommendation systems with strategic content providers. NeuRIPS (2018)

    2018

  4. [4]

    Han Cai, Chuang Gan, Tianzhe Wang, Zhekai Zhang, and Song Han. 2020. Once- for-all: Train one network and specialize it for efficient deployment.ICLR (2020)

    2020

  5. [5]

    Javier Castro, Daniel Gómez, and Juan Tejada. 2009. Polynomial calculation of the Shapley value based on sampling. Computers & operations research (2009)

    2009

  6. [6]

    Hugh Chen, Ian C Covert, Scott M Lundberg, and Su-In Lee. 2023. Algorithms to estimate Shapley value feature attributions. Nature Machine Intelligence 5, 6 (2023), 590–601

    2023

  7. [7]

    Tong Chen, Hongzhi Yin, Yujia Zheng, Zi Huang, Yang Wang, and Meng Wang

  8. [8]

    Yizhou Chen, Guangda Huzhang, Anxiang Zeng, Qingtao Yu, Hui Sun, Heng-Yi Li, Jingyi Li, Yabo Ni, Han Yu, and Zhiming Zhou. 2023. Clustered Embedding Learning for Recommender Systems. In WWW

    work page 2023

Show all 87 references
  1. [9]

    Chi, and Derek Zhiyuan Cheng

    Benjamin Coleman, Wang-Cheng Kang, Matthew Fahrbach, Ruoxi Wang, Lichan Hong, Ed H. Chi, and Derek Zhiyuan Cheng. 2023. Unified Embedding: Battle- tested feature representations for web-scale ML systems. In NeuRIPS

    2023

  2. [10]

    Ian Covert, Chanwoo Kim, Su-In Lee, James Zou, and Tatsunori Hashimoto. 2024. Stochastic Amortization: A Unified Approach to Accelerate Feature and Data Attribution. arXiv preprint arXiv:2401.15866 (2024)

    2024 arXiv

  3. [11]

    Ian Covert and Su-In Lee. 2021. Improving kernelshap: Practical shapley value estimation via linear regression. Artificial Intelligence and Statistics (2021)

    2021

  4. [12]

    Ian Covert, Scott M Lundberg, and Su-In Lee. 2020. Understanding global feature contributions with additive importance measures. NeuRIPS (2020)

    2020

  5. [13]

    Yunfei Fang, Caihong Mu, and Yi Liu. 2023. AutoShape: Automatic Design of Click-Through Rate Prediction Models Using Shapley Value. In Pacific Rim International Conference on Artificial Intelligence

    2023

  6. [14]

    Amirata Ghorbani and James Zou. 2020. Neuron Shapley: discovering the re- sponsible neurons. In NeuRIPS

    2020

  7. [15]

    Xudong Gong et al . 2022. Real-time short video recommendation on mobile devices. In CIKM. 3103–3112

    2022

  8. [16]

    Hui Guan, Andrey Malevich, Jiyan Yang, Jongsoo Park, and Hector Yuen

  9. [17]

    Jiyang Guan, Zhuozhuo Tu, Ran He, and Dacheng Tao. 2022. Few-shot backdoor defense using shapley estimation. In CVPR

    2022

  10. [18]

    Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, and Xiuqiang He. 2017. DeepFM: A Factorization-Machine based Neural Network for CTR Prediction. In IJCAI, Carles Sierra (Ed.). 1725–1731

    2017

  11. [19]

    Jialiang Han, Yun Ma, Qiaozhu Mei, and Xuanzhe Liu. 2021. Deeprec: On-device deep learning for privacy-preserving sequential recommendation in mobile com- merce. In WWW. 900–911

    2021

  12. [20]

    Song Han, Huizi Mao, and William J. Dally. 2016. Deep Compression: Compress- ing Deep Neural Network with Pruning, Trained Quantization and Huffman Coding. In ICLR, Yoshua Bengio and Yann LeCun (Eds.)

    2016

  13. [21]

    Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural collaborative filtering. In WWW. 173–182

    2017

  14. [22]

    Olivier Chapelle Jean-Baptiste Tien, joycenv. 2014. Display Advertising Challenge. https://kaggle.com/competitions/criteo-display-ad-challenge

    2014

  15. [23]

    Neil Jethani, Mukund Sudarshan, Ian Connick Covert, Su-In Lee, and Rajesh Ranganath. 2021. FastSHAP: Real-time shapley value estimation. In ICLR

    2021

  16. [24]

    Joglekar, Cong Li, Jay K

    Manas R. Joglekar, Cong Li, Jay K. Adams, Pranav Khaitan, and Quoc V. Le. 2019. Neural Input Search for Large Scale Recommendation Models. SIGKDD (2019). https://api.semanticscholar.org/CorpusID:195874115

    2019

  17. [25]

    Wang-Cheng Kang et al. 2021. Learning to Embed Categorical Features without Embedding Tables for Recommendation. In SIGKDD. 840–850

    2021

  18. [26]

    Shuming Kong, Weiyu Cheng, Yanyan Shen, and Linpeng Huang. 2022. Autosrh: An embedding dimensionality search framework for tabular data prediction.IEEE Transactions on Knowledge and Data Engineering 35, 7 (2022), 6673–6686

    2022

  19. [27]

    2023.{AdaEmbed}: Adaptive Embedding for{Large-Scale} Rec- ommendation Models

    Fan Lai et al. 2023.{AdaEmbed}: Adaptive Embedding for{Large-Scale} Rec- ommendation Models. In OSDI

    2023

  20. [28]

    Namhoon Lee, Thalaiyasingam Ajanthan, and Philip Torr. 2018. SNIP: Single-shot network pruning based on connection sensitivity. In ICLR

    2018

  21. [29]

    Shiwei Li et al. 2023. Adaptive Low-Precision Training for Embeddings in Click- through Rate Prediction. In AAAI

    2023

  22. [30]

    Shiwei Li et al . 2024. Embedding Compression in Recommender Systems: A Survey. ACM Comput. Surv. (jan 2024)

    2024

  23. [31]

    Xurong Liang, Tong Chen, Lizhen Cui, Yang Wang, Meng Wang, and Hongzhi Yin. 2024. Lightweight Embeddings for Graph Collaborative Filtering. In SIGIR. 1296–1306

    2024

  24. [32]

    Xurong Liang, Tong Chen, Quoc Viet Hung Nguyen, Jianxin Li, and Hongzhi Yin

  25. [33]

    Siyi Liu, Chen Gao, Yihong Chen, Depeng Jin, and Yong Li. 2021. Learnable Embedding Sizes for Recommender Systems. In ICLR. https://openreview.net/ forum?id=vQzcqQWIS0q

    2021

  26. [34]

    Zirui Liu, Hailin Zhang, Boxuan Chen, Zihan Jiang, Yikai Zhao, Yangyu Tao, Tong Yang, and Bin Cui. 2025. CAFE+: Towards Compact, Adaptive, and Fast Embedding for Large-scale Online Recommendation Models. ACM Transactions on Information Systems (2025)

    2025

  27. [35]

    Scott M Lundberg, Gabriel Erion, Hugh Chen, Alex DeGrave, Jordan M Prutkin, Bala Nair, Ronit Katz, Jonathan Himmelfarb, Nisha Bansal, and Su-In Lee. 2020. From local explanations to global understanding with explainable AI for trees. Nature machine intelligence (2020)

    2020

  28. [36]

    Scott M Lundberg and Su-In Lee. 2017. A unified approach to interpreting model predictions. NeuRIPS (2017)

    2017

  29. [37]

    Fuyuan Lyu et al . 2022. OptEmbed: Learning Optimal Embedding Table for Click-through Rate Prediction. In CIKM. 1399–1409

    2022

  30. [38]

    Matteo Marcuzzo, Alessandro Zangari, Andrea Albarelli, and Andrea Gasparetto

  31. [39]

    Lorenzo Minto and Moritz Haller. 2021. Using Federated Learning to Improve Brave’s On-Device Recommendations While Protecting Your Privacy. (2021). https://brave.com/blog/federated-learning/

    2021

  32. [40]

    Pavlo Molchanov, Stephen Tyree, Tero Karras, Timo Aila, and Jan Kautz. 2017. Pruning Convolutional Neural Networks for Resource Efficient Inference. In ICLR

    2017

  33. [41]

    Dheevatsa Mudigere et al . 2022. Software-Hardware Co-Design for Fast and Scalable Training of Deep Learning Recommendation Models. In Proceedings of the 49th Annual International Symposium on Computer Architecture . 993–1011

    2022

  34. [42]

    Markus Nagel, Marios Fournarakis, Rana Ali Amjad, Yelysei Bondarenko, Mart Van Baalen, and Tijmen Blankevoort. 2021. A white paper on neural network quantization. arXiv preprint arXiv:2106.08295 (2021)

    2021 arXiv

  35. [43]

    Niketan Pansare et al. 2022. Learning compressed embeddings for on-device inference. Proceedings of Machine Learning and Systems 4 (2022), 382–397

    2022

  36. [44]

    Liang Qu, Yonghong Ye, Ningzhi Tang, Lixin Zhang, Yuhui Shi, and Hongzhi Yin

  37. [45]

    Yunke Qu, Tong Chen, Quoc Viet Hung Nguyen, and Hongzhi Yin. 2024. Budgeted Embedding Table For Recommender Systems. In WSDM (WSDM ’24) . 557–566

    2024

  38. [46]

    Yunke Qu, Tong Chen, Xiangyu Zhao, Lizhen Cui, Kai Zheng, and Hongzhi Yin

  39. [47]

    Yunke Qu, Liang Qu, Tong Chen, Xiangyu Zhao, Jianxin Li, and Hongzhi Yin

  40. [48]

    Yunke Qu, Liang Qu, Tong Chen, Xiangyu Zhao, Quoc Viet Hung Nguyen, and Hongzhi Yin. 2024. Scalable Dynamic Embedding Size Search for Streaming Recommendation. arXiv preprint arXiv:2407.15411 (2024)

    2024

  41. [49]

    In SIGIR

    Single-shot Embedding Dimension Search in Recommender System. In SIGIR. 513–522

  42. [50]

    Alvin E Roth. 1988. The Shapley value: essays in honor of Lloyd S. Shapley . Cam- bridge University Press

    1988

  43. [51]

    Benedek Rozemberczki and Rik Sarkar. 2021. The Shapley Value of Classifiers in Ensemble Games. In CIKM. 1558–1567

    2021

  44. [52]

    In SIGIR

    Continuous Input Embedding Size Search For Recommender Systems. In SIGIR. 708–717

  45. [53]

    Statista. 2024. Online advertising revenue in the United States from 2000 to

    2024

  46. [54]

    Will Cukierski Steve Wang. 2014. The Avazu Dataset. https://kaggle.com/ competitions/avazu-ctr-prediction

    2014

  47. [55]

    Hung Vinh Tran, Tong Chen, Quoc Viet Hung Nguyen, Zi Huang, Lizhen Cui, and Hongzhi Yin. 2024. A Thorough Performance Benchmarking on Lightweight Embedding-based Recommender Systems. arXiv preprint arXiv:2406.17335 (2024)

    2024 arXiv

  48. [56]

    Steffen Rendle. 2010. Factorization Machines. In ICDM. 995–1000. https://doi. org/10.1109/ICDM.2010.127

    2010 doi

  49. [57]

    Ruoxi Wang et al. 2021. Dcn v2: Improved deep & cross network and practical lessons for web-scale learning to rank systems. In WWW. 1785–1797

    2021

  50. [58]

    Shuyao Wang, Yongduo Sui, Jiancan Wu, Zhi Zheng, and Hui Xiong. 2024. Dy- namic Sparse Learning: A Novel Paradigm for Efficient Recommendation. In WSDM. 740–749

    2024

  51. [59]

    Hao-Jun Michael Shi, Dheevatsa Mudigere, Maxim Naumov, and Jiyan Yang

  52. [60]

    Xin Xia, Hongzhi Yin, Junliang Yu, Qinyong Wang, Guandong Xu, and Quoc Viet Hung Nguyen. 2022. On-device next-item recommendation with self- supervised knowledge distillation. In SIGIR

    2022

  53. [61]

    Xin Xia, Junliang Yu, Qinyong Wang, Chaoqun Yang, Nguyen Quoc Viet Hung, and Hongzhi Yin. 2023. Efficient on-device session-based recommendation. TOIS (2023)

    2023

  54. [62]

    https://www.statista.com/statistics/183816/us-online-advertising- revenue-since-2000/

    (2024). https://www.statista.com/statistics/183816/us-online-advertising- revenue-since-2000/

    2024

  55. [63]

    Zhiqiang Xu, Dong Li, Weijie Zhao, Xing Shen, Tianbo Huang, Xiaoyun Li, and Ping Li. 2021. Agile and accurate CTR prediction model training for massive- scale online advertising systems. In Proceedings of the International Conference on Management of Data (SIGMOD)

    2021

  56. [64]

    Bencheng Yan, Pengjie Wang, Kai Zhang, Wei Lin, Kuang-Chih Lee, Jian Xu, and Bo Zheng. 2021. Learning Effective and Efficient Embedding via an Adaptively- Masked Twins-based Layer. In CIKM

    2021

  57. [65]

    Qinyong Wang, Hongzhi Yin, Tong Chen, Zi Huang, Hao Wang, Yanchang Zhao, and Nguyen Quoc Viet Hung. 2020. Next point-of-interest recommendation on resource-constrained mobile devices. In WWW

    2020

  58. [66]

    Chunxing Yin, Bilge Acun, Xing Liu, and Carole-Jean Wu. 2021. TT-Rec: Ten- sor Train Compression for Deep Learning Recommendation Models. ArXiv abs/2101.11714 (2021). https://api.semanticscholar.org/CorpusID:231719841

    2021 arXiv

  59. [67]

    Hongzhi Yin, Tong Chen, Liang Qu, and Bin Cui. 2024. On-Device Recommender Systems: A Tutorial on The New-Generation Recommendation Paradigm. In Companion Proceedings of the ACM on Web Conference 2024 . 1280–1283

    2024

  60. [68]

    Yejing Wang, Zhaocheng Du, Xiangyu Zhao, Bo Chen, Huifeng Guo, Ruiming Tang, and Zhenhua Dong. 2023. Single-shot feature selection for multi-task recommendations. In SIGIR. 341–351. WWW ’25, April 28–May 2, 2025, Sydney, NSW, Australia. Tran et al. Figure 5: A graphic explanat...

    2023

  61. [69]

    Jiahui Yu, Linjie Yang, Ning Xu, Jianchao Yang, and Thomas Huang. 2019. Slimmable neural networks. ICLR (2019)

    2019

  62. [70]

    Hailin Zhang et al. 2023. Experimental Analysis of Large-scale Learnable Vector Storage Compression. arXiv:2311.15578 [cs.LG]

    2023 arXiv

  63. [71]

    Xin Xia, Junliang Yu, Guandong Xu, and Hongzhi Yin. 2023. Towards communication-efficient model updating for on-device session-based recom- mendation. In CIKM. 2795–2804

    2023

  64. [72]

    Xiangyu Zhao et al. 2021. AutoDim: Field-aware Embedding Dimension Searchin Recommender Systems. In WWW. 3015–3022

    2021

  65. [73]

    Xiangyu Zhaok et al. 2021. AutoEmb: Automated Embedding Dimensionality Search in Streaming Recommendations. In ICDM. 896–905

    2021

  66. [74]

    Jie Amy Yang, Jianyu Huang, Jongsoo Park, Ping Tak Peter Tang, and An- drew Tulloch. 2020. Mixed-precision embedding using a cache. arXiv preprint arXiv:2010.11305 (2020)

    2020 arXiv

  67. [75]

    Jieming Zhu et al. 2022. BARS: Towards Open Benchmarking for Recommender Systems. In SIGIR. 2912–2923

    2022

  68. [76]

    Jieming Zhu, Jinyang Liu, Shuai Yang, Qi Zhang, and Xiuqiang He. 2021. Open benchmarking for click-through rate prediction. In CIKM. 2759–2769. A Proofs Theorem 4.3. Assume player(𝑖,𝑐) in𝑣(·) is corresponding with player(𝑗,𝑐) in𝑢(·) w.r.t. data instance (x,𝑦). Then, denote the...

    2021

  69. [77]

    Hongzhi Yin, Liang Qu, Tong Chen, Wei Yuan, Ruiqi Zheng, Jing Long, Xin Xia, Yuhui Shi, and Chengqi Zhang. 2024. On-Device Recommender Systems: A Comprehensive Survey. arXiv preprint arXiv:2401.11441 (2024)

    2024 arXiv

  70. [80]

    Hailin Zhang, Zirui Liu, Boxuan Chen, Yikai Zhao, Tong Zhao, Tong Yang, and Bin Cui. 2024. CAFE: Towards Compact, Adaptive, and Fast Embedding for Large-scale Recommendation Models. Proceedings of the ACM on Management of Data (2024)

    2024

  71. [83]

    Ruiqi Zheng, Liang Qu, Tong Chen, Kai Zheng, Yuhui Shi, and Hongzhi Yin

  72. [84]

    TKDE (2024)

    Personalized Elastic Embedding Learning for On-Device Recommendation. TKDE (2024)

    2024

  73. [87]

    (26) By applying Cauchy–Bunyakovsky–Schwarz inequality to the above equation, we receive the required Eq. 24. B Further Details of Shaver B.1 Pseudo code The pseudo code is shown in Algorithm 1. For each data point, we first sample a random orderR for set of players in 𝑢(·) (l...

    2025

  74. [2019]

    arXiv preprint arXiv:1911.02079 (2019)

    Post-training 4-bit quantization on embedding tables. arXiv preprint arXiv:1911.02079 (2019)

    2019 arXiv

  75. [2020]

    Compositional Embeddings Using Complementary Partitions for Memory- Efficient Recommendation Systems. In KDD. 165–175

  76. [2021]

    In SIGKDD

    Learning Elastic Embeddings for Customizing On-Device Recommenders. In SIGKDD. 138–147

  77. [2022]

    IEEE Access 10 (2022), 86578–86623

    Recommendation systems: An insight into current development and future research challenges. IEEE Access 10 (2022), 86578–86623

    2022

  78. [2023]

    arXiv:2309.03518 [cs.IR]

    Learning Compact Compositional Embeddings via Regularized Pruning for Recommendation. arXiv:2309.03518 [cs.IR]

    arXiv

  79. [2024]

    arXiv preprint arXiv:2411.12205 (2024)

    Sparser Training for On-Device Recommendation Systems. arXiv preprint arXiv:2411.12205 (2024)

    2024 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.