Pith. sign in

REVIEW 4 major objections 5 minor 3 cited by

TD3: Tucker Decomposition Based Dataset Distillation Method for Sequential Recommendation

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Tucker decomposition shrinks sequential recommendation data to a few percent of its size

desk verdict A coherent new combination of Tucker decomposition and feature alignment for sequential-recommendation distillation, but the experimental claims and cost accounting need tightening. read the letter →

arxiv 2502.02854 v2 pith:RYT5TSYP submitted 2025-02-05 cs.IR cs.LG

Jiaqing Zhang , Mingjia Yin , Hao Wang , Yawen Li , Yuyang Ye , Xingyu Lou , Junping Du , Enhong Chen This is my paper · ORCID
classification cs.IRcs.LG
keywords sequentialrecommendationdatasetdistillationTuckerdecompositionbi-leveloptimizationmeta-learningsyntheticdatafeaturespacealignmentRaT-BPTT
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

The paper claims that a sequential recommendation dataset, composed of ordered user-item interaction sequences, can be condensed into a tiny synthetic summary that preserves and sometimes improves the predictive performance of models trained on the full data. The summary is a three-dimensional probability tensor built from a Tucker decomposition into user, temporal, and item factors plus a small core tensor. A bi-level meta-learning loop trains a learner on the synthetic summary, then updates the summary to match both next-item prediction loss and feature representations on the original data. If this works broadly, dataset distillation becomes practical for sequential recommenders with large item sets, cutting training time and memory by orders of magnitude.

What carries the argument

The central object is the Tucker-decomposed synthetic sequence summary $S = G \times_1 U \times_2 T \times_3 V$, where $U$ is a synthetic user latent factor, $T$ is a temporal dynamics latent factor, $V$ is the shared item embedding table (fixed during distillation), and $G$ is a small relation core tensor coupling the three modes. This decomposition makes distillation tractable by replacing a dense tensor of size $\mu \times \zeta \times |\mathcal{V}|$ with factor tensors whose sizes depend on only one dimension each, removing the vocabulary size from the learnable parameter count. The learning loop is bi-level optimization solved with RaT-BPTT: the inner loop trains a learner on the synthetic summary, and the outer loop updates the summary using a joint objective of next-item prediction loss on original data plus a feature-space alignment term (mean squared error between feature maps of the original-trained and synthetic-trained learners).

What would settle it

Train TD3 on a dataset whose user-item-time interaction tensor has high multilinear rank (for example, a dataset with many long-tail items and highly diverse temporal patterns) and check whether models trained on the distilled summary still match full-data performance; if performance collapses while random sampling or Farzi still work, the low-rank Tucker representation is the bottleneck. A simpler calculation is to vary $d_1$ and $d_2$ at a fixed summary size and observe whether the performance gain saturates well below full-data accuracy.

Watch

Extended reading notes

Core claim

The paper establishes that discrete, sequentially ordered interaction data can be distilled in latent space: the synthetic sequence summary $S = G \times_1 U \times_2 T \times_3 V$ is learned so that a model trained on it generalizes comparably to one trained on the original dataset, even when the summary is only about two to three percent of the original size. Experiments on four public datasets with SASRec as the learner show that on two of them (Magazine and Epinions) models trained on distilled summaries outperform models trained on full data, while on MovieLens the gap to full-data training narrows as the summary grows. The carrying mechanism is the Tucker decomposition, which decouples the three dimensions (number of synthetic users, sequence length, item vocabulary) so that the learnable parameter count scales with $\mu$ and $\zeta$ separately rather than with the product $\mu \times \zeta \times |\mathcal{V}|$, and which reuses the pretrained item embedding table as the item factor so no item-side parameters need to be learned. Ablations show that both the feature-space alignment loss and the augmented inner-loop training contribute to the final performance.

Load-bearing premise

The distilled summary is assumed to be representable as a low-rank Tucker decomposition with a fixed pretrained item embedding, so if the information needed for good next-item prediction cannot be captured in that low-rank subspace, the method's performance ceiling is limited regardless of optimization quality.

Editorial extensions

If this is right

  • If TD3 is correct, sequential recommenders can be trained on a distilled summary a few percent the size of the original dataset, cutting training time from hours to seconds on large datasets.
  • The memory advantage of the decomposition grows with item vocabulary size, so the method should become more attractive as catalogs scale.
  • Models trained on the distilled summary transfer to unseen architectures such as GRU4Rec, NARM, and BERT4Rec nearly as well as models trained on full data, indicating the summary captures general sequential signal rather than overfitting the learner.
  • The one-time distillation cost can be amortized across many downstream training runs, making hyperparameter search and architecture exploration substantially cheaper.
  • Because the summary is a probability distribution over the item set, it can in principle be used to train any differentiable sequential model without changing the distillation procedure.

Reading between the lines

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

  • A direct test of the rank bottleneck would be to fix the summary size and vary the ranks $d_1, d_2$ from very small to near full rank; if performance saturates well below full-data accuracy, the Tucker representation itself, not the optimization, is the limiting factor.
  • Since the item factor $V$ is a fixed pretrained embedding, the quality of that embedding caps what the distilled summary can express, so TD3's ceiling is tied to the quality of the item representation it reuses.
  • The feature-space alignment loss is a surrogate that may matter most when the loss landscape is poorly conditioned; on well-conditioned problems the simpler performance-matching objective might suffice, which the ablations partially support.
  • The method could be extended to cross-domain transfer by distilling a summary in one domain and using it as a regularizer or initialization in another, an idea the authors mention as future work but do not test.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes TD3, a dataset distillation method for sequential recommendation. A synthetic sequence summary S is parameterized by a Tucker decomposition S = G ×1 U ×2 T ×3 V, where V is fixed to a pretrained item embedding table, and G, U, T are learned through bi-level optimization. The inner loop trains a learner (SASRec) on synthetic sequences; the outer loop minimizes a test loss on the original data plus a feature-space alignment loss between the original-data model and the synthetic-trained model. Experiments on Magazine, Epinions, ML-100k, and ML-1M compare TD3 against Farzi and sampling baselines, report cross-architecture generalization, and include ablations of the feature-space alignment and augmented learner training components. The paper claims that training on the distilled summary yields performance comparable to or better than full-data training while substantially reducing training time and memory.

Significance. If the claims held, the paper would be a notable step in dataset distillation for sequential recommendation, with a tensor-factorization parameterization that avoids storing a full μ×ζ×|V| synthetic tensor. The code is released, and the ablations in Table 6 give some evidence for the contribution of the two proposed components. However, the central claim is currently supported only under a favorable reading of the tables: Table 2 is inconsistent with Table 3 on the full-data baseline, TD3 is reported without variance in Table 2, and on two datasets Farzi outperforms TD3 on several metrics. More importantly, the synthetic summary is not self-contained because it embeds a pretrained item embedding V and uses pretrained learner initializations from the original data; this weakens the 'distillation replaces the original dataset' framing and complicates the speedup comparison. The method may still be valuable as a data-compression-plus-transfer technique if these issues are addressed, but the current manuscript overstates what is demonstrated.

major comments (4)
  1. [Table 2 vs. Table 3] Table 2's 'Full-Data' column is inconsistent with Table 3. For Magazine HR@10, Table 2 reports Full-Data = 15.44(±2.68), while Table 3 reports Full-Data = 45.40(±0.76); for ML-1M HR@10, Table 2 reports 15.88(±0.22) vs. Table 3's 78.22(±0.28). This is likely a mislabeled or misaligned column, but as printed it makes the comparison against full-data training unreadable. In addition, the TD3 column in Table 2 has no standard deviations, so the claimed advantages over Farzi and over heuristic sampling cannot be judged statistically.
  2. [§3.2, Eq. (7) and §3.3.3, Algorithm 1] The synthetic summary is not self-contained. In Eq. (7), V is the item embedding table pretrained on the original dataset and is fixed during distillation, and Algorithm 1 initializes θ0 from a pool of checkpoints obtained by pretraining on the original data. Consequently, training on S always leverages representations and initializations derived from the full original dataset. This is a form of information leakage relative to the stated goal of replacing the original dataset with S, and it means that Table 4's 'TD3 distillation process' omits the pretraining cost needed to obtain V and the checkpoint pool. For ML-1M, the reported distillation time is 2h27m vs. 57m for full-data training; adding pretraining could reverse the apparent speedup.
  3. [§4.2, Tables 2 and 4] The claimed superiority over Farzi is not supported by the reported numbers. On Magazine, Farzi beats TD3 on HR@10 (47.87 vs. 45.40) and HR@20 (61.17 vs. 56.98); on Epinions, Farzi beats TD3 on HR@10 (19.86 vs. 19.13), HR@20 (31.06 vs. 30.09), NDCG@10 (10.67 vs. 10.25), and NDCG@20 (13.49 vs. 13.00). The text states 'TD3 is more sample-efficient than Farzi' and 'superior data utilization', but the table and Figure 4 need to be reconciled with these results, and the comparison needs variance estimates for TD3.
  4. [§4.3.1, memory complexity] The theoretical memory comparison omits V from TD3's storage count. V has size |V|×d3, which is the same order as Farzi's decoder matrix d×|V|, so the claimed inequality (μ+ζ)d1 + d1^2 d3 ≪ μζd + d|V| ignores a term that is not negligible for large item sets. Either include V in the TD3 footprint or state explicitly that V is assumed to be already available from a separate pretraining step, which is exactly the leakage issue raised above.
minor comments (5)
  1. [§4.1, Evaluation Metrics] The citation '[3, 71? ]' contains a literal placeholder '71? ' that should be resolved to a proper citation.
  2. [Algorithm 1] Line 14 contains a typo: 'Random Truncked Backpropagation Through time' should be 'Random Truncated Backpropagation Through Time'.
  3. [§4.4, Cross-Architecture Generalization] The text cites 'NARM [25]', but References [24] and [25] are identical; the citation numbering for NARM should be fixed.
  4. [Table 4] The caption says 'Wall-clock time is reported in single A100 (80GB) GPU hours', but the table entries are in minutes and seconds; please clarify the unit (e.g., GPU-minutes or hh:mm:ss).
  5. [§4.2, findings] The first finding says TD3 'achieves comparable training performance', but Table 3 shows ML-1M full-data is substantially higher than every distilled size (e.g., HR@10 78.22 vs. at most 74.45); the wording should be qualified to reflect the actual gaps.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: TD3's synthetic summary is an optimized artifact, not a restatement of its inputs.

full rationale

The paper's central claim is empirical: an optimized synthetic tensor S = G ×1 U ×2 T ×3 V, when used to train a sequential recommender, reaches performance comparable to training on the full data. No step in the derivation equates the output to the input by definition. Like all dataset-distillation methods, the outer loop uses the original data to evaluate the synthetic summary (Eq. 5 and Eq. 9); this is the optimization target, not a hidden circularity. The fixed item factor V is explicitly described as the pretrained item embedding table (Section 3.2), and the learner is initialized from a pool of pretrained checkpoints (Section 3.3.3 and Algorithm 1). These choices import information from the original data and should be disclosed in the speedup accounting, but they do not make the performance comparison a tautology: the synthetic sequences still must encode sequential transition patterns, and the reported cross-architecture results (Table 5) are external checks not used to fit S. No load-bearing self-citation or imported uniqueness theorem exists; citations to the authors' prior work, such as [71], appear only as related work and do not carry the derivation. Therefore, no enumerated circular step is present.

Assumptions & free parameters 3 free parameters · 5 assumptions · 4 invented entities

The main assumptions are that a synthetic sequence can be represented as a low-rank Tucker tensor with a fixed pretrained item embedding, and that feature-space alignment to an original-data model improves generalization. The free parameters are the Tucker ranks, truncation window, and learning rates, which are chosen per dataset rather than derived. No new physical entities are proposed; the synthetic factors are internal learned parameters.

free parameters (3)
  • Tucker ranks d1, d2, d3 = d1=d2 in {8,16} for Magazine/Epinions, {16,32,64} for ML-100k/ML-1M; d3 equals item embedding dimension
    Chosen from a small grid per dataset; determines the capacity and complexity of the decomposition.
  • Random truncation window W = 40
    Set for RaT-BPTT; larger W improves gradient fidelity but increases memory; chosen ad hoc.
  • Inner and outer learning rates = alpha in {0.01, 0.03}, eta in {0.003, 0.005, 0.01}
    Standard hyperparameters tuned per dataset.
assumptions (5)
  • domain assumption The synthetic sequence summary can be represented as a three-dimensional probability tensor over the item set, and optimizing it via KL divergence in latent space preserves enough information for training.
    Section 3.2; this is the core modeling assumption of the method.
  • domain assumption A pretrained item embedding table V from a model trained on the original data can be reused as the item factor in the Tucker decomposition without update.
    Section 3.2 states 'V is shared with the trained item embedding table' and Section 3.3 Step 1 pretrains on original data. The transferability of this embedding to synthetic summary generation is assumed.
  • domain assumption RaT-BPTT provides unbiased or sufficiently low-bias gradients for the bi-level optimization.
    Section 3.3.3, based on reference [6]; the paper relies on this to make optimization work.
  • domain assumption Feature-space alignment to the original-data-trained model improves generalization rather than overfitting to that model.
    Section 3.3.2; this is a design hypothesis, not proven.
  • standard math Standard backpropagation through the learner's unrolled optimization, implemented via the Higher package, is numerically valid.
    Assumed throughout Algorithm 1.
invented entities (4)
  • Synthetic user latent factor U
    purpose: Parameterizes each synthetic user as a low-dimensional vector that, with T and V, generates a synthetic sequence summary.
    Internal learned parameter; no external falsifiable prediction.
  • Temporal dynamics latent factor T
    purpose: Captures position or time information in synthetic sequences.
    Internal learned parameter.
  • Relation core G
    purpose: Models interconnections among user, time, and item factors in the Tucker decomposition.
    Internal learned parameter.
  • Synthetic sequence summary S
    purpose: Learned compact dataset used to train recommendation models in place of the original dataset.
    The output of the method, evaluated on held-out data.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TD3: Tucker Decomposition Based Dataset Distillation Method for Sequential Recommendation." pith.science (2026). https://pith.science/paper/RYT5TSYP

@misc{pith2026250202854,
  author       = {Pith},
  title        = {Pith review of: TD3: Tucker Decomposition Based Dataset Distillation Method for Sequential Recommendation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RYT5TSYP}},
  note         = {Machine review of arXiv:2502.02854}
}
read the original abstract

In the era of data-centric AI, the focus of recommender systems has shifted from model-centric innovations to data-centric approaches. The success of modern AI models is built on large-scale datasets, but this also results in significant training costs. Dataset distillation has emerged as a key solution, condensing large datasets to accelerate model training while preserving model performance. However, condensing discrete and sequentially correlated user-item interactions, particularly with extensive item sets, presents considerable challenges. This paper introduces \textbf{TD3}, a novel \textbf{T}ucker \textbf{D}ecomposition based \textbf{D}ataset \textbf{D}istillation method within a meta-learning framework, designed for sequential recommendation. TD3 distills a fully expressive \emph{synthetic sequence summary} from original data. To efficiently reduce computational complexity and extract refined latent patterns, Tucker decomposition decouples the summary into four factors: \emph{synthetic user latent factor}, \emph{temporal dynamics latent factor}, \emph{shared item latent factor}, and a \emph{relation core} that models their interconnections. Additionally, a surrogate objective in bi-level optimization is proposed to align feature spaces extracted from models trained on both original data and synthetic sequence summary beyond the na\"ive performance matching approach. In the \emph{inner-loop}, an augmentation technique allows the learner to closely fit the synthetic summary, ensuring an accurate update of it in the \emph{outer-loop}. To accelerate the optimization process and address long dependencies, RaT-BPTT is employed for bi-level optimization. Experiments and analyses on multiple public datasets have confirmed the superiority and cross-architecture generalizability of the proposed designs. Codes are released at https://github.com/USTC-StarTeam/TD3.

Figures

Figures reproduced from arXiv: 2502.02854 by the authors.

Figure 1
Figure 1. Comparison of the data-centric recommender sys [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. An illustration of Tucker decomposition. The left [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Illustration of TD3. In step 1, the learner is trained to get the best checkpoint for feature space alignment and item [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Illustration of the performance comparison of the [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. FuXi-\beta: Towards a Lightweight and Fast Large-Scale Generative Recommendation Model

    cs.IR 2025-08 conditional novelty 6.0 of 10

    FuXi-β shows that removing query-key attention and using a functional relative time bias makes generative recommendation Transformers faster and, on industrial datasets, more accurate.

  2. Thought-Augmented Planning for LLM-Powered Interactive Recommender Agent

    cs.CL 2025-06 conditional novelty 6.0 of 10

    TAIRA, a thought-pattern-augmented multi-agent recommender, outperforms prior LLM agents in simulated interactive recommendation, with the largest gains on complex user intents.

  3. DLF: Enhancing Explicit-Implicit Interaction via Dynamic Low-Order-Aware Fusion for CTR Prediction

    cs.IR 2025-05 conditional novelty 5.0 of 10

    DLF is a CTR prediction architecture that combines low-rank, high-rank, and implicit interaction blocks with layer-wise attention fusion, reporting state-of-the-art results on Criteo, Avazu, Movielens, and Frappe.

Reference graph

Works this paper leans on

82 extracted references · 56 canonical work pages · cited by 3 Pith papers

  1. [1]

    Dataset distillation by matching training trajectories

    George Cazenavette, Tongzhou Wang, Antonio Torralba, Alexei A Efros, and Jun-Yan Zhu. Dataset distillation by matching training trajectories. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 4750–4759, 2022

    2022

  2. [2]

    Bilevel optimization: theory, algorithms, applications and a bibliography

    Stephan Dempe. Bilevel optimization: theory, algorithms, applications and a bibliography. Bilevel optimization: advances and next challenges , pages 581–672, 2020

    2020

  3. [3]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018

    arXiv 2018

  4. [4]

    Adam: A method for stochastic optimization

    P Kingma Diederik. Adam: A method for stochastic optimization. (No Title), 2014

    2014

  5. [5]

    Fair graph distillation

    Qizhang Feng, Zhimeng Stephen Jiang, Ruiquan Li, Yicheng Wang, Na Zou, Jiang Bian, and Xia Hu. Fair graph distillation. Advances in Neural Information Processing Systems, 36:80644–80660, 2023

    work page 2023

  6. [6]

    Embar- rassingly simple dataset distillation

    Yunzhen Feng, Shanmukha Ramakrishna Vedantam, and Julia Kempe. Embar- rassingly simple dataset distillation. In The Twelfth International Conference on Learning Representations, 2023

    work page 2023

  7. [7]

    Generalized inner loop meta-learning

    Edward Grefenstette, Brandon Amos, Denis Yarats, Phu Mon Htut, Artem Molchanov, Franziska Meier, Douwe Kiela, Kyunghyun Cho, and Soumith Chin- tala. Generalized inner loop meta-learning. arXiv preprint arXiv:1910.01727 , 2019

    arXiv 1910

  8. [8]

    Efficient dataset distillation via minimax diffusion

    Jianyang Gu, Saeed Vahidian, Vyacheslav Kungurtsev, Haonan Wang, Wei Jiang, Yang You, and Yiran Chen. Efficient dataset distillation via minimax diffusion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, pages 15793–15803, 2024

    work page 2024

Show all 82 references
  1. [9]

    Summarizing stream data for memory-restricted online continual learning

    Jianyang Gu, Kai Wang, Wei Jiang, and Yang You. Summarizing stream data for memory-restricted online continual learning. arXiv preprint arXiv:2305.16645, 2, 2023

    2023 arXiv

  2. [10]

    Scaling new frontiers: Insights into large recommendation models

    Wei Guo, Hao Wang, Luankang Zhang, Jin Yao Chin, Zhongzhou Liu, Kai Cheng, Qiushi Pan, Yi Quan Lee, Wanqi Xue, Tingjia Shen, et al. Scaling new frontiers: Insights into large recommendation models. arXiv preprint arXiv:2412.00714 , 2024

    2024 arXiv

  3. [11]

    Mi- rage: Model-agnostic graph distillation for graph classification

    Mridul Gupta, Sahil Manchanda, Sayan Ranu, and Hariprasad Kodamana. Mi- rage: Model-agnostic graph distillation for graph classification. arXiv preprint arXiv:2310.09486, 2023

    2023 arXiv

  4. [12]

    End4rec: Efficient noise-decoupling for multi- behavior sequential recommendation

    Yongqiang Han, Hao Wang, Kefan Wang, Likang Wu, Zhi Li, Wei Guo, Yong Liu, Defu Lian, and Enhong Chen. End4rec: Efficient noise-decoupling for multi- behavior sequential recommendation. arXiv preprint arXiv:2403.17603, 2024

    2024 arXiv

  5. [13]

    Guesr: A global unsupervised data-enhancement with bucket-cluster sampling for sequential recommendation

    Yongqiang Han, Likang Wu, Hao Wang, Guifeng Wang, Mengdi Zhang, Zhi Li, Defu Lian, and Enhong Chen. Guesr: A global unsupervised data-enhancement with bucket-cluster sampling for sequential recommendation. In International Conference on Database Systems for Advanced Applicatio...

    2023

  6. [14]

    Leveraging large language models for sequential recommendation

    Jesse Harte, Wouter Zorgdrager, Panos Louridas, Asterios Katsifodimos, Dietmar Jannach, and Marios Fragkoulis. Leveraging large language models for sequential recommendation. In Proceedings of the 17th ACM Conference on Recommender Systems, pages 1096–1102, 2023

    2023

  7. [15]

    Fusing similarity models with markov chains for sparse sequential recommendation

    Ruining He and Julian McAuley. Fusing similarity models with markov chains for sparse sequential recommendation. In 2016 IEEE 16th international conference on data mining (ICDM) , pages 191–200. IEEE, 2016

    2016

  8. [16]

    Session-based recommendations with recurrent neural networks

    Balázs Hidasi, Alexandros Karatzoglou, Linas Baltrunas, and Domonkos Tikk. Session-based recommendations with recurrent neural networks. arXiv preprint arXiv:1511.06939, 2015

    2015 arXiv

  9. [17]

    Enhancing sequential recommendation via llm-based semantic embedding learning

    Jun Hu, Wenwen Xia, Xiaolu Zhang, Chilin Fu, Weichang Wu, Zhaoxin Huan, Ang Li, Zuoli Tang, and Jun Zhou. Enhancing sequential recommendation via llm-based semantic embedding learning. In Companion Proceedings of the ACM on Web Conference 2024, pages 103–111, 2024

    2024

  10. [18]

    Overcoming data and model heterogeneities in decentralized federated learning via synthetic anchors

    Chun-Yin Huang, Kartik Srinivas, Xin Zhang, and Xiaoxiao Li. Overcoming data and model heterogeneities in decentralized federated learning via synthetic anchors. arXiv preprint arXiv:2405.11525, 2024

    2024 arXiv

  11. [19]

    Self-attentive sequential recommenda- tion

    Wang-Cheng Kang and Julian McAuley. Self-attentive sequential recommenda- tion. In 2018 IEEE international conference on data mining (ICDM) , pages 197–206. IEEE, 2018

    2018

  12. [20]

    A survey on data-centric recom- mender systems

    Riwei Lai, Li Chen, Rui Chen, and Chi Zhang. A survey on data-centric recom- mender systems. arXiv preprint arXiv:2401.17878, 2024

    2024 arXiv

  13. [21]

    A comprehensive survey of dataset distillation

    Shiye Lei and Dacheng Tao. A comprehensive survey of dataset distillation. IEEE Transactions on Pattern Analysis and Machine Intelligence , 2023

    2023

  14. [22]

    Visualiz- ing the loss landscape of neural nets

    Hao Li, Zheng Xu, Gavin Taylor, Christoph Studer, and Tom Goldstein. Visualiz- ing the loss landscape of neural nets. Advances in neural information processing systems, 31, 2018

    2018

  15. [23]

    Text is all you need: Learning language representations for sequen- tial recommendation

    Jiacheng Li, Ming Wang, Jin Li, Jinmiao Fu, Xin Shen, Jingbo Shang, and Julian McAuley. Text is all you need: Learning language representations for sequen- tial recommendation. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages 1258...

    2023

  16. [25]

    Neural attentive session-based recommendation

    Jing Li, Pengjie Ren, Zhumin Chen, Zhaochun Ren, Tao Lian, and Jun Ma. Neural attentive session-based recommendation. In Proceedings of the 2017 ACM on Conference on Information and Knowledge Management , pages 1419–1428, 2017

    2017

  17. [26]

    How to configure good in-context sequence for visual question answering

    Li Li, Jiawei Peng, Huiyi Chen, Chongyang Gao, and Xu Yang. How to configure good in-context sequence for visual question answering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 26710– 26720, 2024

    2024

  18. [27]

    Data distillation for text classification

    Yongqi Li and Wenjie Li. Data distillation for text classification. arXiv preprint arXiv:2104.08448, 2021

    2021 arXiv

  19. [28]

    Meta knowledge condensation for federated learning

    Ping Liu, Xin Yu, and Joey Tianyi Zhou. Meta knowledge condensation for federated learning. arXiv preprint arXiv:2209.14851, 2022

    2022 arXiv

  20. [29]

    Diffusion augmentation for sequential recommendation

    Qidong Liu, Fan Yan, Xiangyu Zhao, Zhaocheng Du, Huifeng Guo, Ruiming Tang, and Feng Tian. Diffusion augmentation for sequential recommendation. In Proceedings of the 32nd ACM International Conference on Information and Knowledge Management, pages 1576–1586, 2023

    2023

  21. [30]

    Augmenting sequential rec- ommendation with pseudo-prior items via reversely pre-training transformer

    Zhiwei Liu, Ziwei Fan, Yu Wang, and Philip S Yu. Augmenting sequential rec- ommendation with pseudo-prior items via reversely pre-training transformer. In Proceedings of the 44th international ACM SIGIR conference on Research and development in information retrieval, pages 160...

    2021

  22. [31]

    Can pre-trained models assist in dataset distillation? arXiv preprint arXiv:2310.03295, 2023

    Yao Lu, Xuguang Chen, Yuchen Zhang, Jianyang Gu, Tianle Zhang, Yifan Zhang, Xiaoniu Yang, Qi Xuan, Kai Wang, and Yang You. Can pre-trained models assist in dataset distillation? arXiv preprint arXiv:2310.03295, 2023

    2023 arXiv

  23. [32]

    Dataset distillation with attention labels for fine-tuning bert

    Aru Maekawa, Naoki Kobayashi, Kotaro Funakoshi, and Manabu Okumura. Dataset distillation with attention labels for fine-tuning bert. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Vol- ume 2: Short Papers) , pages 119–127, 2023

    2023

  24. [33]

    Dilm: Distilling dataset into language model for text-level dataset distillation

    Aru Maekawa, Satoshi Kosugi, Kotaro Funakoshi, and Manabu Okumura. Dilm: Distilling dataset into language model for text-level dataset distillation. arXiv preprint arXiv:2404.00264, 2024

    2024 arXiv

  25. [34]

    Reducing catastrophic forgetting with learning on synthetic data

    Wojciech Masarczyk and Ivona Tautkute. Reducing catastrophic forgetting with learning on synthetic data. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops , pages 252–253, 2020

    2020

  26. [35]

    Understanding and correcting pathologies in the training of learned optimizers

    Luke Metz, Niru Maheswaranathan, Jeremy Nixon, Daniel Freeman, and Jascha Sohl-Dickstein. Understanding and correcting pathologies in the training of learned optimizers. In International Conference on Machine Learning , pages 4556–4565. PMLR, 2019

    2019

  27. [36]

    Landing ai

    Andrew Ng. Landing ai. Landing AI. A vailable online: https://landing.ai/(accessed on 8 February 2023) , 2023

    2023

  28. [37]

    Truncated backpropagation through time and kalman filter training for neurocontrol

    GV Puskorius and LA Feldkamp. Truncated backpropagation through time and kalman filter training for neurocontrol. In Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN’94), volume 4, pages 2488–2493. IEEE, 1994

    1994

  29. [38]

    Sequence-aware recommender systems

    Massimo Quadrana, Paolo Cremonesi, and Dietmar Jannach. Sequence-aware recommender systems. ACM computing surveys (CSUR), 51(4):1–36, 2018

    2018

  30. [39]

    Scale ai

    Alexander Ratner. Scale ai. Snorkel AI. A vailable online: https://snorkel.ai/(accessed on 8 February 2023) , 2023

    2023

  31. [40]

    Factorizing personalized markov chains for next-basket recommendation

    Steffen Rendle, Christoph Freudenthaler, and Lars Schmidt-Thieme. Factorizing personalized markov chains for next-basket recommendation. In Proceedings of the 19th international conference on World wide web , pages 811–820, 2010

    2010

  32. [41]

    Infi- nite recommendation networks: A data-centric approach

    Noveen Sachdeva, Mehak Dhaliwal, Carole-Jean Wu, and Julian McAuley. Infi- nite recommendation networks: A data-centric approach. Advances in Neural Information Processing Systems, 35:31292–31305, 2022

    2022

  33. [42]

    Farzi data: Autoregressive data distillation

    Noveen Sachdeva, Zexue He, Wang-Cheng Kang, Jianmo Ni, Derek Zhiyuan Cheng, and Julian McAuley. Farzi data: Autoregressive data distillation. arXiv preprint arXiv:2310.09983, 2023

    2023 arXiv

  34. [43]

    Data distillation: A survey

    Noveen Sachdeva and Julian McAuley. Data distillation: A survey. arXiv preprint arXiv:2301.04272, 2023

    2023 arXiv

  35. [44]

    Predictive models in sequential recommendations: Bridging performance laws with data quality insights

    Tingjia Shen, Hao Wang, Chuhan Wu, Jin Yao Chin, Wei Guo, Yong Liu, Huifeng Guo, Defu Lian, Ruiming Tang, and Enhong Chen. Predictive models in sequential recommendations: Bridging performance laws with data quality insights. arXiv preprint arXiv:2412.00430, 2024

    2024 arXiv

  36. [45]

    Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer

    Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang. Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer. In Proceedings of the 28th ACM international conference on information and knowledge management , pages...

    2019

  37. [46]

    Personalized top-n sequential recommendation via con- volutional sequence embedding

    Jiaxi Tang and Ke Wang. Personalized top-n sequential recommendation via con- volutional sequence embedding. In Proceedings of the eleventh ACM international conference on web search and data mining , pages 565–573, 2018

    2018

  38. [47]

    The extension of factor analysis to three-dimensional matrices

    Ledyard R Tucker et al. The extension of factor analysis to three-dimensional matrices. Contributions to mathematical psychology , 110119:110–182, 1964

    1964

  39. [48]

    Unbiased gradient estimation in unrolled computation graphs with persistent evolution strategies

    Paul Vicol, Luke Metz, and Jascha Sohl-Dickstein. Unbiased gradient estimation in unrolled computation graphs with persistent evolution strategies. InInternational Conference on Machine Learning , pages 10553–10563. PMLR, 2021

    2021

  40. [49]

    Scale ai

    Alexandr Wang. Scale ai. Scale AI. A vailable online: https://scale.com/(accessed on 8 February 2023), 2023

    2023

  41. [50]

    Gradient matching for categorical data distillation in ctr prediction

    Cheng Wang, Jiacheng Sun, Zhenhua Dong, Ruixuan Li, and Rui Zhang. Gradient matching for categorical data distillation in ctr prediction. In Proceedings of the 17th ACM Conference on Recommender Systems , pages 161–170, 2023. WWW ’25, April 28-May 2, 2025, Sydney, NSW, Austral...

    2023

  42. [51]

    Hypersorec: Exploiting hyperbolic user and item representations with multiple aspects for social-aware recommendation

    Hao Wang, Defu Lian, Hanghang Tong, Qi Liu, Zhenya Huang, and Enhong Chen. Hypersorec: Exploiting hyperbolic user and item representations with multiple aspects for social-aware recommendation. ACM Transactions on Information Systems (TOIS), 40(2):1–28, 2021

    2021

  43. [52]

    Sequential/session-based recommendations: Challenges, approaches, applications and opportunities

    Shoujin Wang, Qi Zhang, Liang Hu, Xiuzhen Zhang, Yan Wang, and Charu Aggarwal. Sequential/session-based recommendations: Challenges, approaches, applications and opportunities. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Informa...

    2022

  44. [53]

    Dataset distillation

    Tongzhou Wang, Jun-Yan Zhu, Antonio Torralba, and Alexei A Efros. Dataset distillation. arXiv preprint arXiv:1811.10959, 2018

    2018 arXiv

  45. [54]

    An aggregation-free federated learning for tackling data heterogeneity

    Yuan Wang, Huazhu Fu, Renuga Kanagavelu, Qingsong Wei, Yong Liu, and Rick Siow Mong Goh. An aggregation-free federated learning for tackling data heterogeneity. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 26233–26242, 2024

    2024

  46. [55]

    Global context enhanced graph neural networks for session-based recommenda- tion

    Ziyang Wang, Wei Wei, Gao Cong, Xiao-Li Li, Xian-Ling Mao, and Minghui Qiu. Global context enhanced graph neural networks for session-based recommenda- tion. In Proceedings of the 43rd international ACM SIGIR conference on research and development in information retrieval , pa...

    2020

  47. [56]

    An efficient gradient-based algorithm for on- line training of recurrent network trajectories

    Ronald J Williams and Jing Peng. An efficient gradient-based algorithm for on- line training of recurrent network trajectories. Neural computation, 2(4):490–501, 1990

    1990

  48. [57]

    Dataset condensation for recommendation

    Jiahao Wu, Wenqi Fan, Shengcai Liu, Qijiong Liu, Rui He, Qing Li, and Ke Tang. Dataset condensation for recommendation. arXiv preprint arXiv:2310.01038, 2023

    2023 arXiv

  49. [58]

    Leveraging large language models (llms) to empower training-free dataset condensation for content-based recommendation

    Jiahao Wu, Qijiong Liu, Hengchang Hu, Wenqi Fan, Shengcai Liu, Qing Li, Xiao- Ming Wu, and Ke Tang. Leveraging large language models (llms) to empower training-free dataset condensation for content-based recommendation. arXiv preprint arXiv:2310.09874, 2023

    arXiv

  50. [59]

    A survey on large language models for recommendation (2023)

    Likang Wu, Zhi Zheng, Zhaopeng Qiu, Hao Wang, Hongchao Gu, Tingjia Shen, Chuan Qin, Chen Zhu, Hengshu Zhu, Qi Liu, et al. A survey on large language models for recommendation (2023). URL: https://arxiv. org/abs/2305.19860, 2023

    2023 arXiv

  51. [60]

    Session-based recommendation with graph neural networks

    Shu Wu, Yuyuan Tang, Yanqiao Zhu, Liang Wang, Xing Xie, and Tieniu Tan. Session-based recommendation with graph neural networks. In Proceedings of the AAAI conference on artificial intelligence , volume 33, pages 346–353, 2019

    2019

  52. [61]

    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. arXiv preprint arXiv:1803.02021 , 2018

    2018 arXiv

  53. [62]

    Tucker decomposition with frequency attention for temporal knowledge graph completion

    Likang Xiao, Richong Zhang, Zijie Chen, and Junfan Chen. Tucker decomposition with frequency attention for temporal knowledge graph completion. In Findings of the Association for Computational Linguistics: ACL 2023 , pages 7286–7300, 2023

    2023

  54. [63]

    Breaking determinism: Fuzzy modeling of sequential recommendation using discrete state space diffusion model

    Wenjia Xie, Hao Wang, Luankang Zhang, Rui Zhou, Defu Lian, and Enhong Chen. Breaking determinism: Fuzzy modeling of sequential recommendation using discrete state space diffusion model. arXiv preprint arXiv:2410.23994, 2024

    2024 arXiv

  55. [64]

    Feddm: Iterative distribution matching for communication-efficient federated learning

    Yuanhao Xiong, Ruochen Wang, Minhao Cheng, Felix Yu, and Cho-Jui Hsieh. Feddm: Iterative distribution matching for communication-efficient federated learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16323–16332, 2023

    2023

  56. [65]

    Graph contextualized self-attention network for session-based recommendation

    Chengfeng Xu, Pengpeng Zhao, Yanchi Liu, Victor S Sheng, Jiajie Xu, Fuzhen Zhuang, Junhua Fang, and Xiaofang Zhou. Graph contextualized self-attention network for session-based recommendation. In IJCAI, volume 19, pages 3940– 3946, 2019

    2019

  57. [66]

    Multi-granularity interest retrieval and refinement network for long-term user behavior modeling in ctr prediction.arXiv preprint arXiv:2411.15005, 2024

    Xiang Xu, Hao Wang, Wei Guo, Luankang Zhang, Wanshan Yang, Runlong Yu, Yong Liu, Defu Lian, and Enhong Chen. Multi-granularity interest retrieval and refinement network for long-term user behavior modeling in ctr prediction.arXiv preprint arXiv:2411.15005, 2024

    2024 arXiv

  58. [67]

    Cosrec: 2d convolutional neural networks for sequential recommendation

    An Yan, Shuo Cheng, Wang-Cheng Kang, Mengting Wan, and Julian McAuley. Cosrec: 2d convolutional neural networks for sequential recommendation. In Proceedings of the 28th ACM international conference on information and knowledge management, pages 2173–2176, 2019

    2019

  59. [68]

    Does graph distillation see like vision dataset counterpart? Advances in Neural Information Processing Systems , 36, 2024

    Beining Yang, Kai Wang, Qingyun Sun, Cheng Ji, Xingcheng Fu, Hao Tang, Yang You, and Jianxin Li. Does graph distillation see like vision dataset counterpart? Advances in Neural Information Processing Systems , 36, 2024

    2024

  60. [69]

    An efficient dataset condensation plugin and its application to continual learning

    Enneng Yang, Li Shen, Zhenyi Wang, Tongliang Liu, and Guibing Guo. An efficient dataset condensation plugin and its application to continual learning. Advances in Neural Information Processing Systems , 36, 2023

    2023

  61. [70]

    Sequential recommendation with latent relations based on large language model

    Shenghao Yang, Weizhi Ma, Peijie Sun, Qingyao Ai, Yiqun Liu, Mingchen Cai, and Min Zhang. Sequential recommendation with latent relations based on large language model. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Ret...

    2024

  62. [71]

    Dataset regeneration for sequential recommendation

    Mingjia Yin, Hao Wang, Wei Guo, Yong Liu, Suojuan Zhang, Sirui Zhao, Defu Lian, and Enhong Chen. Dataset regeneration for sequential recommendation. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 3954–3965, 2024

    2024

  63. [72]

    Apgl4sr: A generic framework with adaptive and personalized global collaborative information in sequential recommendation

    Mingjia Yin, Hao Wang, Xiang Xu, Likang Wu, Sirui Zhao, Wei Guo, Yong Liu, Ruiming Tang, Defu Lian, and Enhong Chen. Apgl4sr: A generic framework with adaptive and personalized global collaborative information in sequential recommendation. In Proceedings of the 32nd ACM Intern...

    2023

  64. [73]

    Dataset distillation: A compre- hensive review

    Ruonan Yu, Songhua Liu, and Xinchao Wang. Dataset distillation: A compre- hensive review. IEEE Transactions on Pattern Analysis and Machine Intelligence , 2023

    2023

  65. [74]

    Hierar- chical item inconsistency signal learning for sequence denoising in sequential recommendation

    Chi Zhang, Yantong Du, Xiangyu Zhao, Qilong Han, Rui Chen, and Li Li. Hierar- chical item inconsistency signal learning for sequence denoising in sequential recommendation. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management , pages 25...

    2022

  66. [75]

    A unified framework for adaptive representation enhancement and inversed learning in cross-domain recommen- dation

    Luankang Zhang, Hao Wang, Suojuan Zhang, Mingjia Yin, Yongqiang Han, Ji- aqing Zhang, Defu Lian, and Enhong Chen. A unified framework for adaptive representation enhancement and inversed learning in cross-domain recommen- dation. In International Conference on Database Systems...

    2024

  67. [76]

    Learn- ing system dynamics without forgetting

    Xikun Zhang, Dongjin Song, Yushan Jiang, Yixin Chen, and Dacheng Tao. Learn- ing system dynamics without forgetting. arXiv preprint arXiv:2407.00717, 2024

    2024 arXiv

  68. [77]

    Navigating complexity: Toward lossless graph condensation via expanding window matching

    Yuchen Zhang, Tianle Zhang, Kai Wang, Ziyao Guo, Yuxuan Liang, Xavier Bres- son, Wei Jin, and Yang You. Navigating complexity: Toward lossless graph condensation via expanding window matching. arXiv preprint arXiv:2402.05011, 2024

    2024 arXiv

  69. [78]

    Dataset condensation with distribution matching

    Bo Zhao and Hakan Bilen. Dataset condensation with distribution matching. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 6514–6523, 2023

    2023

  70. [79]

    Dataset condensation with gradient matching

    Bo Zhao, Konda Reddy Mopuri, and Hakan Bilen. Dataset condensation with gradient matching. arXiv preprint arXiv:2006.05929, 2020

    2006 arXiv

  71. [80]

    Leveraging social connections to improve personalized ranking for collaborative filtering

    Tong Zhao, Julian McAuley, and Irwin King. Leveraging social connections to improve personalized ranking for collaborative filtering. In Proceedings of the 23rd ACM international conference on conference on information and knowledge management, pages 261–270, 2014

    2014

  72. [81]

    Recbole 2.0: Towards a more up-to-date recommendation library

    Wayne Xin Zhao, Yupeng Hou, Xingyu Pan, Chen Yang, Zeyu Zhang, Zihan Lin, Jingsen Zhang, Shuqing Bian, Jiakai Tang, Wenqi Sun, Yushuo Chen, Lanling Xu, Gaowei Zhang, Zhen Tian, Changxin Tian, Shanlei Mu, Xinyan Fan, Xu Chen, and Ji-Rong Wen. Recbole 2.0: Towards a more up-to-d...

    2022

  73. [82]

    Enhanced self-attention mechanism for long and short term sequential recommendation models

    Xiaoyao Zheng, Xingwang Li, Zhenghua Chen, Liping Sun, Qingying Yu, Liang- min Guo, and Yonglong Luo. Enhanced self-attention mechanism for long and short term sequential recommendation models. IEEE Transactions on Emerging Topics in Computational Intelligence, 2024

    2024

  74. [83]

    Filter-enhanced mlp is all you need for sequential recommendation

    Kun Zhou, Hui Yu, Wayne Xin Zhao, and Ji-Rong Wen. Filter-enhanced mlp is all you need for sequential recommendation. In Proceedings of the ACM web conference 2022, pages 2388–2399, 2022

    2022

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.