Pith. sign in

REVIEW 2 major objections 6 minor 42 references

CSMF: Cascaded Selective Mask Fine-Tuning for Multi-Objective Embedding-Based Retrieval

T0 review · 2 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A single embedding model can retrieve for several objectives at once by splitting its parameters into blocks and reweighting them at serving time, with zero extra storage.

desk verdict CSMF is a promising PackNet-style approach to multi-objective EBR, but the no-overhead fusion formula rests on an unstated block-structure assumption that the paper never verifies. read the letter →

arxiv 2504.12920 v1 pith:FRPN4AU6 submitted 2025-04-17 cs.IR

Hao Deng , Haibo Xing , Kanefumi Matsuyama , Moyu Zhang , Jinxin Hu , Hong Wen , Yu Zhang , Xiaoyi Zeng
show 1 more author
Jing Zhang
This is my paper · ORCID
classification cs.IR
keywords recommendationsystemsembedding-basedretrievalmulti-objectiveoptimizationefficientfine-tuningparameterpruningtwo-towersoftmaxlossonlineserving
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 single two-tower embedding-based retrieval model can be trained to serve several objectives at once, such as exposure, click, and conversion, and then be adjusted online by reweighting objectives with zero additional computation or storage. It proposes CSMF, which sequentially fine-tunes the model on each objective while pruning away redundant neurons from earlier stages to make room for later ones, plus a cumulative percentile pruning method and an adaptive margin loss to manage conflicts. The payoff is Eq. (8): a weighted sum of three objective scores equals a single dot product of a block-weighted user vector with the unchanged item vector, so one approximate-nearest-neighbor index continues to serve arbitrary objective weightings. If true, industrial systems can switch retrieval objectives by changing three scalar weights rather than re-embedding or adding experts.

What carries the argument

The load-bearing identity is Eq. (8), which rewrites the weighted sum of three objective-specific dot products as one dot product of a block-concatenated user vector with the item vector, where the user-side block scalars are accumulated sums of the objective weights. The identity is made available by the CSMF training schedule: a pre-train, selective-mask, accuracy-recovery, fine-tune cascade that partitions the two-tower network into three mutually exclusive parameter sets, one for exposure, one for click, and one for conversion, so that each objective's score is computed by a disjoint prefix of the final embedding. The pruning step is Cumulative Percentile-based Pruning, which per layer prunes the neurons whose cumulative absolute value stays below a fraction $\tau$ of the layer's total, and the conflict-handling step is the Cross-Stage Adaptive Margin Loss, which adaptively widens or narrows the margin between positive and negative items depending on whether upstream objective scores agree.

What would settle it

Take the trained CSMF model, choose any weight triplet, and compute the left and right sides of Eq. (8) over a few thousand random user-item pairs; if the two numbers differ by more than floating-point round-off on a single pair, the serving claim breaks. A cheaper probe is to inspect the pruning masks of both towers and check whether every layer keeps the same neuron indices in both towers.

Watch

Extended reading notes

Core claim

The paper's central discovery is that multi-objective weighted fusion need not increase the embedding dimension or retrieval cost if the model's parameters are partitioned into disjoint blocks, each responsible for one objective. By pre-training on exposure data, then fine-tuning on click data after pruning, then fine-tuning the further-pruned remainder on conversion data, CSMF yields three scores $s_d$, $s_o$, and $s_r$ from the same forward pass. Because each score is a dot product of the user and item sub-vectors of the corresponding parameter block, the linearly weighted score $k_d s_d + k_o s_o + k_r s_r$ collapses, via Eq. (8), to a single dot product in which the user-side block weights are simply $(k_d+k_o+k_r)$, $(k_o+k_r)$, and $k_r$. This lets deployment change the objective mix by reweighting three scalars while the item vectors and the approximate-nearest-neighbor index stay exactly as they were.

Load-bearing premise

The paper's weighted-fusion identity assumes the user tower and item tower are pruned into matching block structures, so the same neuron positions are alive in both towers; if the two towers mask different neurons, the block-weighted dot product no longer equals the summed objective scores.

Editorial extensions

If this is right

  • One unchanged approximate-nearest-neighbor index serves any weighting of exposure, click, and conversion objectives; changing $k_d$, $k_o$, and $k_r$ requires no re-embedding or reindexing.
  • Mixture-of-experts style multi-objective retrieval models can be replaced by CSMF while keeping storage and latency flat; the paper reports 1019.95 MB of vector storage versus 1020.11 MB for a baseline, and ANN time 1.22 ms versus 1.21 ms.
  • On the industrial dataset, CSMF gains 3.51 and 6.61 percent in nDCG@50 and Recall@50 on clicks and 2.23 and 7.51 percent on conversions over the best baseline, with online A/B gains of 0.42 percent RPM, 0.57 percent CTR, and 0.67 percent CVR.
  • Because the weighted fusion uses only the final embeddings, the online serving cost of rebalancing objectives is effectively zero, enabling quick adaptation to changing business priorities.
  • Because cascade training uses the largest dataset first, downstream objectives with sparse data inherit a strong backbone rather than requiring separate expert modules.

Reading between the lines

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

  • The block-reweighting trick generalizes to any number of cascaded objectives: each additional objective adds one parameter block and one more ring of accumulated weights in Eq. (8), so the serving index never grows.
  • Because the weights act only on the user vector, an online system could vary the objective-weight triplet per user, per scene, or even per request without touching the approximate-nearest-neighbor index; the paper only demonstrates fixed global weights in its experiments.
  • A structured variant that fixes the pruning masks of the user and item towers to the same neuron indices at every layer would make Eq. (8) an identity rather than an assumption on the paper's part; checking whether the current masks already align is a cheap and high-value numerical test.
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

2 major / 6 minor

Summary. The paper proposes CSMF, a multi-stage fine-tuning method for two-tower embedding-based retrieval. A backbone is pretrained on exposure data; its least important parameters (under a cumulative-percentile heuristic) are pruned and then fine-tuned on click data; the process is repeated for conversion data. The claimed contribution is that after training, the exposure, click, and conversion scores can be fused online by rescaling blocks of the user embedding and computing a single dot product (Eq. (8)), so that multiple objectives are served from one unchanged ANN index with no extra parameters or retrieval latency. Experiments on an industrial advertising dataset and the AliExpress dataset, plus an online A/B test, report consistent gains over baselines such as MOPPR, MVKE, and DMMP.

Significance. If the central serving claim holds, the paper makes a practically valuable contribution: multi-objective EBR with flexible objective weights at no additional ANN storage or latency. The experimental evidence is substantial, including industrial-scale logs, multiple strong baselines, component ablations, hyperparameter sensitivity, a serving-cost table, and an online A/B test, and the flat storage/latency result in Table 4 is a genuine strength. However, the mathematical derivation that guarantees this efficiency, Eq. (8), depends on a block-decomposition property of the embeddings that the paper neither states nor verifies. The significance is therefore conditional on an architectural clarification and a check of the deployed towers.

major comments (2)
  1. [Section 4.4, Eq. (8)] The derivation of the zero-overhead fusion score is not supported by the training procedure described in Sections 4.1 and 4.2. The equalities e_{θ_d;θ_o} = [e_{θ_d}; e_{θ_o}] and e_θ = [e_{θ_d}; e_{θ_o}; e_{θ_r}] presuppose that the final embeddings are blockwise concatenations of sub-embeddings computed from disjoint parameter blocks. The described CPP operates on individual weights/neurons within each layer and never specifies that the user and item towers are pruned into identical, positionally aligned blocks, nor does it define how the output of a multi-layer network using interleaved θ_d and θ_o decomposes into separate sub-embeddings. Without this architectural constraint, the second and third equalities in Eq. (8) do not follow, and the abstract claim of no additional network parameters and no online retrieval overhead is unsupported. The paper should state the block structure as an explicit construction (e.g., each final-embedding dimension is assigned to one objective block and all upstream connections respect the block partition) and verify it in the deployed system.
  2. [Section 4.2, CPP definition] The pruning rule P(w_kj) = f(c_j ≤ ind_k) is not well-defined unless the neurons are sorted by |w_kj| before forming the cumulative sums C_k. As written, c_j is a cumulative sum over an arbitrary neuron ordering, so pruning the prefix of that ordering has no clear relation to the stated goal of pruning redundant or low-information parameters. The method should either specify that neurons are sorted by magnitude (as in PackNet) or justify a different ordering; otherwise the comparison with fixed-ratio pruning in Table 3 is ambiguous.
minor comments (6)
  1. [Eq. (4) and Section 4.4] The notation {·;·} is overloaded: it denotes parameter-set union in Eq. (4) but vector concatenation in Section 4.4 and Eq. (8). Using distinct notation would prevent confusion.
  2. [Related Work and Table 2] The method name is written as DMTL in Section 2.1 but as DTML in Table 2; the naming should be made consistent.
  3. [Figure 5] The y-axis label in Figure 5 says 'Hitrate@50' while the text and Table 2 use Recall@50; these should be unified.
  4. [Section 4.4] In the sentence describing the weights assigned to the user-side vector, 'e_i_{θ_d}' should likely be 'e_u_{θ_d}'; as written it appears to refer to the item vector.
  5. [Figures 2 and 3] The figure labels in the PDF are difficult to read, with concatenated words and stray symbols (e.g., 'Objective1ExpertsObjective2ExpertsShareExperts'). Redrawing the figures would improve clarity.
  6. [Throughout] The exposure, click, and conversion quantities s_d, s_o, and s_r are called 'probabilities', but Eq. (1) and Eq. (8) define them as dot-product scores; the normalized probability is p_θ in Eq. (2). This terminology should be clarified.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Eq. (8) is a stated block-concatenation identity, and the empirical claims are benchmarked externally.

full rationale

The derivation chain is self-contained. The abstract's serving claim reduces to Eq. (8), which is a linear-algebra identity: if the final user and item embeddings are block concatenations e_u_theta = [e_d; e_o; e_r] and v_i_theta = [v_d; v_o; v_r], then the weighted sum of the three block dot products equals the dot product of a block-rescaled user vector with the item vector. The paper states this decomposition in Section 4.4 ('e_u_theta = {e_u_theta_d; e_u_theta_o; e_u_theta_r}', '{·;·} denotes the vector concatenation operation'), so the equation is a consequence of the stated construction rather than an empirical fit. No fitted parameter is relabeled as a prediction: tau, eta, and the objective weights are ordinary hyperparameters tuned on the same data used for benchmark comparison, which is standard model selection. The performance claims are evaluated against external baselines (YouTubeDNN, DTML, MVKE, DMMP, MOPPR) and a live A/B test, and the ablations compare trained variants rather than re-reporting fit quantities. Citations to prior work by overlapping authors ([39] MOPPR, [7] feature modeling) are used as baselines and background, not as the sole justification of the central mechanism. The block-decomposition and cross-tower mask-alignment assumptions underlying Eq. (8) are under-specified and are correctness risks, but they are not circular: the paper does not use the conclusion as an input to derive itself.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new physical or conceptual entities; its contributions are a training schedule, a pruning rule, a loss modification, and a serving-time weighting identity. The load-bearing assumptions are the cascade structure of the labels, the validity of magnitude-based pruning, and the unstated alignment of pruned blocks between the user and item towers. The central free parameters are tau, eta, and the objective weights, all of which are tuned on the same data that produces the headline numbers.

free parameters (3)
  • pruning ratio tau = 0.75
    Chosen from a sweep over 25% to 95% in Section 5.4, Figure 4; it controls how much parameter space is freed for downstream stages and directly affects the main results.
  • adaptive margin coefficient eta = 1.8
    Selected from the sweep over 1.2 to 2.2 in Section 5.4, Figure 5(a); it scales the corrective margin when upstream and downstream objectives disagree.
  • online objective weights k_d, k_o, k_r = 1, 1.8, 1.2
    Set in Section 5.1.3 and tuned in Section 5.4, Figures 5(b)-(d); the same weights are used in the offline metric tables and the online A/B experiment, so the reported gains are not parameter-free.
assumptions (5)
  • domain assumption User behaviors follow a fixed cascade R subset O subset D subset I, so conversion labels are a subset of click labels, which are a subset of exposure labels.
    Section 3 states this hierarchy, and the entire sequential training order (exposure then click then conversion) depends on it. It matches e-commerce funnels but is not universal for all multi-objective retrieval problems.
  • domain assumption Weight magnitude and cumulative percentile are valid proxies for the information value of a neuron, so pruning low-magnitude parameters preserves most upstream task accuracy.
    Section 4.2 adopts this importance measure from PackNet [24]; CPP only changes the layer-wise ratio. The ability to free space without destroying upstream knowledge rests on this premise.
  • ad hoc to paper The user-tower and item-tower parameter sets are pruned into identical block partitions, so the embedding vectors have matching theta_d, theta_o, and theta_r blocks.
    Equation (8) in Section 4.4 requires block-aligned embeddings. The paper applies CPP per layer without specifying any synchronization rule between towers, so this is an unstated assumption.
  • standard math The inner product is bilinear over block concatenation, so a weighted sum of block dot products equals one dot product of a block-weighted vector with the item vector.
    This is mathematically true for Euclidean vectors and is the basis of Eq. (8), but it is only operationally valid if the block-alignment axiom holds.
  • domain assumption Freezing retained upstream parameters after accuracy recovery does not materially harm downstream task performance.
    Section 4.1 freezes theta_d and later theta_o to prevent catastrophic forgetting; this assumes that frozen upstream features remain useful to downstream tasks after pruning and recovery.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CSMF: Cascaded Selective Mask Fine-Tuning for Multi-Objective Embedding-Based Retrieval." pith.science (2026). https://pith.science/paper/FRPN4AU6

@misc{pith2026250412920,
  author       = {Pith},
  title        = {Pith review of: CSMF: Cascaded Selective Mask Fine-Tuning for Multi-Objective Embedding-Based Retrieval},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FRPN4AU6}},
  note         = {Machine review of arXiv:2504.12920}
}
read the original abstract

Multi-objective embedding-based retrieval (EBR) has become increasingly critical due to the growing complexity of user behaviors and commercial objectives. While traditional approaches often suffer from data sparsity and limited information sharing between objectives, recent methods utilizing a shared network alongside dedicated sub-networks for each objective partially address these limitations. However, such methods significantly increase the model parameters, leading to an increased retrieval latency and a limited ability to model causal relationships between objectives. To address these challenges, we propose the Cascaded Selective Mask Fine-Tuning (CSMF), a novel method that enhances both retrieval efficiency and serving performance for multi-objective EBR. The CSMF framework selectively masks model parameters to free up independent learning space for each objective, leveraging the cascading relationships between objectives during the sequential fine-tuning. Without increasing network parameters or online retrieval overhead, CSMF computes a linearly weighted fusion score for multiple objective probabilities while supporting flexible adjustment of each objective's weight across various recommendation scenarios. Experimental results on real-world datasets demonstrate the superior performance of CSMF, and online experiments validate its significant practical value.

Figures

Figures reproduced from arXiv: 2504.12920 by the authors.

Figure 1
Figure 1. The cascading relationships among user actions. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Methods for multi-objective EBR. (a) Separate EBR models for each objective [ [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Illustration of CSMF Framework. Taking one of the matrices in the user or item tower as an example, the CSMF [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: The performance of CSMF with different pruning [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: The performance of CSMF under different hyperpa [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 7 canonical work pages

  1. [1]

    Martín Abadi, Paul Barham, Jianmin Chen, Zhifeng Chen, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Geoffrey Irving, Michael Isard, et al

  2. [2]

    Zaid Alyafeai, Maged Saeed AlShaibani, and Irfan Ahmad. 2020. A survey on transfer learning in natural language processing. arXiv preprint arXiv:2007.04239 (2020)

    arXiv 2020

  3. [3]

    Zhe Cao, Tao Qin, Tie-Yan Liu, Ming-Feng Tsai, and Hang Li. 2007. Learning to rank: from pairwise approach to listwise approach. In Proceedings of the 24th international conference on Machine learning . 129–136

    2007

  4. [4]

    Michael Crawshaw. 2020. Multi-task learning with deep neural networks: A survey. arXiv preprint arXiv:2009.09796 (2020)

    arXiv 2020

  5. [5]

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

    arXiv 2018

  6. [6]

    Jesse Dodge, Gabriel Ilharco, Roy Schwartz, Ali Farhadi, Hannaneh Hajishirzi, and Noah Smith. 2020. Fine-tuning pretrained language models: Weight initializations, data orders, and early stopping. arXiv preprint arXiv:2002.06305 (2020)

    arXiv 2020

  7. [7]

    Zhifang Fan, Dan Ou, Yulong Gu, Bairan Fu, Xiang Li, Wentian Bao, Xin-Yu Dai, Xiaoyi Zeng, Tao Zhuang, and Qingwen Liu. 2022. Modeling users’ contextualized page-wise feedback for click-through rate prediction in e-commerce search. In Proceedings of the Fifteenth ACM International Conference on Web Search and Data Mining. 262–270

    work page 2022

  8. [8]

    Tianyu Gao, Xingcheng Yao, and Danqi Chen. 2021. Simcse: Simple contrastive learning of sentence embeddings. arXiv preprint arXiv:2104.08821 (2021)

    arXiv 2021

Show all 42 references
  1. [9]

    Soufiane Hayou, Nikhil Ghosh, and Bin Yu. 2024. Lora+: Efficient low rank adaptation of large models. arXiv preprint arXiv:2402.12354 (2024)

    2024 arXiv

  2. [10]

    Yunzhong He, Yuxin Tian, Mengjiao Wang, Feier Chen, Licheng Yu, Mao- long Tang, Congcong Chen, Ning Zhang, Bin Kuang, and Arul Prakash. 2023. Que2engage: Embedding-based retrieval for relevant and engaging products at facebook marketplace. In Companion Proceedings of the ACM W...

    2023

  3. [11]

    Geoffrey Hinton. 2015. Distilling the Knowledge in a Neural Network. arXiv preprint arXiv:1503.02531 (2015)

    2015 arXiv

  4. [12]

    Jeremy Howard and Sebastian Ruder. 2018. Universal language model fine-tuning for text classification. arXiv preprint arXiv:1801.06146 (2018)

    2018 arXiv

  5. [13]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685 (2021)

    2021 arXiv

  6. [14]

    Jui-Ting Huang, Ashish Sharma, Shuying Sun, Li Xia, David Zhang, Philip Pronin, Janani Padmanabhan, Giuseppe Ottaviano, and Linjun Yang. 2020. Embedding- based retrieval in facebook search. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery &...

    2020

  7. [15]

    Po-Sen Huang, Xiaodong He, Jianfeng Gao, Li Deng, Alex Acero, and Larry Heck. 2013. Learning deep structured semantic models for web search using clickthrough data. In Proceedings of the 22nd ACM international conference on Information & Knowledge Management . 2333–2338

    2013

  8. [16]

    Yuchen Jiang, Qi Li, Han Zhu, Jinbei Yu, Jin Li, Ziru Xu, Huihui Dong, and Bo Zheng. 2022. Adaptive domain interest network for multi-domain recommenda- tion. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management. 3212–3221

    2022

  9. [17]

    Jeff Johnson, Matthijs Douze, and Hervé Jégou. 2019. Billion-scale similarity search with GPUs. IEEE Transactions on Big Data (2019), 535–547

    2019

  10. [18]

    Morgan Kaufinann. 2006. Data mining: Concepts and techniques. (2006), 4

    2006

  11. [19]

    Yoon Kim. 2014. Convolutional neural networks for sentence classification.arXiv preprint arXiv:1408.5882 (2014)

    2014 arXiv

  12. [20]

    Shih-Yang Liu, Chien-Yi Wang, Hongxu Yin, Pavlo Molchanov, Yu-Chiang Frank Wang, Kwang-Ting Cheng, and Min-Hung Chen. 2024. Dora: Weight-decomposed low-rank adaptation. arXiv preprint arXiv:2402.09353 (2024)

    2024 arXiv

  13. [21]

    Jiaqi Ma, Zhe Zhao, Xinyang Yi, Jilin Chen, Lichan Hong, and Ed H Chi. 2018. Modeling task relationships in multi-task learning with multi-gate mixture-of- experts. In Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining . 1930–1939

    2018

  14. [22]

    Xiao Ma, Liqin Zhao, Guan Huang, Zhi Wang, Zelin Hu, Xiaoqiang Zhu, and Kun Gai. 2018. Entire space multi-task model: An effective approach for estimating post-click conversion rate. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrie...

    2018

  15. [23]

    Arun Mallya, Dillon Davis, and Svetlana Lazebnik. 2018. Piggyback: Adapting a single network to multiple tasks by learning to mask weights. In Proceedings of the European conference on computer vision (ECCV) . 67–82

    2018

  16. [24]

    Arun Mallya and Svetlana Lazebnik. 2018. Packnet: Adding multiple tasks to a single network by iterative pruning. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition . 7765–7773

    2018

  17. [25]

    Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. 2013. Distributed representations of words and phrases and their compositionality. Advances in neural information processing systems (2013)

    2013

  18. [26]

    pengcheng Li, Runze Li, Qing Da, An-Xiang Zeng, and Lijun Zhang. 2020. Im- proving Multi-Scenario Learning to Rank in E-commerce by Exploiting Task Relationships in the Label Space. In proceedings of the 28th ACM International Conference on Information and Knowledge Management...

    2020

  19. [27]

    Badrul Sarwar, George Karypis, Joseph Konstan, and John Riedl. 2001. Item-based collaborative filtering recommendation algorithms. In Proceedings of the 10th international conference on World Wide Web. 285–295

    2001

  20. [28]

    Atefeh Shahroudnejad. 2021. A survey on understanding, visualizations, and explanation of deep neural networks. arXiv preprint arXiv:2102.01792 (2021)

    2021 arXiv

  21. [29]

    Hongyan Tang, Junning Liu, Ming Zhao, and Xudong Gong. 2020. Progressive layered extraction (ple): A novel multi-task learning (mtl) model for personalized recommendations. In Proceedings of the 14th ACM Conference on Recommender Systems. 269–278

    2020

  22. [30]

    Xu Wang, Jiangxia Cao, Zhiyi Fu, Kun Gai, and Guorui Zhou. 2024. HoME: Hierarchy of Multi-Gate Experts for Multi-Task Learning at Kuaishou. arXiv preprint arXiv:2408.05430 (2024)

    2024 arXiv

  23. [31]

    Yuhao Wang, Ha Tsz Lam, Yi Wong, Ziru Liu, Xiangyu Zhao, Yichao Wang, Bo Chen, Huifeng Guo, and Ruiming Tang. 2023. Multi-task deep recommender systems: A survey. arXiv preprint arXiv:2302.03525 (2023)

    2023 arXiv

  24. [32]

    Yi Xin, Siqi Luo, Haodi Zhou, Junlong Du, Xiaohong Liu, Yue Fan, Qing Li, and Yuntao Du. 2024. Parameter-efficient fine-tuning for pre-trained vision models: A survey. arXiv preprint arXiv:2402.02242 (2024)

    2024

  25. [33]

    Zhenhui Xu, Meng Zhao, Liqun Liu, Lei Xiao, Xiaopeng Zhang, and Bifeng Zhang

  26. [34]

    Qingqing Yi, Jingjing Tang, Yujian Zeng, Xueting Zhang, and Weiqi Xu. 2024. DMMP: A distillation-based multi-task multi-tower learning model for personal- ized recommendation. Knowledge-Based Systems 284 (2024), 111236

    2024

  27. [35]

    Xinyang Yi, Ji Yang, Lichan Hong, Derek Zhiyuan Cheng, Lukasz Heldt, Aditee Kumthekar, Zhe Zhao, Li Wei, and Ed Chi. 2019. Sampling-bias-corrected neural modeling for large corpus item recommendations. In Proceedings of the 13th ACM conference on recommender systems . 269–277

    2019

  28. [36]

    Tianhe Yu, Saurabh Kumar, Abhishek Gupta, Sergey Levine, Karol Hausman, and Chelsea Finn. 2020. Gradient surgery for multi-task learning. Advances in Neural Information Processing Systems 33 (2020), 5824–5836

    2020

  29. [37]

    Jianjin Zhang, Zheng Liu, Weihao Han, Shitao Xiao, Ruicheng Zheng, Yingxia Shao, Hao Sun, Hanqing Zhu, Premkumar Srinivasan, Weiwei Deng, et al. 2022. Uni-retriever: Towards learning the unified embedding based retriever in bing sponsored search. In Proceedings of the 28th ACM...

    2022

  30. [38]

    Zhong Zhao, Yanmei Fu, Hanming Liang, Li Ma, Guangyao Zhao, and Hongwei Jiang. 2021. Distillation based multi-task learning: A candidate generation model for improving reading duration. arXiv preprint arXiv:2102.07142 (2021)

    2021 arXiv

  31. [39]

    Yukun Zheng, Jiang Bian, Guanghao Meng, Chao Zhang, Honggang Wang, Zhixuan Zhang, Sen Li, Tao Zhuang, Qingwen Liu, and Xiaoyi Zeng. 2022. Multi-Objective Personalized Product Retrieval in Taobao Search. arXiv preprint arXiv:2210.04170 (2022)

    2022 arXiv

  32. [40]

    Hongyun Zhou, Xiangyu Lu, Wang Xu, Conghui Zhu, and Tiejun Zhao. 2024. LoRA-drop: Efficient LoRA Parameter Pruning based on Output Evaluation.arXiv preprint arXiv:2402.07721 (2024)

    2024 arXiv

  33. [2016]

    In 12th USENIX symposium on operating systems design and implementation (OSDI 16)

    {TensorFlow}: a system for {Large-Scale} machine learning. In 12th USENIX symposium on operating systems design and implementation (OSDI 16) . 265–283

  34. [2022]

    In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    Mixture of virtual-kernel experts for multi-objective user profile modeling. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 4257–4267

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.