Pith. sign in

REVIEW 3 major objections 6 minor 106 references

OneShot's claim: train the retrieval index itself under the ranking objective, and billion-scale search can escape the dot-product bottleneck — with measured +20% recall and 10x efficiency gains in a deployed recommender.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-03 01:38 UTC pith:ZQXKM66M

load-bearing objection A plausible, well-engineered industrial retrieval paper whose big causal claim — that index-in-ranking, not just neural scoring capacity, drives the +20% recall — is not actually isolated by its own experiments. the 3 major comments →

arxiv 2607.27475 v2 pith:ZQXKM66M submitted 2026-07-29 cs.IR cs.LG

OneShot: Index-in-Ranking with Neural Scoring for Large-Scale Retrieval

Ziwei Li , Shuyao Li , Xufeng Cai , Xue Zou , Yiming Ma , Huiting Lu , Wujie Yan , Zhichen Zhao
show 7 more authors
Yang Lu Zhe Wang Rui Luo Zhengyu Su Dan Zhang Yimin Tan Ji Liu
This is my paper
classification cs.IR cs.LG
keywords in-model indexingindex-in-rankingneural scoringretrievalrecommendation systemsstochastic compositional optimizationindex balancingvector quantization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper attacks a mismatch at the heart of large-scale recommendation: the model that ranks items is trained on user behavior, but the index that finds candidates is built afterwards, by clustering item embeddings so they can be searched quickly. The two objectives do not point the same way, and the separate index forces retrieval scoring to stay a simple dot product. OneShot proposes to learn the index inside the ranking model — each item gets a multi-layer one-hot code, and the ranking loss is applied directly to the summed code scores — so the index and the ranker optimize the same thing. That alignment is what lets the scoring functions become arbitrary neural networks, and a theory-grounded balancing term keeps the learned index from collapsing into overused clusters. If the paper is right, retrieval can be simultaneously more accurate (a reported +20% offline recall) and cheaper (10x fewer items ranked at matched recall), with the approach validated in full production.

Core claim

The central claim is that index and ranker can be one model. OneShot assigns each item a multi-layer one-hot code, scores each code layer against the user with a — possibly nonlinear — network, and trains the whole code hierarchy directly under the sampled-softmax ranking loss, with a dedicated balancing term (Eq. 4.7). Because the codebooks are optimized by the ranking objective itself, item embeddings no longer need the dot-product-friendly geometry that approximate-nearest-neighbor indexes require, so neural scoring becomes viable in retrieval. The paper shows recall rising with interaction width and depth (0.4016 to 0.4613), reports a 2-layer index matching a 16384-cluster baseline while

What carries the argument

The central object is the index-in-ranking loss of Eq. (4.7): per layer, the sampled-softmax ranking loss is applied to the sum of code scores over layers up to and including that layer — a nested summation that works as residual boosting — plus a dense-embedding ranking term and a balancing term. Three mechanisms carry it. The Straight-Through Estimator keeps the forward pass hard (one-hot codes) while backpropagating through soft assignment probabilities, so the discrete index stays differentiable. A decoupled dense embedding frees the codes from reconstruction objectives and lets the scorers be arbitrary neural networks on user–item pairs. And the balancing surrogate, Eq. (4.9) — the inne

Load-bearing premise

The convergence guarantee for the balancing loss is proven only for soft cluster assignments, but the deployed system tracks hard one-hot assignments through a straight-through gradient surrogate; the balanced-index property — and with it the recall and efficiency gains — depends on that unproven surrogate behaving like the proven one.

What would settle it

Two checks could settle the load-bearing claim. First, the convergence theorem covers softly-assigned objectives while the deployed system tracks hard one-hot assignments with a straight-through gradient proxy: train identical models with the soft tracker and with the hard tracker, and see whether recall or the max/mean cluster ratio diverges. Second, the balancing term is meant to be global: re-measure cluster-size statistics on a deduplicated corpus an order of magnitude larger than the evaluation set; if the P99/mean ratio (2.26 under OneShot) drifts toward the no-balancing baseline (190),

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Retrieval interaction models can now scale the way ranking models do: recall rises monotonically as the neural scorer's width and depth grow (0.4016 to 0.4613), a scaling regime previously reserved for the ranking stage.
  • At matched recall, indexing gets roughly 10x cheaper: a 2-layer OneShot index ranks 90% fewer items than the 16384-cluster ANN baseline at the same recall.
  • Discrete codes carry most of the dense model's predictive power — layer 4 of a 4-layer index reaches about 80% of dense like@10 — so the index itself can act as the item representation when serving compute is scarce.
  • Assignments matter more than codebooks: with a frozen semantic-ID codebook, merely learning the item-to-code assignment recovers most of the fully-learned EID recall (0.4230 vs 0.4308, versus 0.1354 for frozen SIDs), which points toward token-based generative recommenders.
  • The balancing surrogate transfers across domains: as a Mixture-of-Experts load-balancing loss in LLM pretraining, the KL form matches the standard Switch loss at usual strength and, under reduced balancing weight, leaves no dead experts.

Where Pith is reading between the lines

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

  • If the ranking objective can fully determine the index, the classic pipeline's two separate steps — learn embeddings, then build a proximity index — become one learning problem. A sharper test of that thesis would be training retrieval with no proximity-or-reconstruction term at all, supervised purely by engagement, which this paper approaches but does not quite isolate.
  • The Appendix's KL-versus-Switch comparison exposes a general principle about balancing any discrete allocation: a logarithmically growing penalty protects under-used buckets (raising the valley, eliminating dead experts) at the cost of a higher peak. That principle should transfer to other load-balancing problems — learned hash tables, cluster assignment in distributed storage, token bucket alloca
  • The EID result suggests codes group items by co-engagement rather than meaning; a hybrid that appends learned engagement codes to semantic IDs is the paper's own suggested direction. A natural test is cold start: semantic structure should matter most for items with little engagement history, so a hybrid's advantage over pure EIDs should be largest there.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. OneShot proposes an end-to-end, in-model hierarchical index for large-scale retrieval, replacing the conventional decoupled ANN index with multi-layer one-hot codebooks trained by the same sampled-softmax ranking losses as the dense scoring stage. The method adds a logit-level boosting scheme across layers, swaps dot-product scoring for nonlinear neural scoring on both code and dense branches, and regularizes codebook balance using a surrogate derived from stochastic compositional gradient descent (SCGD). Experiments on Instagram short-video retrieval compare OneShot against k-means ANN, reporting +20% relative recall at the 1% ranking-volume operating point, roughly 10x efficiency gains in a two-layer setting, improved cluster-balance statistics, and small positive online A/B metrics. The paper also applies the balancing surrogate to MoE routing and compares OneShot-generated item IDs with semantic IDs.

Significance. If the claims hold, this is a significant contribution to the retrieval literature: it directly addresses the long-standing misalignment between index construction and ranking objectives, and it demonstrates that nonlinear interaction models can be used in the indexing stage rather than being confined to late-stage ranking. The paper's strengths include a clean architectural formulation, a principled SCGD derivation of the balancing surrogate, and a realistic industrial evaluation that uses the full serving pipeline and reports deployment metrics. The balancing method is also shown to transfer to MoE load balancing, which broadens the impact. However, the central causal claim about index-in-ranking is not yet isolated from the capacity gains of neural scoring, and the theoretical guarantee is proved only for a soft-assignment surrogate while the deployed system uses hard assignments with straight-through estimation. Reproducibility is further limited by the absence of code/data and by the lack of statistical error bars on the online metrics.

major comments (3)
  1. [§5.1.1, Fig. 3, Table 2] The central claim that 'index-in-ranking' is what enables neural scoring and produces the reported recall gains is confounded. Every offline comparison pits the full OneShot system (neural scoring for both code and dense stages, jointly trained index) against a k-means ANN baseline that uses dot-product scoring and a decoupled index. The +20% recall improvement could therefore be due primarily to the added capacity of the neural scorer, independent of how the index is trained. The paper needs a controlled ablation, for example: (a) a fixed k-means index combined with the same neural scoring functions NN_l and NN_d used by OneShot, optionally with the scorer trained after clustering; and (b) OneShot with dot-product scoring in place of neural scoring. Without these conditions, the paper's foundational contribution—that aligning index learning with ranking objectives is what makes neural s
  2. [§4.3, Appendix B, Eq. (B.5)] Theorem 4.1 proves the surrogate property and convergence guarantee for the soft-assignment objective, where the inner map is p_i^(l). In deployment, however, the usage tracker qhat is updated from hard assignments e_i^(l) (Eq. B.5), and gradients are obtained by replacing the derivative of the hard assignment with that of p_i^(l) via straight-through estimation. The theorem's almost-sure convergence guarantee does not apply to this hard-assignment/STE objective. The paper transparently acknowledges this gap, but then continues to describe the method as 'theoretically grounded.' Because the balancing mechanism is central to the index-quality and efficiency claims, the authors should either extend the convergence/bias analysis to the deployed hard-assignment tracker under explicitly stated conditions, or reframe the hard-assignment version as a heuristic and clearly separate the proven so
  3. [§5.1.2, Table 3] The online A/B results are presented as percentage gains without confidence intervals, significance tests, experiment duration, or number of users/days. Given that the reported gains are small in absolute terms (+0.035% daily sessions, +0.136% time-spent), the statement that these are 'significant wins' cannot be verified from the information provided. Please add statistical significance measures, confidence intervals, and a description of the A/B testing methodology (e.g., number of users, number of days, any guardrail metrics). Without this, the online validation is not quantitatively assessable.
minor comments (6)
  1. [§1, §2] The description of NANN's 'flat' constraint on the neural scoring landscape is asserted without a precise reference to the specific equation or formal condition in Chen et al. (2022). Please add a citation to the relevant definition or lemma.
  2. [§4.3, Table 4] The hyperparameters λ_marg, λ_joint, and the EMA coefficient ρ are not reported for the main balancing experiments. Please provide the values used in Table 4 and any sensitivity analysis.
  3. [§4.3, Eq. (4.9)] The definition of B_item should be made more precise: it is a deduplicated item minibatch uniformly sampled from the candidate corpus. Clarify how this differs from the user-item impression minibatch B and how it is constructed in practice.
  4. [§5.1.1, Table 2] The caption states that K=0 is 'a simple linear projection from U^T V to task logits.' This is still not the exact dot-product scoring used by the k-means ANN baseline. Please clarify whether the baseline was also given this linear projection, and state explicitly which components differ between the two systems.
  5. [Appendix C] The MoE experiments would benefit from additional details on the C4 data preprocessing, tokenizer, exact training compute, and optimizer schedules. As written, the comparison is difficult to reproduce independently.
  6. [Figures 2 and 6] The figures are dense and the text labels are small. Please increase font sizes and consider separating the forward/backward gradient paths more clearly in Figure 2.

Circularity Check

0 steps flagged

No circular dependency: OneShot's empirical claims are measurements and its balanced-index surrogate is an external-theory-backed derivation.

full rationale

Walking the claimed derivation chain: Eq. (4.1) defines index-in-ranking as L_rank applied to code embeddings; Eq. (4.7) sums such terms with dense ranking and balancing losses. These are constructions, not predictions of the observed recall/efficiency numbers. The recall, cluster-balance, and online metrics in Tables 2–4 and Figure 3 are measured on held-out serving data or A/B tests, with no fitted constant later reported as a forecast. The balancing regularizer (4.9) is derived algebraically as an SCGD surrogate, and the convergence statement explicitly invokes the external theorem of Wang et al. (2017); no self-citation carries the load and the derivation does not assume the recall result. The one admitted gap—Theorem 4.1 is proved for soft assignments while deployment uses hard-assignment EMA plus STE (Section 4.3, Appendix B Eq. B.5)—is a soundness/coverage gap, not circular reasoning: the hard-assignment variant is not claimed to follow from, nor to be equivalent to, the theorem. The causal attribution of the +20% recall to index-in-ranking rather than to neural-scoring capacity is under-supported by ablations, but that is an experimental identification issue, not a definitional reduction of the result to its inputs.

Axiom & Free-Parameter Ledger

7 free parameters · 6 axioms · 1 invented entities

The paper's empirical claims rest mostly on standard assumptions about sampled softmax, STE training, and SCGD convergence. The main unproved step is the hard-assignment STE variant of the balancing regularizer, which is load-bearing for the deployed index quality. Several industrial hyperparameters are undisclosed.

free parameters (7)
  • Codebook sizes and number of layers (L, N_l) = 1-layer 16384; 2-layer 2048/1024; 4-layer 2048 each
    Architecture capacity hyperparameters controlling index selectivity versus serving volume; varied across experiments.
  • Sampled-softmax inverse temperature beta = not reported
    Temperature in Eq. (3.1) controls contrastive loss sharpness; not disclosed.
  • Soft-assignment temperature T = not reported
    Eq. (4.4) temperature controlling hard/soft assignment sharpness.
  • Gumbel noise strength gamma = 0.1 (vs 0)
    Appendix A reports gamma=0.1 gives +0.5% recall but slightly worse cluster balance; treated as a small tuned default.
  • Segment-wise interaction dimensions d', m, n = d'=256, m=12, n=1 (m also swept)
    Section 4.2: shapes chosen to keep O(B^2) nonlinear evaluations tractable.
  • Balancing coefficients lambda_marg, lambda_joint = e.g., (10,0) and (5,3) in Table 6
    Eq. (B.7) weights marginal versus joint path balancing; tuned per experiment.
  • EMA coefficient rho for usage tracker = not reported
    Appendix B Eq. (B.5) tracker smoothing hyperparameter.
axioms (6)
  • standard math SCGD convergence theorem of Wang et al. (2017) applies to the soft-assignment surrogate (4.9).
    Used in the proof of Theorem 4.1 to claim almost-sure convergence to a stationary point for the soft objective.
  • domain assumption Sampled softmax with logQ correction provides an unbiased estimate of the full-corpus softmax loss.
    Footnote 1 invokes Yi et al. 2019, Wu et al. 2024, Jean et al. 2015; essential for ranking-loss soundness under negative sampling.
  • domain assumption Balancing against the uniform distribution over the deduplicated candidate corpus is the right serving-efficiency target.
    Section 4.3: uniform cluster usage is assumed necessary for beam-search serving; if uniform balance is not sufficient, efficiency gains could differ.
  • ad hoc to paper The hard-assignment EMA tracker with STE gradients inherits the SCGD guarantee of Theorem 4.1.
    Section 4.3/Appendix B: theorem is proved for soft p, but deployed qhat tracks hard e with gradient proxy through soft p. This is an unproved bridge.
  • ad hoc to paper Straight-through estimation gives usable gradients for discrete one-hot codes.
    Section 4.1.2: standard practice but biased; no formal guarantee for the resulting discrete index optimization.
  • standard math Gumbel-Max trick produces exact categorical samples from the softmax assignment distribution.
    Appendix A Fact A.1; used for stochastic code exploration with gamma=0.1.
invented entities (1)
  • Engagement ID (EID) independent evidence
    purpose: Represents each item by its multi-layer one-hot code path, serving as a discrete item identifier for generative recommendation and retrieval.
    Section 5.3 Table 5 compares EID with semantic IDs in the same retrieval model, providing an internal offline metric but not an external benchmark.

pith-pipeline@v1.3.0-alltime-deepseek · 18168 in / 14184 out tokens · 154412 ms · 2026-08-03T01:38:18.324750+00:00 · methodology

0 comments
read the original abstract

In modern recommendation systems, retrieval serves as a primary stage responsible for filtering billions of candidate items down to thousands prior to refined ranking. To make this massive search effective and efficient, the system relies on ranking accuracy and indexing efficiency. However, these two objectives are traditionally misaligned: while the former optimizes for the alignment between ranking predictions and user behavior, the latter optimizes for a structural grouping of item representations which enables fast search among billions of candidates. Thus, despite extensive efforts to scale up interaction modeling for retrieval, they remain fundamentally limited by the structural misalignment between the ranking objectives and the proximity-learned index. In this work, we address this long-standing dichotomy by proposing a new holistic retrieval framework, OneShot. It is an end-to-end, in-model index learning framework that natively aligns index learning with ranking objectives. Using this joint learning as a structural foundation, OneShot pushes the boundaries of retrieval expressiveness by scaling interaction modeling with neural scoring beyond the persistent dot-product bottleneck. OneShot is fully deployed in Instagram's industrial short-video recommendation system, driving significant wins in user daily sessions, engagement, and time-spent. Additionally, OneShot achieves a $20\%$ recall gain at the operational ranking volume and a 10x efficiency improvement at an equivalent recall level.

Review history (2 revisions) →

discussion (0)

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

Reference graph

Works this paper leans on

106 extracted references · 14 linked inside Pith

  1. [1]

    Deng, Jiaxin and Wang, Shiyao and Cai, Kuo and Ren, Lejian and Hu, Qigen and Ding, Weifeng and Luo, Qiang and Zhou, Guorui , eprint=

  2. [2]

    IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    Autoregressive image generation using residual quantization , author=. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

  3. [3]

    IEEE Transactions on Communications , volume=

    An algorithm for vector quantizer design , author=. IEEE Transactions on Communications , volume=. 1980 , publisher=

    1980

  4. [4]

    International Joint Conference on Artificial Intelligence (IJCAI) , year=

    Latent class models for collaborative filtering , author=. International Joint Conference on Artificial Intelligence (IJCAI) , year=

  5. [5]

    IEEE International Conference on Data Mining (ICDM) , year=

    A scalable collaborative filtering framework based on co-clustering , author=. IEEE International Conference on Data Mining (ICDM) , year=

  6. [6]

    ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

    Predictive discrete latent factor models for large scale dyadic data , author=. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

  7. [7]

    ACM International Conference on Information and Knowledge Management (CIKM) , year=

    Jointly Optimizing Query Encoder and Product Quantization to Improve Retrieval Performance , author=. ACM International Conference on Information and Knowledge Management (CIKM) , year=

  8. [8]

    ACM International Conference on Web Search and Data Mining (WSDM) , year=

    Learning Discrete Representations via Constrained Clustering for Effective and Efficient Dense Retrieval , author=. ACM International Conference on Web Search and Data Mining (WSDM) , year=

  9. [9]

    1603.04467 , archivePrefix=

    Tensorflow: Large-scale machine learning on heterogeneous distributed systems , author=. 1603.04467 , archivePrefix=

    Pith/arXiv arXiv

  10. [10]

    ACM Conference on Recommender Systems (RecSys) , year=

    Speeding up the xbox recommender system using a euclidean transformation for inner-product spaces , author=. ACM Conference on Recommender Systems (RecSys) , year=

  11. [11]

    International Conference on Machine Learning (ICML) , year=

    Fast nearest neighbor retrieval for bregman divergences , author=. International Conference on Machine Learning (ICML) , year=

  12. [12]

    Statistical Analysis and Data Mining: The ASA Data Science Journal , volume=

    Dual-tree fast exact max-kernel search , author=. Statistical Analysis and Data Mining: The ASA Data Science Journal , volume=

  13. [13]

    SIAM International Conference on Data Mining (SDM) , year=

    Fast exact max-kernel search , author=. SIAM International Conference on Data Mining (SDM) , year=

  14. [14]

    ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

    Embedding-based retrieval in facebook search , author=. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

  15. [15]

    ACM International Conference on Information and Knowledge Management (CIKM) , year=

    Multi-interest network with dynamic routing for recommendation at Tmall , author=. ACM International Conference on Information and Knowledge Management (CIKM) , year=

  16. [16]

    Information Systems , volume=

    Approximate nearest neighbor algorithm based on navigable small world graphs , author=. Information Systems , volume=

  17. [17]

    1908.06887 , archivePrefix=

    Relevance Proximity Graphs for Fast Relevance Retrieval , author=. 1908.06887 , archivePrefix=

    Pith/arXiv arXiv 1908

  18. [18]

    The VLDB Journal , volume=

    Searching in metric spaces by spatial approximation , author=. The VLDB Journal , volume=

  19. [19]

    ACM International Conference on Information and Knowledge Management (CIKM) , year=

    Search-based user interest modeling with lifelong sequential behavior data for click-through rate prediction , author=. ACM International Conference on Information and Knowledge Management (CIKM) , year=

  20. [20]

    ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

    Maximum inner-product search using cone trees , author=. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

  21. [21]

    ACM Web Conference (WWW) , year=

    Asymmetric minwise hashing for indexing binary inner products and set containment , author=. ACM Web Conference (WWW) , year=

  22. [22]

    ACM International Conference on Web Search and Data Mining (WSDM) , year=

    Fast item ranking under neural network based measures , author=. ACM International Conference on Web Search and Data Mining (WSDM) , year=

  23. [23]

    ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

    Deep interest network for click-through rate prediction , author=. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

  24. [24]

    ACM Conference on Recommender Systems (RecSys) , year=

    Field-aware factorization machines for CTR prediction , author=. ACM Conference on Recommender Systems (RecSys) , year=

  25. [25]

    IEEE International Conference on Data Mining (ICDM) , year=

    Factorization machines , author=. IEEE International Conference on Data Mining (ICDM) , year=

  26. [26]

    ACM International Conference on Information and Knowledge Management (CIKM) , year=

    Candidate generation with binary codes for large-scale top-n recommendation , author=. ACM International Conference on Information and Knowledge Management (CIKM) , year=

  27. [27]

    ACM Web Conference (WWW) , year=

    Item-based collaborative filtering recommendation algorithms , author=. ACM Web Conference (WWW) , year=

  28. [28]

    com recommendations: Item-to-item collaborative filtering , author=

    Amazon. com recommendations: Item-to-item collaborative filtering , author=. IEEE Internet Computing , volume=. 2003 , publisher=

    2003

  29. [29]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Probabilistic matrix factorization , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  30. [30]

    Computer , volume=

    Matrix factorization techniques for recommender systems , author=. Computer , volume=. 2009 , publisher=

    2009

  31. [31]

    1703.04247 , archivePrefix=

    DeepFM: a factorization-machine based neural network for CTR prediction , author=. 1703.04247 , archivePrefix=

    Pith/arXiv arXiv

  32. [32]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Optimized product quantization , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2014 , publisher=

    2014

  33. [33]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Scalable nearest neighbor algorithms for high dimensional data , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2014 , publisher=

    2014

  34. [34]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Rank-based similarity search: Reducing the dimensional dependence , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2015 , publisher=

    2015

  35. [35]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Joint optimization of tree-based index and deep model for recommender systems , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  36. [36]

    1806.09464 , archivePrefix=

    Learning K-way D-dimensional discrete codes for compact embedding representations , author=. 1806.09464 , archivePrefix=

    Pith/arXiv arXiv

  37. [37]

    1703.05160 , archivePrefix=

    A New Unbiased and Efficient Class of LSH-Based Samplers and Estimators for Partition Function Computation in Log-Linear Models , author=. 1703.05160 , archivePrefix=

    Pith/arXiv arXiv

  38. [38]

    Asymmetric

    Shrivastava, Anshumali and Li, Ping , booktitle=. Asymmetric

  39. [39]

    International Conference on Machine Learning (ICML) , year=

    Learning optimal tree models under beam search , author=. International Conference on Machine Learning (ICML) , year=

  40. [40]

    Journal of the Royal Statistical Society: Series B (Methodological) , volume=

    Maximum likelihood from incomplete data via the EM algorithm , author=. Journal of the Royal Statistical Society: Series B (Methodological) , volume=. 1977 , publisher=

    1977

  41. [41]

    Extreme classification in log memory using count-min sketch: A case study of

    Medini, Tharun Kumar Reddy and Huang, Qixuan and Wang, Yiqiu and Mohan, Vijai and Shrivastava, Anshumali , booktitle=. Extreme classification in log memory using count-min sketch: A case study of

  42. [42]

    You, Ronghui and Zhang, Zihan and Wang, Ziye and Dai, Suyang and Mamitsuka, Hiroshi and Zhu, Shanfeng , booktitle=

  43. [43]

    International Conference on Machine Learning (ICML) , year=

    Bayes optimal multilabel classification via probabilistic classifier chains , author=. International Conference on Machine Learning (ICML) , year=

  44. [44]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Efficient loss-based decoding on graphs for extreme classification , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  45. [45]

    1611.01964 , archivePrefix=

    Log-time and log-space extreme classification , author=. 1611.01964 , archivePrefix=

    Pith/arXiv arXiv

  46. [46]

    2601.10639 , archivePrefix=

    STEM: Scaling Transformers with Embedding Modules , author=. 2601.10639 , archivePrefix=

    arXiv

  47. [47]

    He, Ruining and Heldt, Lukasz and Hong, Lichan and Keshavan, Raghunandan and Mao, Shifan and Mehta, Nikhil and Su, Zhengyang and Tsai, Alicia and Wang, Yueqi and Wang, Shao-Chuan and others , booktitle=

  48. [48]

    , author=

    Adaptive subgradient methods for online learning and stochastic optimization. , author=. Journal of Machine Learning Research (JMLR) , volume=

  49. [49]

    2309.06497 , archivePrefix=

    A distributed data-parallel pytorch implementation of the distributed shampoo optimizer for training neural networks at-scale , author=. 2309.06497 , archivePrefix=

    Pith/arXiv arXiv

  50. [50]

    International Conference on Machine Learning (ICML) , year=

    Shampoo: Preconditioned stochastic tensor optimization , author=. International Conference on Machine Learning (ICML) , year=

  51. [51]

    Annual Meeting of the Association for Computational Linguistics and International Joint Conference on Natural Language Processing (ACL-IJCNLP) , year=

    On Using Very Large Target Vocabulary for Neural Machine Translation , author=. Annual Meeting of the Association for Computational Linguistics and International Joint Conference on Natural Language Processing (ACL-IJCNLP) , year=

  52. [52]

    Entropy , VOLUME =

    Ramos, Daniel and Franco-Pedroso, Javier and Lozano-Diez, Alicia and Gonzalez-Rodriguez, Joaquin , TITLE =. Entropy , VOLUME =. 2018 , NUMBER =

    2018

  53. [53]

    ACM Conference on Recommender Systems (RecSys) , year=

    Sampling-bias-corrected neural modeling for large corpus item recommendations , author=. ACM Conference on Recommender Systems (RecSys) , year=

  54. [54]

    ACM Transactions on Information Systems , volume=

    On the effectiveness of sampled softmax loss for item recommendation , author=. ACM Transactions on Information Systems , volume=. 2024 , publisher=

    2024

  55. [55]

    ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

    Real-time Indexing for Large-scale Recommendation by Streaming Vector Quantization Retriever , author=. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

  56. [56]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Unsupervised Learning of Visual Features by Contrasting Cluster Assignments , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  57. [57]

    Deep Neural Networks for

    Covington, Paul and Adams, Jay and Sargin, Emre , booktitle=. Deep Neural Networks for

  58. [58]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Sinkhorn Distances: Lightspeed Computation of Optimal Transport , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  59. [59]

    Du, Nan and Huang, Yanping and Dai, Andrew M and Tong, Simon and Lepikhin, Dmitry and Xu, Yuanzhong and Krikun, Maxim and Zhou, Yanqi and Yu, Adams Wei and Firat, Orhan and others , booktitle=

  60. [60]

    Fedus, William and Zoph, Barret and Shazeer, Noam , journal=

  61. [61]

    International Conference on Machine Learning (ICML) , year=

    Accelerating Large-Scale Inference with Anisotropic Vector Quantization , author=. International Conference on Machine Learning (ICML) , year=

  62. [62]

    2401.04088 , archivePrefix=

    Mixtral of Experts , author=. 2401.04088 , archivePrefix=

    Pith/arXiv arXiv

  63. [63]

    Spherical

    Hornik, Kurt and Feinerer, Ingo and Kober, Martin and Buchta, Christian , journal=. Spherical

  64. [64]

    Billion-Scale Similarity Search with

    Johnson, Jeff and Douze, Matthijs and J. Billion-Scale Similarity Search with. IEEE Transactions on Big Data , volume=. 2021 , publisher=

    2021

  65. [65]

    IEEE International Conference on Data Mining (ICDM) , year=

    Self-Attentive Sequential Recommendation , author=. IEEE International Conference on Data Mining (ICDM) , year=

  66. [66]

    Lepikhin, Dmitry and Lee, HyoukJoong and Xu, Yuanzhong and Chen, Dehao and Firat, Orhan and Huang, Yanping and Krikun, Maxim and Shazeer, Noam and Chen, Zhifeng , booktitle=

  67. [67]

    ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) , year=

    Mamba4Rec: Towards Efficient Sequential Recommendation with Selective State Space Models , author=. ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) , year=

  68. [68]

    2405.03110 , archivePrefix=

    Vector Quantization for Recommender Systems: A Review and Outlook , author=. 2405.03110 , archivePrefix=

    Pith/arXiv arXiv

  69. [69]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Neural Discrete Representation Learning , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  70. [70]

    Matrix Factorization Revisited , author=

    Neural Collaborative Filtering vs. Matrix Factorization Revisited , author=. ACM Conference on Recommender Systems (RecSys) , year=

  71. [71]

    International Conference on Learning Representations (ICLR) , year=

    Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer , author=. International Conference on Learning Representations (ICLR) , year=

  72. [72]

    ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

    Scalable and Sustainable Deep Learning via Randomized Hashing , author=. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

  73. [73]

    IEEE/ACM Transactions on Audio, Speech, and Language Processing , volume=

    SoundStream: An End-to-End Neural Audio Codec , author=. IEEE/ACM Transactions on Audio, Speech, and Language Processing , volume=. 2021 , publisher=

    2021

  74. [74]

    2504.05314 , archivePrefix=

    Multimodal Quantitative Language for Generative Recommendation , author=. 2504.05314 , archivePrefix=

    Pith/arXiv arXiv

  75. [75]

    2410.09560 , archivePrefix=

    Towards Scalable Semantic Representation for Recommendation , author=. 2410.09560 , archivePrefix=

    Pith/arXiv arXiv

  76. [76]

    ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

    Learning Tree-based Deep Model for Recommender Systems , author=. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

  77. [77]

    ACM Web Conference (WWW) , year=

    Retrieval with Learned Similarities , author=. ACM Web Conference (WWW) , year=

  78. [78]

    ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

    Revisiting Neural Retrieval on Accelerators , author=. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , year=

  79. [79]

    2412.19437 , archivePrefix=

    Pith/arXiv arXiv

  80. [80]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Recommender Systems with Generative Retrieval , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

Showing first 80 references.