pith. sign in

arxiv: 2605.25584 · v1 · pith:LGQ4CQDInew · submitted 2026-05-25 · 💻 cs.RO · cs.AI

Acting on the Unseen: Communication-Free Collaborative Filtering for Decentralized Multi-Robot Task Allocation

Alexander Apartsin , Yigal Meshulam , Yehudit Aperstein This is my paper

Pith reviewed 2026-06-29 21:50 UTC · model grok-4.3

classification 💻 cs.RO cs.AI
keywords multi-robot task allocationcollaborative filteringdecentralized systemslow-rank matrix completionzero-knowledge MRTAonline learningSwarmCF
0
0 comments X

The pith

Each robot can act well on unseen tasks by running online low-rank collaborative filtering over broadcasted teammate outcomes without communication or prior models.

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

The paper examines multi-robot task allocation in a zero-knowledge regime with no prior knowledge of task models, no communication or parameter sharing, and only partial noisy private views of a public stream of teammates' outcomes. It shows that a hidden low-rank structure in robot-task suitability lets each robot run online low-rank collaborative filtering independently over the broadcast to generalize to most unattempted pairs. This yields a categorical advantage because any structure-free learner stays at the prior-mean error floor on unseen pairs. The approach proves per-robot sample complexity linear in rank d rather than task count n, an anytime separation under scarcity, and a deterministic condition for exact decentralized recovery from the masked broadcast.

Core claim

In Zero-Knowledge MRTA a robot team with no prior knowledge, no communication, and only a partial privately-noisy view of a public stream of teammates' outcomes can still allocate tasks effectively because a hidden low-rank structure governs robot-task suitability. Each robot runs online low-rank collaborative filtering over the broadcast (SwarmCF) to act well on tasks it never attempted and to onboard new tasks. This delivers matching per-robot sample complexity Ω(d) versus Ω(n), an anytime cumulative-reward separation under task scarcity, and a deterministic condition under which decentralized recovery from the masked broadcast is exact.

What carries the argument

SwarmCF: online low-rank collaborative filtering run independently by each robot on the broadcasted masked outcomes to generalize to unseen (robot, task) pairs.

If this is right

  • Per-robot unseen-pair skill rises with team size.
  • The method recovers about 80 percent of centralized full-communication earned skill.
  • Performance holds under capacity-1 contention and in robotics-grounded sensing instances.
  • Anytime separation under task scarcity is stronger than other low-rank methods.

Where Pith is reading between the lines

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

  • Broadcasting outcomes could enable similar generalization in other decentralized multi-agent systems whose interactions admit low-rank structure.
  • The scaling law implies that adding agents improves individual performance on novel tasks even when agents never interact directly.
  • The anytime property could support extensions to settings with streaming tasks or changing team membership.
  • The exact-recovery condition may apply to other distributed matrix-completion problems where each agent sees only a masked subset.

Load-bearing premise

A hidden low-rank structure governs which robot suits which task.

What would settle it

If the suitability matrix is randomized to destroy low-rank structure, performance on unseen pairs should fall to the prior-mean error floor with no categorical advantage remaining.

Figures

Figures reproduced from arXiv: 2605.25584 by Alexander Apartsin, Yehudit Aperstein, Yigal Meshulam.

Figure 1
Figure 1. Figure 1: The setting. The robot × task reward is hidden and low-rank, 𝑅 = 𝑃𝑈 ⊤ (capability traits 𝑝𝑖 , requirement traits 𝑢𝑗 ). A focal robot (blue row) must act on its whole row, including the many pairs it never engaged ('?'), using only its own clean engagements (blue circles), a partial and per-observer-noisy view of the teammates it can sense (black dots; greyed rows are persistently invisible to it), and no c… view at source ↗
Figure 2
Figure 2. Figure 2: Unseen-pair skill versus broadcast rate 𝜌. Structure-free learners are pinned at the floor at every broadcast rate; SwarmCF acts on the unseen throughout and is robust under masking, while the batch variant SwarmCF-batch decays as observation becomes partial. Means with bootstrap 95% CIs over 16 seeds. 6.2 The operational (anytime) separation Final-policy quality can overstate a method that explores cheapl… view at source ↗
Figure 3
Figure 3. Figure 3: Anytime cumulative-reward skill (𝜌 = 0.25). SwarmCF earns from round one; explore-then-commit pays a probe phase; Independent-UCB stays near the random floor, while 𝜀-greedy tabular earns only by re-exploiting tasks it has already engaged. Means with bootstrap 95% CIs over 16 seeds. 6.3 Why a swarm: the value of the broadcast and a positive scaling law Two experiments isolate what the team and the broadcas… view at source ↗
Figure 4
Figure 4. Figure 4: (a) Value of the broadcast: unseen skill versus 𝜌 (left edge = isolated). (b) Positive scaling: unseen skill versus team size 𝑚 at fixed broadcast rate 𝜌 = 0.5 (and fixed 𝑛, horizon 𝑇). In both panels our online SwarmCF and its batch variant SwarmCF-batch rise (the gain is structural), while structure-free learners stay flat. Means with bootstrap 95% CIs over 16 seeds. 6.4 The cost of communication-free op… view at source ↗
Figure 5
Figure 5. Figure 5: The categorical separation under capacity-1 contention (LatentSwarm: signed low-rank reward, persistent partial and private broadcast, task scarcity). Left: earned (anytime) skill, SwarmCF best among learners and far above the structure-free floor (independent-UCB goes negative under contention). Right: unseen-pair skill, SwarmCF the only method significantly above the floor. Bars are means with bootstrap … view at source ↗
Figure 6
Figure 6. Figure 6: The cost of no de-confliction under capacity-1 contention (LatentSwarm). (a) Earned skill versus collision rate on the all-tasks menu: structure-free independent-UCB collides on almost every engagement and earns below random, while SwarmCF spreads robots across tasks via its heterogeneous learned models and earns most. (b) Collision rate per method under the all-tasks menu versus a size-𝑐 menu: restricting… view at source ↗
Figure 7
Figure 7. Figure 7: Robustness to the guessed rank 𝑑̂ (true rank 𝑑 = 5). Under-ranking (𝑑̂ < 𝑑) is mis-specified and degrades smoothly; over-ranking (𝑑̂ > 𝑑, up to 3𝑑) is regularized away and stays flat. Lines are means with bootstrap 95% CIs over 16 seeds. 6.7 A robotics-grounded instance The experiments above use abstract latent traits and a Bernoulli visibility mask. To check the separation is not an artifact of that abstr… view at source ↗
Figure 8
Figure 8. Figure 8: A robotics-grounded instance (LatentSwarm): non-negative sensing-modality traits and a geometry-induced range-limited line-of-sight mask with distance-dependent noise, under capacity-1 contention. (a) An example patrol layout and its line-of-sight visibility graph; (b) earned (anytime) skill; (c) unseen-pair skill. SwarmCF stays well above the structure-free floor, so the categorical separation survives a … view at source ↗
Figure 9
Figure 9. Figure 9: Offer-size robustness. (a) Unseen-pair skill versus broadcast rate 𝜌 under the body's size-𝑐 menu (𝑐 = 20) and the unrestricted all-tasks menu (𝑐 = 𝑛), for SwarmCF and the batch completer (the structure-free floor is shown once): every low-rank method stays above the floor at both menu sizes (the categorical separation is offer-size invariant), but SwarmCF's lead over the batch method narrows under the all… view at source ↗
Figure 10
Figure 10. Figure 10: Persistent versus i.i.d. masking. Unseen-pair and anytime skill are essentially invariant to the masking model (left, center); decentralization (state-uniqueness of the robots' learned models) is durable under the persistent mask but transient under the i.i.d. mask, which averages out blind spots over the mission (right). Theorem 1's strict regime. The body's anytime comparison ( [PITH_FULL_IMAGE:figures… view at source ↗
Figure 11
Figure 11. Figure 11: Two Appendix-F ablations. (a) Strict scarce-offer regime (𝑐 = 3, 𝑐𝑇 = 𝑜(𝑛), 𝜌 = 0.25): structure-free learners' cumulative skill collapses to the floor (Theorem 1) while SwarmCF still earns. (b) Under the all-tasks menu (𝑐 = 𝑛), a short UCB probe restores the online estimator's unseen-pair lead over batch completion that pure greedy selection gives up (structure-free floor shown for reference). Means over… view at source ↗
Figure 12
Figure 12. Figure 12: Approximate-low-rank robustness: unseen-pair skill versus a full-rank perturbation of strength 𝜀 (top axis: the resulting effective rank at 99% energy), at the masked headline 𝜌 = 0.25. SwarmCF and its batch variant degrade gracefully as the reward leaves the rank-𝑑 subspace and approach the structure-free floor only when little low-rank structure remains; the structure-free learner is at the floor for ev… view at source ↗
read the original abstract

Multi-robot task allocation usually assumes some combination of communication, known task models, or a coordinator. We study the opposite extreme, a regime common in practice but overlooked in theory, which we name Zero-Knowledge MRTA (ZK-MRTA): a robot team with no prior knowledge (no task models, not even the latent rank), no communication (no messages, no parameter sharing, no coordinator), and only a partial and privately-noisy view of a public stream of teammates' outcomes. A hidden low-rank structure governs which robot suits which task, and there are far more tasks than rounds, so most (robot, task) pairs are never attempted. Yet each robot can act well on tasks it never attempted, and onboard new tasks, by running online low-rank collaborative filtering over the broadcast (SwarmCF). The advantage over any structure-free learner is categorical, not a constant factor: a structure-free learner is provably at the prior-mean error floor on unseen pairs. We prove a matching per-robot sample complexity ({\Theta}(d) versus {\Theta}(n), in the rank d and the task count n), an anytime (cumulative-reward) separation under task scarcity, and a deterministic condition under which decentralized recovery from the masked broadcast is exact (validated empirically). Experiments quantify the value of the broadcast, a positive scaling law (per-robot unseen-pair skill rises with team size), and the strongest masking-robustness and anytime profile among low-rank methods, recovering most (about 80% on earned skill) of a centralized full-communication ceiling, and holding under capacity-1 contention and in a robotics-grounded sensing instance.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The manuscript introduces Zero-Knowledge Multi-Robot Task Allocation (ZK-MRTA), a regime with no prior task models, no communication or parameter sharing, and only partial privately-noisy observations of a public stream of teammate outcomes. It proposes SwarmCF, an online low-rank collaborative filtering algorithm that exploits an assumed hidden low-rank structure in the robot-task suitability matrix to enable robots to act on unseen tasks. The central claims are a matching per-robot sample complexity of Θ(d) versus Θ(n) for structure-free learners, an anytime cumulative-reward separation under task scarcity, and a deterministic condition for exact decentralized recovery from the masked broadcast, with empirical validation showing positive scaling with team size and recovery of ~80% of a centralized ceiling.

Significance. If the low-rank assumption holds and the stated bounds and recovery condition are correctly derived, the work supplies a communication-free mechanism for decentralized MRTA that achieves a categorical (not merely constant-factor) advantage over structure-free methods on unseen pairs, together with matching sample-complexity guarantees and an anytime profile. These elements, if substantiated, would be a substantive contribution to scalable multi-robot systems operating under severe communication and knowledge constraints.

major comments (1)
  1. Abstract: the claim of 'matching per-robot sample complexity (Θ(d) versus Θ(n))' and the 'deterministic condition under which decentralized recovery ... is exact' are load-bearing for the categorical separation result; the manuscript must state the precise theorem statements, including all assumptions on the low-rank matrix and the masking process, in a dedicated theorem environment so that the matching and exactness can be verified directly.
minor comments (2)
  1. Abstract, paragraph 2: the phrase 'a hidden low-rank structure governs which robot suits which task' should be accompanied by the explicit matrix factorization model (e.g., suitability ≈ U V^T with rank d) at first use.
  2. The empirical claim of 'strongest masking-robustness ... among low-rank methods' requires an explicit list of the compared baselines and the precise metric (e.g., earned skill under 80% masking) in the abstract or a table caption.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comment on improving the verifiability of our central claims. We address it point by point below.

read point-by-point responses

  1. Referee: [—] Abstract: the claim of 'matching per-robot sample complexity (Θ(d) versus Θ(n))' and the 'deterministic condition under which decentralized recovery ... is exact' are load-bearing for the categorical separation result; the manuscript must state the precise theorem statements, including all assumptions on the low-rank matrix and the masking process, in a dedicated theorem environment so that the matching and exactness can be verified directly.

    Authors: We agree that the load-bearing claims require explicit formal statements for direct verification. The current manuscript states the results in prose in the abstract and introduction but does not isolate them as numbered theorems with complete assumption lists. In the revised version we will add two dedicated theorem environments (likely in the analysis section following the algorithm description): Theorem 1 will state the per-robot sample complexity bound of Θ(d) for unseen pairs under the low-rank model, and Theorem 2 will state the deterministic exact-recovery condition from the masked broadcast. Both will enumerate all assumptions on the low-rank matrix (rank d, incoherence parameter, minimum singular-value gap) and on the masking process (broadcast observation model, private noise distribution). This change directly addresses the referee's request without altering any proofs or results. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper's derivation begins from the explicit modeling assumption of an approximately low-rank robot-task suitability matrix and derives all performance guarantees (categorical separation from structure-free learners, Θ(d) per-robot sample complexity, anytime cumulative-reward bounds, and deterministic exact decentralized recovery) as mathematical consequences of that assumption together with the masked-broadcast observation model. No claimed prediction is obtained by fitting a parameter to the same data and renaming the fit, no load-bearing uniqueness theorem is imported via self-citation, and no ansatz is smuggled through prior work. The low-rank structure is presented as an external premise rather than an internally derived or self-referential quantity, rendering the central claims self-contained under the stated hypothesis.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the existence of a low-rank structure in the unknown robot-task matrix and on the availability of a public broadcast channel that delivers noisy outcome observations without direct robot-to-robot messaging.

axioms (1)
  • domain assumption There exists a hidden low-rank structure governing robot-task suitability
    Invoked in abstract paragraph 2 as the property that enables generalization to unseen pairs via collaborative filtering.
invented entities (1)
  • SwarmCF no independent evidence
    purpose: Decentralized online low-rank collaborative filtering performed independently by each robot on masked broadcast data
    New algorithm introduced to solve the ZK-MRTA problem; no independent evidence outside the paper is provided.

pith-pipeline@v0.9.1-grok · 5840 in / 1482 out tokens · 38787 ms · 2026-06-29T21:50:48.292832+00:00 · methodology

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

54 extracted references · 5 canonical work pages · 3 internal anchors

  1. [1]

    The impact of diversity on optimal control policies for heterogeneous robot swarms

    Prorok A, Hsieh MA, Kumar V. The impact of diversity on optimal control policies for heterogeneous robot swarms. IEEE Transactions on Robotics. 2017 Jan 16;33(2):346 -58

    2017

  2. [2]

    Learning task requirements and agent capabilities for multi -agent task allocation

    Fu B, Smith W, Rizzo D, Castanier M, Ghaffari M, Barton K. Learning task requirements and agent capabilities for multi -agent task allocation. arXiv preprint arXiv:2211.03286. 2022 Nov 7

    work page arXiv 2022

  3. [3]

    Personality traits and the marriage market

    Dupuy A, Galichon A. Personality traits and the marriage market. Journal of Political Economy. 2014 Dec 1;122(6):1271 -319. 20

    2014

  4. [4]

    Schilling, F., Schiano, F., & Floreano, D. (2021). Vision -based drone flocking in outdoor environments. IEEE Robotics and Automation Letters , 6(2), 2954 -2961

    2021

  5. [5]

    & Gao, F

    Zhou, X., Wen, X., Wang, Z., Gao, Y., Li, H., Wang, Q., ... & Gao, F. (2022). Swarm of micro flying robots in the wild. Science robotics , 7(66), eabm5954

    2022

  6. [6]

    & Arkin, R

    Hudson, N., Talbot, F., Cox, M., Williams, J., Hines, T., Pitt, A., ... & Arkin, R. C. (2022). Heterogeneous ground and air platforms, homogeneous sensing: Team CSIRO Data61's approach to the DARPA subterranean challenge. Field Robotics , 2, 595-636

    2022

  7. [7]

    & Alexis, K

    Tranzatto, M., Dharmadhikari, M., Bernreiter, L., Camurri, M., Khattak, S., Mascarich, F., ... & Alexis, K. (2024). Team cerberus wins the darpa subterranean challenge: Technical overview and lessons learned. Field Robotics , 4, 349-312

    2024

  8. [8]

    Brambilla, M., Ferrante, E., Birattari, M., & Dorigo, M. (1935). Swarm robotics: a review from the swarm engineering perspective. Swarm Intell. 7 (1), 1 –41 (2013)

    1935

  9. [9]

    (2004, July)

    Şahin, E. (2004, July). Swarm robotics: From sources of inspiration to domains of application. In International workshop on swarm robotics (pp. 10-20). Berlin, Heidelberg: Springer Berlin Heidelberg

    2004

  10. [10]

    Cortes, J., Martinez, S., Karatas, T., & Bullo, F. (2004). Coverage control for mobile sensing networks. IEEE Transactions on robotics and Automation , 20(2), 243 -255

    2004

  11. [11]

    P., & Matarić, M

    Gerkey, B. P., & Matarić, M. J. (2004). A formal analysis and taxonomy of task allocation in multi -robot systems. The International journal of robotics research , 23(9), 939 -954

    2004

  12. [12]

    A., Stentz, A., & Dias, M

    Korsah, G. A., Stentz, A., & Dias, M. B. (2013). A comprehensive taxonomy for multi -robot task allocation. The International journal of robotics research , 32(12), 1495 -1512

    2013

  13. [13]

    Nunes, E., Manner, M., Mitiche, H., & Gini, M. (2017). A taxonomy for task allocation problems with temporal and ordering constraints. Robotics and Autonomous Systems , 90, 55-70

    2017

  14. [14]

    B., Zlot, R., Kalra, N., & Stentz, A

    Dias, M. B., Zlot, R., Kalra, N., & Stentz, A. (2006). Market -based multirobot coordination: A survey and analysis. Proceedings of the IEEE , 94(7), 1257 -1270

    2006

  15. [15]

    L., Brunet, L., & How, J

    Choi, H. L., Brunet, L., & How, J. P. (2009). Consensus -based decentralized auctions for robust task allocation. IEEE transactions on robotics , 25(4), 912 -926

    2009

  16. [16]

    Khamis, A., Hussein, A., & Elmogy, A. (2015). Multi -robot task allocation: A review of the state -of-the- art. Cooperative robots and sensor networks 2015 , 31-51

    2015

  17. [17]

    J., & Sofge, D

    Otte, M., Kuhlman, M. J., & Sofge, D. (2020). Auctions for multi -robot task allocation in communication limited environments. Autonomous robots , 44(3), 547 -584

    2020

  18. [18]

    Gielis, J., Shankar, A., & Prorok, A. (2022). A critical review of communications in multi -robot systems. Current robotics reports , 3(4), 213 -225

    2022

  19. [19]

    (2019, June)

    Notomista, G., Mayya, S., Hutchinson, S., & Egerstedt, M. (2019, June). An optimal task allocation strategy for heterogeneous multi-robot systems. In 2019 18th European control conference (ECC) (pp. 2071-2076). IEEE

    2019

  20. [20]

    Ravichandar, H., Shaw, K., & Chernova, S. (2020). STRATA: unified framework for task assignments in large teams of heterogeneous agents. Autonomous Agents and Multi -Agent Systems , 34(2), 38

    2020

  21. [21]

    KA, A., & Subramaniam, U. (2024). A systematic literature review on multi -robot task allocation. ACM Computing Surveys , 57(3), 1-28

    2024

  22. [22]

    (2016, June)

    Rosenski, J., Shamir, O., & Szlak, L. (2016, June). Multi -player bandits –a musical chairs approach. In International Conference on Machine Learning (pp. 155 -163). PMLR

    2016

  23. [23]

    Boursier, E., & Perchet, V. (2019). SIC-MMAB: Synchronisation involves communication in multiplayer multi-armed bandits. Advances in Neural Information Processing Systems , 32

    2019

  24. [24]

    Bistritz, I., & Leshem, A. (2018). Distributed multi-player bandits-a game of thrones approach. Advances in Neural Information Processing Systems , 31

    2018

  25. [25]

    Yu, C., Velu, A., Vinitsky, E., Gao, J., Wang, Y., Bayen, A., & Wu, Y. (2022). The surprising effectiveness of ppo in cooperative multi -agent games. Advances in neural information processing systems , 35, 24611- 24624

    2022

  26. [26]

    S., Farquhar, G., Foerster, J., & Whiteson, S

    Rashid, T., Samvelyan, M., De Witt, C. S., Farquhar, G., Foerster, J., & Whiteson, S. (2020). Monotonic value function factorisation for deep multi -agent reinforcement learning. Journal of Machine Learning Research , 21(178), 1 -51. 21

    2020

  27. [27]

    Value-Decomposition Networks For Cooperative Multi-Agent Learning

    Sunehag, P., Lever, G., Gruslys, A., Czarnecki, W. M., Zambaldi, V., Jaderberg, M., ... & Graepel, T. (2017). Value-decomposition networks for cooperative multi -agent learning. arXiv preprint arXiv:1706.05296

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  28. [28]

    I., Tamar, A., Harb, J., Pieter Abbeel, O., & Mordatch, I

    Lowe, R., Wu, Y. I., Tamar, A., Harb, J., Pieter Abbeel, O., & Mordatch, I. (2017). Multi-agent actor-critic for mixed cooperative-competitive environments. Advances in neural information processing systems , 30

    2017

  29. [29]

    Sukhbaatar, S., & Fergus, R. (2016). Learning multiagent communication with backpropagation. Advances in neural information processing systems , 29

    2016

  30. [30]

    (2019, May)

    Das, A., Gervet, T., Romoff, J., Batra, D., Parikh, D., Rabbat, M., & Pineau, J. (2019, May). Tarmac: Targeted multi-agent communication. In International Conference on machine learning (pp. 1538 -1546). PMLR

    2019

  31. [31]

    A., De Freitas, N., & Whiteson, S

    Foerster, J., Assael, I. A., De Freitas, N., & Whiteson, S. (2016). Learning to communicate with deep multi- agent reinforcement learning. Advances in neural information processing systems , 29

    2016

  32. [32]

    Goarin, M., & Loianno, G. (2024). Graph neural network for decentralized multi -robot goal assignment. IEEE Robotics and Automation Letters , 9(5), 4051 -4058

    2024

  33. [33]

    Wang, Z., & Gombolay, M. (2020). Learning scheduling policies for multi -robot coordination with graph attention networks. IEEE Robotics and Automation Letters , 5(3), 4509 -4516

    2020

  34. [34]

    Federated Collaborative Filtering for Privacy-Preserving Personalized Recommendation System

    Ammad-Ud-Din, M., Ivannikova, E., Khan, S. A., Oyomno, W., Fu, Q., Tan, K. E., & Flanagan, A. (2019). Federated collaborative filtering for privacy-preserving personalized recommendation system. arXiv preprint arXiv:1901.09888

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  35. [35]

    McMahan, B., Moore, E., Ramage, D., Hampson, S., & y Arcas, B. A. (2017, April). Communication - efficient learning of deep networks from decentralized data. In Artificial intelligence and statistics (pp. 1273- 1282). Pmlr

    2017

  36. [36]

    S., Givan, R., Immerman, N., & Zilberstein, S

    Bernstein, D. S., Givan, R., Immerman, N., & Zilberstein, S. (2002). The complexity of decentralized control of Markov decision processes. Mathematics of operations research , 27(4), 819 -840

    2002

  37. [37]

    Candes, E., & Recht, B. (2012). Exact matrix completion via convex optimization. Communications of the ACM, 55(6), 111 -119

    2012

  38. [38]

    H., Montanari, A., & Oh, S

    Keshavan, R. H., Montanari, A., & Oh, S. (2010). Matrix completion from a few entries. IEEE transactions on information theory , 56(6), 2980 -2998

    2010

  39. [39]

    Recht, B. (2011). A simpler approach to matrix completion. Journal of Machine Learning Research, 12(12)

    2011

  40. [40]

    (2013, June)

    Jain, P., Netrapalli, P., & Sanghavi, S. (2013, June). Low-rank matrix completion using alternating minimization. In Proceedings of the forty -fifth annual ACM symposium on Theory of computing (pp. 665- 674)

    2013

  41. [41]

    Koren, Y., Bell, R., & Volinsky, C. (2009). Matrix factorization techniques for recommender systems. Computer, 42(8), 30-37

    2009

  42. [42]

    Salakhutdinov, R., & Mnih, A. (2008). Bayesian probablistic matrix factorization using MCMC. In Proceedings of the International Conference on Machine Learning (ICML), Helsinki, Finland (Vol. 10, No. 1390156.1390267)

    work page arXiv 2008

  43. [43]

    Mazumder, R., Hastie, T., & Tibshirani, R. (2010). Spectral regularization algorithms for learning large incomplete matrices. The Journal of Machine Learning Research , 11, 2287-2322

    2010

  44. [44]

    (2008, December)

    Hu, Y., Koren, Y., & Volinsky, C. (2008, December). Collaborative filtering for implicit feedback datasets. In 2008 Eighth IEEE international conference on data mining (pp. 263 -272). Ieee

    2008

  45. [45]

    Rendle, S., Freudenthaler, C., Gantner, Z., & Schmidt -Thieme, L. (2012). BPR: Bayesian personalized ranking from implicit feedback. arXiv preprint arXiv:1205.2618

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  46. [46]

    J., & Lee, T

    Kang, Y., Hsieh, C. J., & Lee, T. C. M. (2022). Efficient frameworks for generalized low-rank matrix bandit problems. Advances in Neural Information Processing Systems , 35, 19971-19983

    2022

  47. [47]

    S., Willett, R., Wright, S., & Nowak, R

    Jun, K. S., Willett, R., Wright, S., & Nowak, R. (2019, May). Bilinear bandits with low -rank structure. In International Conference on Machine Learning (pp. 3163 -3172). PMLR

    2019

  48. [48]

    (2021, March)

    Lu, Y., Meisami, A., & Tewari, A. (2021, March). Low-rank generalized linear bandit problems. In International conference on artificial intelligence and statistics (pp. 460 -468). PMLR

    2021

  49. [49]

    (2014, June)

    Gentile, C., Li, S., & Zappella, G. (2014, June). Online clustering of bandits. In International conference on machine learning (pp. 757 -765). PMLR

    2014

  50. [50]

    (2012, March)

    Ling, Q., Xu, Y., Yin, W., & Wen, Z. (2012, March). Decentralized low -rank matrix completion. In 2012 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (pp. 2925 -2928). IEEE. 22

    2012

  51. [51]

    Murphy, A. H. (1988). Skill scores based on the mean square error and their relationships to the correlation coefficient. Monthly weather review , 116(12), 2417 -2424

    1988

  52. [52]

    (2002, December)

    Sarwar, B., Karypis, G., Konstan, J., & Riedl, J. (2002, December). Incremental singular value decomposition algorithms for highly scalable recommender systems. In Fifth international conference on computer and information science (Vol. 1, No. 012002, pp. 27 -8)

    2002

  53. [53]

    Auer, P., Cesa -Bianchi, N., & Fischer, P. (2002). Finite-time analysis of the multiarmed bandit problem. Machine learning , 47(2), 235 -256

    2002

  54. [54]

    Kuhn, H. W. (1955). The Hungarian method for the assignment problem. Naval research logistics quarterly , 2(1‐2), 83 -97. Appendix A. Proofs of the main results We give proofs of the main results below (full for Proposition 1, Lemma 1, Theorem 1, and Theorem 2; Theorem 3 follows as a corollary of Theorem 2), each closed by a short remark on the proof tech...

    1955