pith. sign in

arxiv: 2605.17568 · v2 · pith:SYP4OISAnew · submitted 2026-05-17 · 💻 cs.LG

Structured Neural Marked Point Processes for Interpretable Event Interaction Modeling

Zhitong Xu , Qiwei Yuan , Yinghao Chen , Shandian Zhe , Bin Shen This is my paper

Pith reviewed 2026-05-21 07:53 UTC · model grok-4.3

classification 💻 cs.LG
keywords marked point processesneural point processesinterpretable modelsevent interaction modelinginfluence kernelstemporal decaystochastic processes
0
0 comments X

The pith

Structured neural marked point process factors influences into signed class networks and monotonic temporal components.

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

This paper introduces a structured neural marked point process that represents event influences as the product of a signed interaction network over event types and a delay-aware monotonic temporal network. A sympathetic reader would care because standard neural point process models deliver accurate predictions but treat all interactions as opaque black boxes, whereas this factorization makes it possible to read out explicit topologies of excitation, inhibition, and neutrality between event classes along with flexible timing patterns. The approach maintains modeling power for complex streams while supporting direct discovery of structured dependencies, and it trains efficiently with a stratified Monte Carlo estimator. Experiments on synthetic and real benchmarks show the model recovers meaningful relationships without sacrificing predictive performance.

Core claim

Our model constructs a product-form neural influence kernel composed of a signed interaction network over event types and a delay-aware monotonic temporal network. This design enables explicit characterization of inter-class influence topology -- including excitation, inhibition, and neutrality -- while flexibly capturing diverse temporal decay patterns and potential influence delays.

What carries the argument

The product-form neural influence kernel, which multiplies a signed interaction network over event types with a delay-aware monotonic temporal network to separate class-wise relational structure from temporal dynamics.

Load-bearing premise

The true inter-event influences can be faithfully represented by a product of a class-wise signed network and a monotonic temporal network without substantial loss of modeling power or introduction of spurious structure.

What would settle it

On synthetic data with known ground-truth signed influences, the learned network fails to recover the correct excitation or inhibition signs, or predictive log-likelihood falls substantially below that of unstructured neural point process baselines on real event streams.

Figures

Figures reproduced from arXiv: 2605.17568 by Bin Shen, Qiwei Yuan, Shandian Zhe, Yinghao Chen, Zhitong Xu.

Figure 1
Figure 1. Figure 1: The conditional intensity function on a validation sequence from [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 1
Figure 1. Figure 1: Graphical Illustration of SNMPP. 3 Methodology 3.1 Model To enable explicit event-wise and type-wise interaction discovery while retaining flexibility to capture complex temporal dependencies, we model the marked conditional intensity as λk(t | Ht) = σ  αk + X tn<t fkn→k(t − tn)  , (1) where αk is a latent baseline parameter capturing the spontaneous tendency of event type k, and fkn→k(∆t) denotes the in… view at source ↗
Figure 2
Figure 2. Figure 2: The conditional intensity function on a validation sequence from [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Learned influence kernel on PP1. Note that SNMPP learns a single unified influence kernel shared across all event types, as defined in (2) [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 3
Figure 3. Figure 3: The conditional intensity function on a validation sequence from [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Learned influence kernel from PP2. For all other type pairs, including E1 → E1, E2 → E2, and E2 → E1, there is no true interaction; the corresponding ground-truth kernels are identically zero. The learned kernels faithfully recover this null structure, with estimated delay parameters close to zero. In addition, SNMPP accurately estimates the baseline intensities for both processes. Detailed quantitative co… view at source ↗
Figure 5
Figure 5. Figure 5: Influence kernels learned by SNMPP on event data generated by a simulated supply-chain system. the inventory first reaches zero, a stockout event Eout is recorded, and thereafter customer orders are physically suppressed (i.e., not recorded) until inventory is replenished. When the inventory falls below the reorder threshold r = 5 and no restock is pending, a replenishment order E2 is triggered, which sche… view at source ↗
Figure 5
Figure 5. Figure 5: Learned influence kernel from PP2. Structure Discovery. We then examined whether SNMPP can recover the interaction structure among event types and the temporal evolution of influence strengths. To this end, we visualize the learned influence functions for every ordered pair of event types. As shown in Figures 4 and 5, SNMPP not only correctly identifies the type of interaction — excitation, inhibition, or … view at source ↗
Figure 6
Figure 6. Figure 6: Graphical Illustration of SNMPP. Comparing this with the variance of the stratified estimator yields Varh Ibmci = Varh Ibstrati + L 2 n Q2 X Q q=1 (µq − µ¯) 2 . The second term is nonnegative, and hence Varh Ibstrati ≤ Varh Ibmci . The inequality is strict whenever the stratum-wise means µ1, . . . , µQ are not all equal. Lemma B.1 shows that standard Monte Carlo contains both within-stratum variability and… view at source ↗
Figure 6
Figure 6. Figure 6: Influence kernels learned by SNMPP on event data generated by a simulated supply-chain system. Third, we evaluated SNMPP on a simulated supply-chain system designed to reflect realistic oper￾ational logic. The event sequences are generated from stochastic dynamics governed by physical constraints and inventory decision rules, rather than being derived from any temporal point process model. This setting ena… view at source ↗
Figure 7
Figure 7. Figure 7: Conditional intensity function of E1 (customer order) inferred by SNMPP on a sequence from the simulated supply-chain system. E.1 Event types The system produces four event types: (i) customer order E1, (ii) replenishment order E2, (iii) stock arrival E3, and (iv) recorded stockout Eout, which is emitted when inventory first reaches zero. Their physical meanings are summarized in [PITH_FULL_IMAGE:figures/… view at source ↗
Figure 7
Figure 7. Figure 7: Conditional intensity function of E1 (customer order) inferred by SNMPP on a sequence from the simulated supply-chain system. E Supply-Chain System Simulation We simulate event sequences using a hidden-state inventory generator that follows operational constraints and decision rules rather than a temporal point process. The model only observes the resulting event times and types. E.1 Event types The system… view at source ↗
Figure 8
Figure 8. Figure 8: Next-event time RMSE and event-type prediction accuracy over training epochs for the global Monte Carlo estimator (GMCE) and our stratified Monte Carlo estimator (Q = 1). F.2 Smoothness Parameter s Throughout our experiments, we set the smoothness parameter to s = 0.1 in the soft-clipping transformation (3). In this section, we examine the sensitivity of SNMPP to this choice. We vary 20 [PITH_FULL_IMAGE:f… view at source ↗
Figure 9
Figure 9. Figure 9: Next-event time RMSE and event-type prediction accuracy versus training epochs under different choices of Q. s ∈ {0.01, 0.05, 0.1, 0.5, 1.0, 10.0} and evaluate SNMPP on the StackOverflow (SO) dataset. We track next-event type prediction accuracy and next-event time RMSE over training epochs. As shown in [PITH_FULL_IMAGE:figures/full_fig_p021_9.png] view at source ↗
Figure 9
Figure 9. Figure 9: Next-event time RMSE and event-type prediction accuracy versus training epochs under different choices of Q. F.2 Smoothness Parameter s Throughout our experiments, we set the smoothness parameter to s = 0.1 in the soft-clipping transformation (3). In this section, we examine the sensitivity of SNMPP to this choice. We vary s ∈ {0.01, 0.05, 0.1, 0.5, 1.0, 10.0} and evaluate SNMPP on the StackOverflow (SO) d… view at source ↗
Figure 10
Figure 10. Figure 10: Next-event time RMSE and event-type prediction accuracy over training epochs with different choices of the smoothness parameter s in the soft-clipping transformation (3). 21 [PITH_FULL_IMAGE:figures/full_fig_p021_10.png] view at source ↗
read the original abstract

Multi-class event streams arise in numerous real-world applications, where uncovering structured, interpretable inter-event relationships, together with accurate prediction, remains a central challenge. Existing neural point process models are highly expressive but encode event interactions in a black-box manner, preventing explicit discovery of structured dependencies. In this paper, we propose a structured neural marked point process (SNMPP) that achieves high modeling flexibility while enabling explicit event-wise and class-wise relationship discovery from data. Our model constructs a product-form neural influence kernel composed of a signed interaction network over event types and a delay-aware monotonic temporal network. This design enables explicit characterization of inter-class influence topology -- including excitation, inhibition, and neutrality -- while flexibly capturing diverse temporal decay patterns and potential influence delays. For efficient learning, we develop a stratified Monte Carlo estimator for stochastic training. Extensive experiments on synthetic and real-world benchmark datasets validate the ability of our approach to uncover structured relationships and deliver strong predictive performance.

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

2 major / 2 minor

Summary. The paper proposes a Structured Neural Marked Point Process (SNMPP) for multi-class event streams. It constructs a product-form neural influence kernel as the product of a signed interaction network (over event types, capturing excitation/inhibition/neutrality) and a delay-aware monotonic temporal network (for flexible decay and delays). This enables explicit recovery of inter-class influence topology while retaining modeling flexibility. A stratified Monte Carlo estimator is introduced for stochastic training, with validation claimed on synthetic and real-world benchmarks for both structured relationship discovery and predictive performance.

Significance. If the separability assumption holds with low fidelity loss, the approach would meaningfully advance interpretable neural point processes by providing explicit topology recovery (including inhibition) without sacrificing expressiveness, which is valuable for domains like epidemiology or user behavior modeling where understanding signed influences matters. The stratified estimator and monotonic temporal component are practical contributions if empirically supported.

major comments (2)
  1. [Model construction (around the definition of the neural influence kernel)] The central interpretability claim rests on the product-form kernel faithfully approximating general inter-event influences. No section derives or bounds the approximation error of factoring influences into a class-wise signed matrix times a shared monotonic temporal function relative to an unrestricted bivariate kernel; this is load-bearing for the topology recovery guarantee and the claim that the design incurs 'no substantial loss of modeling power'.
  2. [Experiments] Abstract and experiments section assert validation on synthetic and real datasets with strong predictive performance and relationship discovery, yet no quantitative results, ablation studies on the separability assumption, or comparisons against post-hoc fitting baselines are referenced in the provided text. This undermines assessment of whether the explicit topology is recovered accurately or is an artifact of the factorization.
minor comments (2)
  1. [Model] Notation for the signed interaction network and the delay-aware temporal network should be introduced with explicit equations early in the model section to improve readability.
  2. [Learning] Clarify whether the monotonicity constraint on the temporal network is enforced via architecture (e.g., positive weights) or regularization, and discuss any impact on optimization stability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful and constructive comments on our manuscript. We address each of the major comments in detail below and describe the revisions we intend to make to strengthen the paper.

read point-by-point responses

  1. Referee: [Model construction (around the definition of the neural influence kernel)] The central interpretability claim rests on the product-form kernel faithfully approximating general inter-event influences. No section derives or bounds the approximation error of factoring influences into a class-wise signed matrix times a shared monotonic temporal function relative to an unrestricted bivariate kernel; this is load-bearing for the topology recovery guarantee and the claim that the design incurs 'no substantial loss of modeling power'.

    Authors: We agree that providing a formal bound on the approximation error would further support the claims. However, our model is proposed as a structured alternative rather than an approximation to a fully general bivariate kernel. The product form allows us to explicitly recover the signed interaction matrix, which is the key for interpretability in applications where understanding excitation and inhibition is important. The neural networks in each factor provide flexibility within the structured form. We did not include a theoretical analysis of the error in the original submission. In the revision, we will add a paragraph in the model section discussing the implications of the separability assumption and include additional experiments comparing to more flexible baselines to empirically assess any loss in modeling power. revision: yes

  2. Referee: [Experiments] Abstract and experiments section assert validation on synthetic and real datasets with strong predictive performance and relationship discovery, yet no quantitative results, ablation studies on the separability assumption, or comparisons against post-hoc fitting baselines are referenced in the provided text. This undermines assessment of whether the explicit topology is recovered accurately or is an artifact of the factorization.

    Authors: The full paper includes extensive quantitative evaluations in the experiments section. Specifically, we report predictive performance metrics such as negative log-likelihood and mean absolute error for event times on both synthetic data (with ground-truth interactions) and real-world datasets. We also present precision and recall for recovered interaction signs and topologies. Ablation studies are included to assess the contribution of the signed interaction network and the monotonic temporal network. Additionally, we compare the discovered relationships to those obtained by applying post-hoc interpretation techniques to a standard neural marked point process. We will update the abstract to reference these specific results and ensure all quantitative claims are backed by explicit references to tables and figures in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

Model proposal is self-contained design choice with no reduction to inputs

full rationale

The paper proposes a new SNMPP model whose central feature is the explicit construction of a product-form neural influence kernel from a signed interaction network over event types and a delay-aware monotonic temporal network. This architecture is presented as an engineering decision that directly enables the claimed interpretability properties (explicit excitation/inhibition/neutrality topology and flexible temporal patterns). No derivation chain is shown in which a 'prediction' or 'first-principles result' is obtained by fitting a parameter and then re-labeling a closely related quantity as an output; the separability is an input modeling assumption rather than a derived claim that collapses back to fitted data. The provided text contains no self-citations, uniqueness theorems, or ansatzes imported from prior author work that would make the central claim load-bearing on unverified self-reference. The work is therefore a self-contained model definition whose performance claims rest on empirical validation rather than circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

Review performed on abstract only; ledger entries are inferred from stated model components and standard point-process assumptions. Full paper likely contains additional fitted parameters and background results.

axioms (1)
  • domain assumption Multi-class event streams can be represented as marked point processes whose conditional intensity factors into an influence kernel.
    Standard modeling choice in the temporal point process literature referenced by the abstract.
invented entities (2)
  • Signed interaction network no independent evidence
    purpose: To encode excitation, inhibition, or neutrality between event classes.
    Core component of the product-form kernel introduced to achieve interpretability.
  • Delay-aware monotonic temporal network no independent evidence
    purpose: To capture arbitrary temporal decay shapes and possible influence delays.
    Second factor in the product-form kernel.

pith-pipeline@v0.9.0 · 5698 in / 1306 out tokens · 40919 ms · 2026-05-21T07:53:05.331131+00:00 · methodology

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

273 extracted references · 273 canonical work pages · 13 internal anchors

  1. [1]

    Translational and Clinical Pharmacology , volume=

    Pharmacodynamic principles and the time course of delayed and cumulative drug effects , author=. Translational and Clinical Pharmacology , volume=. 2018 , publisher=

    work page 2018

  2. [2]

    Doubly stochastic

    Grandell, Jan , year=. Doubly stochastic

    work page

  3. [3]

    International Conference on Learning Representations , year=

    Intensity-Free Learning of Temporal Point Processes , author=. International Conference on Learning Representations , year=

    work page

  4. [4]

    Residual

    Yuan, Ruoxin and Fang, Guanhua , booktitle=. Residual

    work page

  5. [5]

    Transactions on Machine Learning Research , issn=

    Exploring Generative Neural Temporal Point Process , author=. Transactions on Machine Learning Research , issn=. 2022 , url=

    work page 2022

  6. [6]

    Advances in Neural Information Processing Systems , volume=

    Add and thin: Diffusion for temporal point processes , author=. Advances in Neural Information Processing Systems , volume=

    work page

  7. [7]

    The Fourteenth International Conference on Learning Representations , year=

    Edit-Based Flow Matching for Temporal Point Processes , author=. The Fourteenth International Conference on Learning Representations , year=

    work page

  8. [8]

    Gavin Kerrigan and Kai Nelson and Padhraic Smyth , booktitle=. Event. 2026 , url=

    work page 2026

  9. [9]

    Charlin, Laurent and Ranganath, Rajesh and McInerney, James and Blei, David M , booktitle=. Dynamic

    work page

  10. [10]

    2004 , publisher=

    Convex Optimization , author=. 2004 , publisher=

    work page 2004

  11. [11]

    Mathematical Programming , year=

    Smooth minimization of non-smooth functions , author=. Mathematical Programming , year=

    work page

  12. [12]

    2016 , publisher=

    Deep Learning , author=. 2016 , publisher=

    work page 2016

  13. [13]

    , author=

    Scalable recommendation with hierarchical poisson factorization. , author=. UAI , pages=

    work page

  14. [14]

    Advances in neural information processing systems , volume=

    Content-based recommendations with Poisson factorization , author=. Advances in neural information processing systems , volume=

    work page

  15. [15]

    Journal of the American Statistical Association , volume=

    Regression methods for Poisson process data , author=. Journal of the American Statistical Association , volume=. 1987 , publisher=

    work page 1987

  16. [16]

    Healthcare , volume=

    The impact of delayed symptomatic treatment implementation in the intensive care unit , author=. Healthcare , volume=. 2021 , organization=

    work page 2021

  17. [17]

    Asia Pacific Management Review , volume=

    The effect of lead-time on supply chain resilience performance , author=. Asia Pacific Management Review , volume=. 2019 , publisher=

    work page 2019

  18. [18]

    The International Journal of Advanced Manufacturing Technology , volume=

    A study of lead time variation impact on supply chain performance , author=. The International Journal of Advanced Manufacturing Technology , volume=. 2009 , publisher=

    work page 2009

  19. [19]

    Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages =

    Yuan, Yuan and Ding, Jingtao and Shao, Chenyang and Jin, Depeng and Li, Yong , title =. Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages =. 2023 , isbn =. doi:10.1145/3580305.3599511 , abstract =

    work page doi:10.1145/3580305.3599511 2023

  20. [20]

    Advances in neural information processing systems , volume=

    Monotonic networks , author=. Advances in neural information processing systems , volume=

    work page

  21. [21]

    Infinite

    Xu, Zenglin and Yan, Feng and Qi, Yuan , booktitle=. Infinite

    work page

  22. [22]

    Advances in neural information processing systems , volume=

    Neural ordinary differential equations , author=. Advances in neural information processing systems , volume=

    work page

  23. [23]

    Dintucker: Scaling up

    Zhe, Shandian and Qi, Yuan and Park, Youngja and Xu, Zenglin and Molloy, Ian and Chari, Suresh , booktitle=. Dintucker: Scaling up

    work page

  24. [24]

    2012 , school=

    Scalable inference for structured Gaussian process models , author=. 2012 , school=

    work page 2012

  25. [25]

    Forty-second International Conference on Machine Learning , year=

    Toward Efficient Kernel-Based Solvers for Nonlinear PDEs , author=. Forty-second International Conference on Machine Learning , year=

    work page

  26. [26]

    Solving High Frequency and Multi-Scale PDEs with

    Shikai Fang and Madison Cooley and Da Long and Shibo Li and Robert Kirby and Shandian Zhe , booktitle=. Solving High Frequency and Multi-Scale PDEs with

    work page

  27. [27]

    International Conference on Learning Representations , year=

    Neural Spatio-Temporal Point Processes , author=. International Conference on Learning Representations , year=

    work page

  28. [28]

    Proceedings of The 4th Annual Learning for Dynamics and Control Conference , pages =

    Neural Point Process for Learning Spatiotemporal Event Dynamics , author =. Proceedings of The 4th Annual Learning for Dynamics and Control Conference , pages =. 2022 , editor =

    work page 2022

  29. [29]

    International Conference on Learning Representations , year=

    Decoupled Weight Decay Regularization , author=. International Conference on Learning Representations , year=

    work page

  30. [30]

    Automatic Integration for Spatiotemporal Neural Point Processes , url =

    Zhou, Zihao and Yu, Rose , booktitle =. Automatic Integration for Spatiotemporal Neural Point Processes , url =

    work page

  31. [31]

    2024 , booktitle=

    EasyTPP: Towards Open Benchmarking Temporal Point Processes , author=. 2024 , booktitle=

    work page 2024

  32. [32]

    arXiv preprint arXiv:2307.08097 , year=

    Easytpp: Towards open benchmarking temporal point processes , author=. arXiv preprint arXiv:2307.08097 , year=

    work page arXiv

  33. [33]

    Annals of the Institute of Statistical Mathematics , volume=

    Space-time point-process models for earthquake occurrences , author=. Annals of the Institute of Statistical Mathematics , volume=. 1998 , publisher=

    work page 1998

  34. [34]

    Biometrics , volume=

    A space--time conditional intensity model for invasive meningococcal disease occurrence , author=. Biometrics , volume=. 2012 , publisher=

    work page 2012

  35. [35]

    2016 , school=

    Point process modeling with spatiotemporal covariates for predicting crime , author=. 2016 , school=

    work page 2016

  36. [36]

    Advances in neural information processing systems , volume=

    Denoising diffusion probabilistic models , author=. Advances in neural information processing systems , volume=

    work page

  37. [37]

    2008 , publisher=

    An introduction to the theory of point processes: volume II: general theory and structure , author=. 2008 , publisher=

    work page 2008

  38. [38]

    Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages=

    Recurrent marked temporal point processes: Embedding event history to vector , author=. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages=

    work page

  39. [39]

    2008 , publisher=

    Graphical models, exponential families, and variational inference , author=. 2008 , publisher=

    work page 2008

  40. [40]

    Introducing TensorFlow Feature Columns , Year =

    TensorFlowTeam , Institution =. Introducing TensorFlow Feature Columns , Year =

    work page

  41. [41]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Variational Inference for Sparse Gaussian Process Modulated Hawkes Process , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  42. [42]

    Journal of Machine Learning Research , volume=

    Efficient Inference for Nonparametric Hawkes Processes Using Auxiliary Latent Variables , author=. Journal of Machine Learning Research , volume=

    work page

  43. [43]

    International Conference on Machine Learning , pages=

    Variational inference for Gaussian process modulated Poisson processes , author=. International Conference on Machine Learning , pages=. 2015 , organization=

    work page 2015

  44. [44]

    Proceedings of the International Conference on Learning Representations (ICLR) , year=

    EFFICIENT INFERENCE OF FLEXIBLE INTERACTION IN SPIKING-NEURON NETWORKS , author=. Proceedings of the International Conference on Learning Representations (ICLR) , year=

    work page

  45. [45]

    International Conference on Machine Learning , pages=

    Transformer hawkes process , author=. International Conference on Machine Learning , pages=. 2020 , organization=

    work page 2020

  46. [46]

    Proceedings of the Tenth International Conference on Learning Representations (ICLR) , year=

    Transformer Embeddings of Irregularly Spaced Events and Their Participants , author=. Proceedings of the Tenth International Conference on Learning Representations (ICLR) , year=

    work page

  47. [47]

    Rubanova, Yulia and Chen, Ricky T. Q. and Duvenaud, David K , booktitle =. Latent Ordinary Differential Equations for Irregularly-Sampled Time Series , url =

    work page

  48. [48]

    Advances in Neural Information Processing Systems , pages=

    Fully neural network based model for general temporal point processes , author=. Advances in Neural Information Processing Systems , pages=

    work page

  49. [49]

    Verlag New York Berlin Heidelberg: Springer , year=

    An introduction to the theory of point processes, volume 1: Elementary theory and methods , author=. Verlag New York Berlin Heidelberg: Springer , year=

    work page

  50. [50]

    International Conference on Machine Learning , pages=

    Self-attentive hawkes process , author=. International Conference on Machine Learning , pages=. 2020 , organization=

    work page 2020

  51. [51]

    Proceedings of the 27th ACM International Conference on Information and Knowledge Management , pages=

    Regularizing matrix factorization with user and item embeddings for recommendation , author=. Proceedings of the 27th ACM International Conference on Information and Knowledge Management , pages=

    work page

  52. [52]

    Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval , pages=

    Warm Up Cold-start Advertisements: Improving CTR Predictions via Learning to Learn ID Embeddings , author=. Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval , pages=

    work page

  53. [53]

    Categorical Reparameterization with Gumbel-Softmax

    Categorical reparameterization with gumbel-softmax , author=. arXiv preprint arXiv:1611.01144 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  54. [54]

    Stochastic gradient descent with b iased but consistent gradient estimators,

    Stochastic gradient descent with biased but consistent gradient estimators , author=. arXiv preprint arXiv:1807.11880 , year=

    work page arXiv

  55. [55]

    Tensorflow:

    Abadi, Mart. Tensorflow:. 12th \ USENIX \ Symposium on Operating Systems Design and Implementation ( \ OSDI \ 16) , pages=

    work page

  56. [56]

    Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages=

    SEISMIC: A Self-Exciting Point Process Model for Predicting Tweet Popularity , author=. Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages=. 2015 , organization=

    work page 2015

  57. [57]

    Advances in Neural Information Processing Systems , pages=

    The neural hawkes process: A neurally self-modulating multivariate point process , author=. Advances in Neural Information Processing Systems , pages=

    work page

  58. [58]

    Advances in Neural Information Processing Systems , pages=

    Online learning for multivariate Hawkes processes , author=. Advances in Neural Information Processing Systems , pages=

    work page

  59. [59]

    The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables

    The concrete distribution: A continuous relaxation of discrete random variables , author=. arXiv preprint arXiv:1611.00712 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  60. [60]

    Advances in Neural Information Processing Systems , pages=

    Stochastic Nonparametric Event-Tensor Decomposition , author=. Advances in Neural Information Processing Systems , pages=

    work page

  61. [61]

    Pour une analyse krigeante des donn

    Matheron, G , journal=. Pour une analyse krigeante des donn

    work page

  62. [62]

    Mathematical Geology , volume=

    Linear coregionalization model: tools for estimation and choice of cross-variogram matrix , author=. Mathematical Geology , volume=. 1992 , publisher=

    work page 1992

  63. [63]

    Auto-Encoding Variational Bayes

    Auto-encoding variational bayes , author=. arXiv preprint arXiv:1312.6114 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  64. [64]

    Adam: A Method for Stochastic Optimization

    Adam: A method for stochastic optimization , author=. arXiv preprint arXiv:1412.6980 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  65. [65]

    The 22nd International Conference on Artificial Intelligence and Statistics , pages=

    Scalable High-Order Gaussian Process Regression , author=. The 22nd International Conference on Artificial Intelligence and Statistics , pages=

    work page

  66. [66]

    1978 , publisher=

    Mining geostatistics , author=. 1978 , publisher=

    work page 1978

  67. [67]

    Artificial Intelligence and Statistics , pages=

    Deep gaussian processes , author=. Artificial Intelligence and Statistics , pages=

    work page

  68. [68]

    Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , volume=

    Nonlinear information fusion algorithms for data-efficient multi-fidelity modelling , author=. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , volume=. 2017 , publisher=

    work page 2017

  69. [69]

    Deep Gaussian Processes for Multi-fidelity Modeling

    Deep Gaussian Processes for Multi-fidelity Modeling , author=. arXiv preprint arXiv:1903.07320 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 1903

  70. [70]

    Biometrika , volume=

    Predicting the output from a complex computer code when fast approximations are available , author=. Biometrika , volume=. 2000 , publisher=

    work page 2000

  71. [71]

    Mathematics of computation , volume=

    Numerical solution of the Navier-Stokes equations , author=. Mathematics of computation , volume=

    work page

  72. [72]

    Journal of Structural Mechanics , volume=

    On the design of compliant mechanisms using topology optimization , author=. Journal of Structural Mechanics , volume=. 1997 , publisher=

    work page 1997

  73. [73]

    Fast forward selection to speed up sparse

    Seeger, Matthias and Williams, Christopher and Lawrence, Neil , booktitle=. Fast forward selection to speed up sparse

    work page

  74. [74]

    Advances in Neural Information Processing Systems 15 , year=

    Anton Schwaighofer and Volker Tresp , title=. Advances in Neural Information Processing Systems 15 , year=

    work page

  75. [75]

    Journal of Computational Physics , volume=

    Numerical study of viscous flow in a cavity , author=. Journal of Computational Physics , volume=. 1973 , publisher=

    work page 1973

  76. [76]

    Advances in neural information processing systems , pages=

    Multi-task Gaussian process prediction , author=. Advances in neural information processing systems , pages=

    work page

  77. [77]

    International Conference on Machine Learning , pages=

    Kernel interpolation for scalable structured Gaussian processes (KISS-GP) , author=. International Conference on Machine Learning , pages=

    work page

  78. [78]

    International Conference on Artificial Intelligence and Statistics , pages=

    Scalable Gaussian Processes with Billions of Inducing Inputs via Tensor Train Decomposition , author=. International Conference on Artificial Intelligence and Statistics , pages=

    work page

  79. [79]

    SIAM Journal on Matrix Analysis and Applications , volume=

    Decompositions of a higher-order tensor in block terms?Part II: Definitions and uniqueness , author=. SIAM Journal on Matrix Analysis and Applications , volume=. 2008 , publisher=

    work page 2008

  80. [80]

    SIAM Journal on Scientific Computing , volume=

    Tensor-train decomposition , author=. SIAM Journal on Scientific Computing , volume=. 2011 , publisher=

    work page 2011

Showing first 80 references.