Pith. sign in

REVIEW 4 major objections 4 minor 13 references

Meta Learning with Relational Information for Short Sequences

T0 review · 4 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read HARMLESS models short event sequences as a mixture of Hawkes processes tied together by a relational graph, and claims this improves prediction of the next event time over baselines.

desk verdict Useful empirical contribution, but the model is not a coherent generative process; the held-out evaluation mostly survives, and the paper needs reframing plus code. read the letter →

arxiv 1909.02105 v1 pith:4SGN3ZPU submitted 2019-09-04 cs.LG stat.ML

Yujia Xie , Haoming Jiang , Feng Liu , Tuo Zhao , Hongyuan Zha This is my paper · ORCID
classification cs.LGstat.ML
keywords meta-learningHawkesprocesspointshorteventsequencesmixedmembershipstochasticblockmodelrelationalnetworkvariationalinferenceprediction
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Short event sequences, such as a user's tweets or job changes, are too sparse for reliable per-subject inference, yet subjects share behavior within overlapping communities. The paper proposes HARMLESS, which gives each subject a mixture of Hawkes processes—one per identity—and ties the mixture weights to a mixed-membership stochastic blockmodel on a relational graph. Each identity's Hawkes parameters are adapted from a shared meta-parameter by one MAML-type gradient step, so short sequences borrow strength from the community while retaining individual flexibility. The paper argues that joint training of the sequence likelihoods and the graph structure yields better predictions of the next event timestamp than ignoring either source of information, and supports this with synthetic experiments, four real datasets, and ablations.

What carries the argument

The load-bearing object is the coupled generative model in Figure 1: per-subject identity proportions $\pi_i$ (with Dirichlet prior $\alpha$) feed both a mixture-of-Hawkes likelihood $p(\tau_i)=\sum_{k}\pi_{i,k}L_i(\tilde{\theta}_k^{(i)})$ and an MMB edge likelihood $Y_{ij}\sim\text{Bernoulli}(z_{i\to j}^T B z_{i\leftarrow j})$. The identities $z_i$ for sequences and $z_{i\to j},z_{i\leftarrow j}$ for edges are linked by sharing $\pi_i$, which is what carries relational information into the point processes. The second mechanism is Eq. (9), $\tilde{\theta}_k^{(i)}=\theta_k-\eta D(\log L_i,\theta_k)$: each individual Hawkes parameter is one gradient step from the community common model, so the adapted models are close within a community without being identical. Variational mean-field updates propagate the graph and sequence evidence into the community assignments, and a gradient ascent update refines the common Hawkes parameters.

What would settle it

On a synthetic dataset generated from known communities, run HARMLESS twice from different random initializations of $\theta_k$ and different inner step sizes, then check whether the fixed-point equation $\tilde{\theta}_k^{(i)}=\theta_k-\eta\nabla\log L_i(\tilde{\theta}_k^{(i)})$ admits materially different adapted parameters; if the predicted next-event log-likelihood changes across initializations, the self-referential adaptation is not well-defined.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that relational information can be folded into meta-learning for point processes. The generative model assumes each subject $i$ has identity proportions $\pi_i$; the same proportions drive (i) which Hawkes process generates the subject's event sequence and (ii) which identity is active when the subject forms a network edge. Edges are modeled by a mixed-membership stochastic blockmodel, so subjects in the same mixed community share Hawkes parameters, while MAML adaptation lets each subject's parameters deviate slightly from the community's common $\theta_k$. The resulting variational meta expectation maximization algorithm alternates closed-form updates for identity assignments and gradient updates for $\theta_k$. Experimental results show higher held-out log-likelihood for next-event prediction than four baselines and than a two-step procedure that learns communities from the graph first; the paper reports this advantage as consistent across synthetic and real data.

Load-bearing premise

The individual Hawkes parameters in Eq. (9) are defined by a gradient step that depends on the likelihood of the same sequence whose generative probability uses those very parameters; the paper does not prove that this self-referential equation has a unique or stable solution, and if it does not, the reported predictive gains may be artifacts of the update schedule.

Editorial extensions

If this is right

  • If HARMLESS's central claim is right, then in settings where each subject's observed history is short, the relational graph can substitute for longer histories: the same identity proportions govern both who a subject links to and how the subject's events arrive.
  • Jointly estimating community structure from events and edges should beat a two-stage pipeline that first detects communities from the adjacency matrix alone and then fits event models; the paper reports exactly this on synthetic data.
  • All three adaptation variants (MAML, FOMAML, Reptile) should inherit the relational benefit, though Reptile's utility may be dataset-dependent, as the paper notes on synthetic data.
  • Ablation results imply that removing either graph information, within-community heterogeneity, or multiple identities lowers predictive log-likelihood on the LinkedIn dataset, so all three are load-bearing ingredients.
  • The variational meta EM procedure is designed to scale to large collections, which the paper demonstrates on the StackOverflow dataset with over thirteen thousand subjects.

Reading between the lines

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

  • Because the relational side of the model only uses identity proportions, the same coupling could be applied to other event models such as Poisson or neural point processes, giving a general recipe for injecting graph structure into meta-learning for sequences.
  • The paper's Eq. (9) defines each individual parameter in terms of the likelihood it generates; a natural robustness check is to initialize $\theta_k$ differently and see whether the converged predictive likelihood is unaffected, since the authors give no fixed-point guarantee.
  • The community assignments learned while fitting HARMLESS can be read as behavioral profiles shared across the graph; one testable extension is to compare those profiles with observable user attributes on public datasets.
  • Appendix A hints that the model could be rewritten as a fully Bayesian hierarchical meta-learner; if that rewrite is carried out, the variational EM would produce posterior intervals over individual adapted parameters instead of point estimates.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes HARMLESS, a method for learning heterogeneous Hawkes processes from short event sequences when a relational graph among the subjects is available. The model combines a mixed-membership stochastic blockmodel for the graph with a mixture of Hawkes processes for the sequences, and uses MAML-type updates to adapt per-identity parameters from shared common parameters. Inference is performed by a stochastic variational meta expectation-maximization algorithm. The authors report improvements in held-out log-likelihood over four baselines on synthetic data and four real-world datasets, and include an ablation study on LinkedIn data. The central methodological issue is that the proposed generative model is self-referential and is not shown to define a valid probability distribution, which undermines the probabilistic interpretation of the reported likelihood scores.

Significance. If the model definition were coherent, the combination of mixed-membership blockmodels with MAML-style adaptation for Hawkes processes would be a novel and potentially useful contribution to meta-learning for event sequences. The paper addresses an important practical setting, short and heterogeneous sequences with relational side information, and the empirical evaluation covers several real datasets with an ablation study. The algorithm is scalable via stochastic updates and the paper explicitly discusses the MAML/FOMAML/Reptile variants. However, the core generative model is not well-defined as stated: Equation (9) makes the per-sequence parameter a function of the likelihood of the same sequence used in the mixture density, so the paper does not establish that Equation (3) is a valid density or that the variational EM objective is an ELBO. Because the claimed gains are reported as log-likelihoods, this affects the central claim rather than being a presentation issue.

major comments (4)
  1. [Section 3, Eq. (3); Section 4, Eq. (9); Appendix C, p(T|z)] The model is self-referential as written. Substituting Eq. (9) into Eq. (3) gives p(τ_i) = Σ_k π_ik L(θ_k − η D(log L_i, θ_k); τ_i), where L_i is the likelihood of the very same sequence τ_i. Thus the parameter of the Hawkes component is a function of the observation whose probability is being computed. The paper never proves that this plug-in function integrates to one over τ_i, and the generative process in Section 3 cannot be simulated because drawing τ_i requires θ~_ik, which depends on τ_i. Consequently, the ELBO in Eq. (4) is not a lower bound on a marginal likelihood of a well-defined model, and the 'log-likelihood' values in Table 2 and Figure 2 are plug-in scores without a common normalization. The authors should either prove that Eq. (3) with Eq. (9) defines a proper density (possibly by making the adaptation depend only on a training prefix and providing a normalizing constant), or reformulate HARMLESS as optimizing a well-defined predictive objective and revise all Bayesian/generative claims accordingly.
  2. [Section 5.1, data generation] The synthetic data are not generated from the proposed model. The paper states that the per-sequence parameters are generated as ~µ_zi ~ N(µ_zi, 0.01), ~δ_zi ~ N(δ_zi, 0.01), ~ω_zi ~ N(ω_zi, 0.05), which is an additive perturbation mechanism, not the MAML update in Eq. (9). As a result, the synthetic experiments do not test whether the self-referential likelihood in Eq. (3) is coherent or whether the adaptation rule in Eq. (9) recovers the generating parameters. The authors should generate data from the actual model (including the Eq. (9) adaptation) or clearly state that the synthetic study only validates the algorithm under a different generative mechanism.
  3. [Section 5.1 and Table 2] The claim that HARMLESS is 'consistently better than the baselines' is not supported by the reported numbers. In Table 2, on StackOverflow, HARMLESS (MAML) obtains 1.3886 ± 0.0082, which is worse than MLE-Com (1.5594 ± 0.0134) and MTL (1.4910 ± 0.0089). On 911-Calls, the HARMLESS variants have means that straddle MLE-Com within one standard error (e.g., MAML 4.5208 ± 0.3256 vs. MLE-Com 4.5111 ± 0.3192, Reptile 4.4929 ± 0.3503), and only HARMLESS (FOMAML) is reported to be significantly better via a paired t-test. The statement should be restricted to the settings and variants where it actually holds, or additional paired tests and error bars in Figure 2 should be provided.
  4. [Section 4 and Appendix D] The relationship between the training objective and the evaluation metric needs clarification. If the held-out last timestamp is excluded when computing L_i and the variational parameters, then the predictive density in Appendix D is a conditionally normalized one-step-ahead density, which makes the held-out comparison meaningful despite the training objective being non-standard. However, the paper does not explicitly state that γ_i,k and θ~_ik are computed using only the training prefix, and Appendix D writes 'the posterior distribution of z_i is q2(z_i)' using parameters trained on the data. The authors should specify the exact information used to construct the predictive density for a test timestamp; otherwise the evaluation could be circular if the full sequence, including the test event, enters the adaptation step.
minor comments (4)
  1. [Section 5, first paragraph] There is a typo: 'oue experiment settings' should be 'our experiment settings'.
  2. [Section 3, Generative process] The generative process lists the parameters as α, θ~_k^(i), and B, but the common parameters θ_k are the actual objects being optimized in Eq. (10). Please clarify that θ_k are also parameters of the model and specify their role in the generative story.
  3. [Appendix E] The paper says 'we adopt α = 1_K' in the experiments, but Section 4 includes an M-step update for α in Eq. (11). Please state clearly whether α is updated or fixed in the reported experiments, and if fixed, remove Eq. (11) from the algorithm or mark it as optional.
  4. [Section 5.1 and Figure 2] The two-step procedure is described only briefly and its details are not given in Appendix E. Please specify how the MMB model is trained, how identities are fixed for the subsequent HARMLESS training, and whether the same hyperparameter tuning is used. Also, Figure 2 appears to have no error bars, making it difficult to assess the claimed consistent improvement over the two-step procedure.

Circularity Check

1 steps flagged · score 7.0 of 10

Eq. (3) defines the sequence 'mixture likelihood' using adapted Hawkes parameters that depend on the very same sequence's log-likelihood, so the claimed generative process and reported log-likelihoods are self-referential plug-in scores rather than a well-defined probability model; the held-out next-event prediction itself is not circular.

  1. self definitional [Section 3, Eq. (3) and the 'Generative process' bullet; Section 4, Eq. (9); Appendix D evaluation description.]
    "The likelihood for the i-th sequence τ_i is p(τ_i) = Σ_{k=1}^K π_{i,k} L_i(~θ^{(i)}_k). (3) ... we adapt the k-th common model H(θ_k) to sequence i using MAML-type updates, ~θ^{(i)}_k = θ_k − η D(log L_i, θ_k). (9) ... Sample the i-th sequence τ_i from the mixture of Hawkes processes described in (3). ... In the generative process, for subject i, we first sample z_i, then use parameter ~θ^{(i)}_{z_i} = θ_{z_i} − η D(L_i, θ_{z_i})."

    Substituting (9) into (3) gives p(τ_i) = Σ_k π_{i,k} L(θ_k − η ∇_θ log L(θ_k; τ_i); τ_i), so the parameter of the density over τ_i is defined through the log-likelihood of the same τ_i. The paper's own generative-process bullet requires sampling τ_i from (3), but the adapted parameters needed for (3) cannot be computed until τ_i is observed. No fixed-point, normalization, or proper-scoring argument is supplied; the Appendix D statement that to predict the next arrival one first samples z_i, then uses parameter ~θ^{(i)}_{z_i} = θ_{z_i} − η D(L_i, θ_{z_i}) presupposes the L_i of the sequence being generated. Hence Eq.

full rationale

The held-out evaluation is not circular: the last timestamp is removed before training, and the conditional predictive density for the next arrival is evaluated given the training prefix and adapted parameters, so no test timestamp is used to fit Eq. (9). The circularity is in the model definition. Eq. (3) is announced as 'The likelihood for the i-th sequence' and the generative process says sequences are sampled from it, while Eq. (9) makes the mixture component parameters functions of the same sequence's log-likelihood. The paper never proves that the resulting plug-in function integrates to one or corresponds to a well-defined generative mechanism. This is not a matter of self-citation: MAML (Finn et al., 2017) is an external, legitimate source, and the graph/community components are standard MMB machinery. The flaw is internal to the paper's equations: the 'Bayesian mixture' interpretation of Section 3 is not supported by a realizable generative process. Because the paper's reported scores are explicitly called log-likelihoods and are used for the 'consistently better' claim, the self-referential definition of Eq. (3)/(9) directly affects the central quantitative claim, though it does not by itself force the empirical ranking (the next-event prediction could still be a reasonable conditional score). I therefore score this as a substantial, partial circularity in the model construction, not a full reduction of the experiments to the input. No additional circular steps involving self-citation or fitted-input-called-prediction were found.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The model relies on several standard building blocks (Hawkes likelihood, MMB, mean-field VI). The distinctive new assumption is the self-referential MAML adaptation in Eq. (9), and the hyperparameters K, learning rates, alpha, and nu are tuned or fixed by hand. No fundamentally new entities are postulated.

free parameters (5)
  • K (number of identities/communities) = 6 for synthetic ground truth; 3 for most real experiments; searched over 2,3,5 (real) and 3,6,10 (synthetic)
    The number of mixture components is chosen by model selection on validation, not determined by the model.
  • Learning rates eta (inner), eta_theta (outer), eta_alpha = Selected via grid search, e.g., eta=5e-1 for LinkedIn MAML, eta=1e-3 for StackOverflow MAML
    The MAML-type adaptation and the outer gradient updates require step sizes that are tuned per dataset.
  • alpha (Dirichlet concentration) = 1_K (vector of ones)
    Fixed after trial experiments reported in Appendix E; no data-driven estimate.
  • nu (regularization coefficient) = 1e-2
    Chosen as a small constant after testing that results are insensitive from 1e-10 to 1e-1.
  • Initial values for theta_k and B = Not specified in the paper
    The algorithm's behavior depends on initialization, which is not reported.
assumptions (6)
  • standard math Hawkes process likelihood formula (Eq. 2) is valid for the exponential-decay intensity in Eq. (13)
    Standard point process likelihood from Laub et al. (2015), cited in the paper.
  • domain assumption MMB generative process for edges, with edge probability z^T B z and Dirichlet prior on pi
    Adopted from Airoldi et al. (2008); it assumes each edge is generated by a single Bernoulli draw from the two incident identity indicators.
  • standard math Mean-field variational family provides a good enough approximation for inference
    Standard variational inference approximation, used in Appendix C.
  • ad hoc to paper Individual Hawkes parameters theta_i,k are well-defined as theta_k - eta D(log L_i, theta_k)
    This is a new assumption introduced in Eq. (9). It makes each parameter a function of the observed sequence's likelihood, which is not a standard generative modeling assumption.
  • domain assumption One-step MAML adaptation captures within-community heterogeneity for short sequences
    The paper assumes that a single gradient step from the common model is enough to model individual variation; no analysis is provided.
  • domain assumption Sequences and edges are conditionally independent given the latent identities
    The graphical model in Fig. 1 assumes no direct dependence between the event sequence and the edge variables except through the identity proportions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Meta Learning with Relational Information for Short Sequences." pith.science (2026). https://pith.science/paper/4SGN3ZPU

@misc{pith2026190902105,
  author       = {Pith},
  title        = {Pith review of: Meta Learning with Relational Information for Short Sequences},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4SGN3ZPU}},
  note         = {Machine review of arXiv:1909.02105}
}
read the original abstract

This paper proposes a new meta-learning method -- named HARMLESS (HAwkes Relational Meta LEarning method for Short Sequences) for learning heterogeneous point process models from short event sequence data along with a relational network. Specifically, we propose a hierarchical Bayesian mixture Hawkes process model, which naturally incorporates the relational information among sequences into point process modeling. Compared with existing methods, our model can capture the underlying mixed-community patterns of the relational network, which simultaneously encourages knowledge sharing among sequences and facilitates adaptive learning for each individual sequence. We further propose an efficient stochastic variational meta expectation maximization algorithm that can scale to large problems. Numerical experiments on both synthetic and real data show that HARMLESS outperforms existing methods in terms of predicting the future events.

Figures

Figures reproduced from arXiv: 1909.02105 by the authors.

Figure 1
Figure 1. Probabilisitic graph of the suggested model. The yel￾low nodes are parameters, white nodes are latent variables, and the gray nodes are observed variables. The solid arrows represent proba￾bilistic mapping, while the hol￾low arrows represent the deter￾ministic mapping. We next introduce the meta learning method for analyzing short sequences. Suppose we are given a collection of sequences T = {τ1, τ2 · · · , τN }. We… view at source ↗
Figure 2
Figure 2. Plot of synthetic data. S = 1. Benefit of joint training. To validate the benefit of joint training on graphs and sequences, we compare HARMLESS result with a two step procedure: We first train an MMB model and obtain the identities, and train HARMLESS (MAML) with fixed identities. In [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

13 extracted references · 9 canonical work pages

  1. [1]

    , BACRY, E

    ACHAB , M. , BACRY, E. , GAÏFFAS , S. , MASTROMATTEO , I. and MUZY, J.-F. (2017). Uncovering causality from multivariate hawkes integrated cumulants. The Journal of Machine Learning Research, 18 6998–7025. AIROLDI , E. M. , BLEI , D. M. , FIENBERG , S. E. and XING , E. P. (2008). Mixed membership stochastic blockmodels. Journal of machine learning researc...

    work page 2017

  2. [2]

    LAUB, P. J. , TAIMRE , T. and POLLETT , P. K. (2015). Hawkes processes. arXiv preprint arXiv:1507.02822. LI, L. and ZHA, H. (2013). Dyadic event attribution in social networks with mixtures of hawkes processes. In Proceedings of the 22nd ACM international conference on Information & Knowledge Management. ACM. LINDERMAN , S. and ADAMS , R. (2014). Discover...

    arXiv 2015

  3. [6]

    JMLR. org. FINN , C. , XU, K. and LEVINE , S. (2018). Probabilistic model-agnostic meta-learning. In Advances in Neural Information Processing Systems. FOX, E. W. , SHORT, M. B. , SCHOENBERG , F. P., CORONGES , K. D. and BERTOZZI , A. L. (2016). Modeling e-mail networks and inferring leadership using self-exciting point processes. Journal of the American ...

    arXiv 2018

  4. [8]

    JMLR. org. NICHOL , A. , ACHIAM , J. and SCHULMAN , J. (2018). On first-order meta-learning algorithms. arXiv preprint arXiv:1803.02999. NICHOL , A. and SCHULMAN , J. (2018). Reptile: a scalable metalearning algorithm. arXiv preprint arXiv:1803.02999. 10 OGATA, Y. (1999). Seismicity analysis through point-process modeling: A review. In Seismicity patterns,...

    arXiv 2018

  5. [9]

    SANTORO , A

    Wiley New York. SANTORO , A. , BARTUNOV , S. , BOTVINICK , M. , WIERSTRA , D. and LILLICRAP , T. (2016). Meta-learning with memory-augmented neural networks. In International conference on machine learning. SNELL , J., SWERSKY , K. and ZEMEL , R. (2017). Prototypical networks for few-shot learning. In Advances in Neural Information Processing Systems. SUN...

    work page 2016

  6. [11]

    JMLR. org. XU, H. and ZHA, H. (2017). A dirichlet mixture model of hawkes processes for event sequence clustering. In Advances in Neural Information Processing Systems. YANG , S.-H. and ZHA, H. (2013). Mixture of mutually exciting processes for viral diffusion. In International Conference on Machine Learning. ZAREZADE , A. , KHODADADI , A. , FARAJTABAR , ...

    arXiv 2017

  7. [13]

    Update forq3 andq4 The derivation of update forq3 andq4 is very similar to the update forq2, so we will not elaborate on that

    After some derivation, we have γi,k←L i(θk−ηD(Li,θk)) exp ( fdg(βi,k)−fdg( ∑ 𝓁 βi,𝓁) ) , (16) γi,k← γi,k∑ 𝓁γi,𝓁 , (17) wherefdg is the digamma function. Update forq3 andq4 The derivation of update forq3 andq4 is very similar to the update forq2, so we will not elaborate on that. Readers who are interested might also refer to Airoldi et al. (2008). The upd...

    work page 2008

  8. [40]

    CHALMERS , D

    John Wiley & Sons. CHALMERS , D. J. (1991). The evolution of learning: An experiment in genetic connectionism. In Connectionist Models. Elsevier, 81–90. CLEEREMANS , A. and MCCLELLAND , J. L. (1991). Learning the structure of event sequences. Journal of Experimental Psychology: General, 120

    work page 1991

Show all 13 references
  1. [43]

    and CARIN , L

    XU, H., LUO, D., CHEN , X. and CARIN , L. (2017a). Benefits from superposed hawkes processes. arXiv preprint arXiv:1710.05115. XU, H., LUO, D. and ZHA, H. (2017b). Learning hawkes processes from short doubly-censored event sequences. In Proceedings of the 34th International Con...

    2017 arXiv

  2. [70]

    JMLR. org. FARAJTABAR , M. , YE, X. , HARATI , S. , SONG , L. and ZHA, H. (2016). Multistage campaigning in social networks. In Advances in Neural Information Processing Systems. 9 FINN , C. , ABBEEL , P. and LEVINE , S. (2017). Model-agnostic meta-learning for fast adaptation...

    2016

  3. [157]

    and HAUTSCH , N

    BAUWENS , L. and HAUTSCH , N. (2009). Modelling financial high frequency data using point processes. In Handbook of financial time series. Springer, 953–979. BENGIO , Y., BENGIO , S. and CLOUTIER , J. (1990). Learning a synaptic learning rule. Université de Montréal, Département...

    2009

  4. [235]

    , DAHLHAUS , R

    EICHLER , M. , DAHLHAUS , R. and DUECK , J. (2017). Graphical modeling for multivariate hawkes processes with nonparametric link functions. Journal of Time Series Analysis, 38 225–242. FARAJTABAR , M. , YANG , J. , YE, X. , XU, H. , TRIVEDI , R. , KHALIL , E. , LI, S. , SONG ,...

    2017

  5. [2017]

    Reptile (Nichol and Schulman, 2018), a variant of MAML, is proposed to simplify the computation of MAML

    opens another line of research, i.e., it designs an optimization scheme so that the model can fast adapt to new tasks. Reptile (Nichol and Schulman, 2018), a variant of MAML, is proposed to simplify the computation of MAML. None of those works, however, considers the relationa...

    2018

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.