Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

A unifying framework for generalised Bayesian online learning in non-stationary environments

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A single framework rewrites non-stationary online learning as five interchangeable choices.

desk verdict BONE's taxonomy is a keeper; its new method has a data-peeking problem and thin stats in two key experiments. read the letter →

arxiv 2411.10153 v3 pith:6QKS2YPX submitted 2024-11-15 stat.ML cs.LG

Gerardo Duran-Martin , Leandro Sánchez-Betancourt , Alexander Y. Shestopaloff , Kevin Murphy This is my paper · ORCID
classification stat.MLcs.LG MSC 62F1562M2068T05
keywords onlinelearningnon-stationaryenvironmentsBayesianchangepointdetectionrunlengthgeneralisedBayesprequentialforecastingcontinualstate-spacemodels
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

This paper claims that probabilistic online learning in changing environments reduces to five independent choices: what model generates measurements, what auxiliary variable tracks the changes, what prior over parameters that variable implies, how the parameter posterior is approximated, and how hypotheses about the auxiliary variable are weighted. It packages these five slots as the BONE framework and shows that a wide range of existing methods, from Kalman filters to Bayesian online changepoint detection to continual-learning and bandit algorithms, are instances of this structure. The paper also fills the slots with a new combination, RL[1]-OUPR*, which tracks a single runlength hypothesis and switches between gradual Ornstein-Uhlenbeck drift and a hard reset depending on the estimated probability that a changepoint has occurred. On electricity-load forecasting, drifting classification, bandits, and segmentation tasks, the new combination is reported to match or beat standard baselines while using only one retained hypothesis. If the claims hold, researchers get a common vocabulary for comparing, reusing, and extending methods whose connections were previously scattered across subfields.

What carries the argument

The load-bearing object is the five-slot predictive decomposition given in equation (6) of the paper: measurement model $h$ with loss $\ell$, auxiliary variable $\psi_t$, conditional prior $\pi$, parameter posterior $q$, and weighting function $\nu$. The decomposition separates 'how the world changes' from 'how much to trust each hypothesis,' so swapping one slot yields a different algorithm while the update skeleton stays fixed. The paper's signature mechanism is the thresholded hybrid prior of equations (19)-(20): a convex Ornstein-Uhlenbeck combination of the previous belief and the initial prior when the posterior continuation probability is high, and a hard reset to the initial prior when that probability falls below $\epsilon$.

What would settle it

Run RL[1]-OUPR* on a synthetic stream whose changepoints are known and whose segments contain slow drift, sweeping $\epsilon$ from 0 to 1. If no intermediate $\epsilon$ beats both the always-Ornstein-Uhlenbeck and always-reset extremes, or if the best $\epsilon$ shifts erratically across runs of the same task family, the claim that one threshold handles both gradual and abrupt change fails in that setting.

Watch

Extended reading notes

Core claim

The central discovery is a decomposition of one-step-ahead prediction in a hierarchical state-space model into five named components. For parameters $\theta_t$ and an auxiliary variable $\psi_t$ that encodes non-stationarity, the predictive expectation factorizes as $\hat{y}_{t+1}=\sum_{\psi_t}\nu_t(\psi_t)\int h(\theta_t,x_{t+1})q(\theta_t;\psi_t,D_{1:t})\,d\theta_t$, where $h$ is the measurement model, $\nu_t$ weights hypotheses about the auxiliary variable, and $q$ is the parameter posterior. The paper argues that every known probabilistic online method is a specific choice of these slots plus the conditional prior $\pi(\theta_t;\psi_t,D_{1:t-1})$, and it catalogs published filtering, segmentation, continual-learning, and bandit methods in this taxonomy. The proposed instance, RL[1]-OUPR*, uses the runlength as the auxiliary variable, keeps only the single most likely runlength hypothesis, and defines the conditional prior as an Ornstein-Uhlenbeck pull back toward the initial prior when the continuation probability $\nu_t(r^{(1)}_t)$ exceeds a threshold $\epsilon$, and as a hard reset to the initial prior otherwise. This thresholded hybrid is the paper's answer to environments that change both gradually and abruptly.

Load-bearing premise

The new method works only if the fixed threshold $\epsilon$ and the single retained runlength hypothesis can be set so that continuation-versus-reset decisions are right in both gradual and abrupt phases, and the paper gives no principled rule for choosing $\epsilon$.

Editorial extensions

If this is right

  • A Kalman filter and a changepoint detector can be combined or compared by swapping one slot, so methods developed for filtering become usable for forecasting and vice versa.
  • The new single-hypothesis method RL[1]-OUPR* gives a low-memory option that, in the paper's reported experiments, performs competitively on data with both gradual and abrupt changes.
  • Outlier-robust likelihoods can be dropped into the framework without changing the rest of the algorithm, as demonstrated by the WoLF+RL[inf]-PR variant.
  • The framework exposes a memory-accuracy tradeoff: exact weighting over a growing auxiliary space is expensive, while DA[K] with K retained hypotheses bounds the cost at the price of pruning the hypothesis set.
  • Because the auxiliary variable is a free design choice, a method built for segmentation with runlengths can be transplanted to bandits or continual learning, which the paper demonstrates experimentally.

Reading between the lines

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

  • Beyond the paper, the five-slot decomposition suggests a search space: one could treat the slots as hyperparameters and systematically search over auxiliary-variable types, priors, and thresholds on a fixed benchmark suite.
  • The threshold $\epsilon$ in the new method is the paper's most brittle hand-set hyperparameter, and nothing in the paper prevents learning it online with a second auxiliary variable, though the authors do not pursue this.
  • The framework implies a clean ablation discipline: to isolate the benefit of modelling gradual change, one compares methods that differ only in the prior slot, which the experiments begin to do.
  • The reported results hint that the right auxiliary-variable complexity is task-dependent, since RL[K]-PR overtakes the single-hypothesis method on some tasks when more hypotheses are kept; identifying when one hypothesis suffices is a natural next question.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes BONE, a five-component framework for online probabilistic learning in non-stationary environments, with modelling choices (M.1 measurement model, M.2 auxiliary process, M.3 conditional prior) and algorithmic choices (A.1 posterior approximation, A.2 weighting for the auxiliary variable). It shows how many existing methods from filtering, changepoint detection, continual learning, and bandits can be expressed as BONE instances (Table 3), and introduces a new method, RL[1]-OUPR*, which combines a single runlength hypothesis with an Ornstein-Uhlenbeck drift and a threshold-based reset. The framework is evaluated on electricity load forecasting, synthetic classification with drift and jumps, Bernoulli bandits, piecewise autoregression, and heavy-tailed regression.

Significance. The modular taxonomy is a genuinely useful organizing contribution: Table 3 convincingly maps a broad literature into a small number of components, and the accompanying JAX library is a concrete, reproducible artifact. The recursions in Section 2 are correct, and the generalization from Bayesian posteriors to loss-based updates and ad-hoc weights is coherent and clearly explained. If the proposed method were a clean instance of the framework and the empirical claims were supported by repeated trials, the paper would be a valuable reference for both practitioners and researchers. The main reservations concern the internal consistency of RL[1]-OUPR* with the formal BONE definition, and the strength of the stochastic classification evidence.

major comments (3)
  1. [Section 2.5, Eqs. (19)-(20); Algorithm 4; Table 1] The M.3 prior is formally defined as π_t(θ_t; ψ_t) := π(θ_t; ψ_t, D_{1:t-1}) in Table 1, i.e., it may depend only on data before the current observation. However, Eqs. (19)-(20) and Algorithm 4 lines 2-16 construct the prior using ν_t(r_t) = p(r_t | D_{1:t}), which is computed from the current observation y_t through the predictive likelihoods in Algorithm 4 lines 2-4. Algorithm 4 then multiplies this prior by the same y_t as a likelihood in line 16. Thus y_t is used twice: once to set the runlength/mixing weight and once as the observation in the parameter update. This makes RL[1]-OUPR* inconsistent with the BONE definition in Table 1 and with the generalized posterior update in Eq. (7). Please either revise the definition of M.3 (and the update in Eq. (7)) to permit priors that depend on D_{1:t}, or present RL[1]-OUPR* as a heuristic extension of BONE rather than an instance of it; if the latter, the abstract's claim that BONE 'allows us to propose new methods' needs qualification.
  2. [Sections 4.1.2 and 4.1.3, Figures 7 and 9] The classification experiments use a stochastic data-generating process (Bernoulli observations in both cases, and random parameter jumps in Section 4.1.3), yet the results are reported as single-run misclassification rates without repeated trials or error bars. The statements that RL[1]-OUPR* 'works the best' and 'significantly outperforms' RL[1]-PR are therefore not statistically supported. Please add multiple seeds with means and confidence intervals, and report how hyperparameters, including the threshold ε, were selected for each method.
  3. [Section 2.5, Eqs. (19)-(20); Sections 4.1-4.3] The threshold ε is a new free parameter of RL[1]-OUPR*, but the paper provides no principled rule for setting it beyond the limits ε=1 (always reset, no learning) and ε=0 (OU-type update). Since a misspecified ε collapses the method to either never adapting or always resetting, the experiments should state the ε values used and include a sensitivity analysis. This is necessary to assess whether the reported advantages of RL[1]-OUPR* over one-pass baselines are robust or are driven by a favorable threshold choice.
minor comments (6)
  1. [Eqs. (19)-(20)] The notation µ(r_t) and Σ(r_t) in the ν_t(r_t) > ε branch is inconsistent with Eq. (18) and with Algorithm 4 line 8, where the previous posterior (µ_{t-1}, Σ_{t-1}) is used; please align the notation.
  2. [Section 2.2] The heading 'Details of BONE' is followed by no text; either fill in the section or remove the heading.
  3. [Figure 3 caption] The caption says 'rolling relative absolute error' while the text in Section 4.1.1 says 'rolling mean absolute error (MAE)'; these should be made consistent.
  4. [Section 4.1.1] The phrase 'between March 4 2020 and March March 8 2020' contains a duplicated 'March'; please fix the typo.
  5. [Table 2] The cardinality for CPT appears as '2 t' and is presumably 2^{t+1}; the value notation '2{0,1,...,t}' also needs an explicit explanation.
  6. [Algorithm 4, line 10] Given the preceding 'if ν(r_t^{(1)}) > ε' branch, the condition 'else if ν(r_t^{(1)}) ≤ ε' is equivalent to 'else'; this is not an error, but the asymmetry may confuse readers.

Circularity Check

1 steps flagged · score 6.0 of 10

RL[1]-OUPR*'s M.3 prior is built from ν_t=p(r_t|D_{1:t}) computed with the current y_t, then Algorithm 4 conditions on the same y_t as the likelihood, making the new method not a BONE instance by the paper's own definition.

  1. self definitional [Section 2.5, Eqs. (19)-(20); Eq. (37); Algorithm 4; Table 1]
    "gt(rt,D1:t−1) = { µ0 (1−νt(rt)) +µ(rt)νt(rt) νt(rt)>ε, µ0 νt(rt)≤ε, ... Here νt(rt) =p(rt|D 1:t), with rt =rt−1 + 1, is the probability we are continuing a segment ... For details on how to compute νt(rt), see(37)."

    Table 1 defines the M.3 prior as π_t(θ_t;ψ_t):=π(θ_t;ψ_t,D_{1:t-1}), i.e., using only data before y_t. Equations (19)-(20) instead set the prior mean/covariance from ν_t(r_t)=p(r_t|D_{1:t}), and Eq. (37) (Algorithm 4 lines 2-4) computes this ν_t from the predictive likelihood p(y_t|r_t,x_t,D_{1:t-1}) of the current observation. Algorithm 4 then reuses the same y_t as the likelihood in line 16, q_t∝N(θ_t|bar-μ_t,bar-Σ_t)p(y_t|θ^T x_t,R_t). Hence y_t enters twice: once to construct the prior and once as the likelihood. The new method is therefore not an instance of BONE as defined, and any advantage over one-pass baselines may reflect this double-counting rather than the framework.

full rationale

The BONE taxonomy itself is a re-description of existing recursions (Eqs. 2-7) and does not derive new predictive results from fitted quantities; existing-method classifications are supported by the cited external papers, and the experimental comparisons tune hyperparameters on a warmup period and evaluate on held-out data. The only load-bearing circularity I can exhibit is the self-definitional construction of RL[1]-OUPR*: its M.3 prior depends on p(r_t|D_{1:t}) from Eq. (37), which is a function of y_t, while Table 1 restricts M.3 to D_{1:t-1}; Algorithm 4 then updates with the same y_t as likelihood. This invalidates the claim that RL[1]-OUPR* is an instance of BONE, though it does not reduce the whole framework to a fit. Score 6 reflects that the central new-method demonstration is constructed from the very observation it then treats as evidence.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The framework uses standard Bayesian recursions plus two modeling assumptions (factorization and known hazard) and one ad hoc threshold in the new method. No free parameters are needed for the unification claim itself; the new method's empirical performance depends on kappa (tuned) and epsilon (fixed).

free parameters (2)
  • kappa = tuned per experiment (values not reported)
    Prior probability of a changepoint (hazard rate) in Eqs. (30) and (36); tuned on the warmup period for RL-based methods in Sections 4.1.1 and 4.1.2.
  • epsilon = 0.5 (default, not tuned)
    Reset threshold in Eqs. (19)-(20); controls whether RL[1]-OUPR* does a soft OU update or a hard reset; set to 0.5 in experiments without sensitivity analysis.
assumptions (4)
  • standard math Standard probability calculus: Bayes rule and marginalization
    Equations (1)-(5) and (28)-(29) rely on standard probability identities.
  • domain assumption Factorized joint density over theta_t and psi_t
    The joint model in Eq. (1) assumes observations are conditionally independent given theta_t, and that non-stationarity is captured separately by psi_t; methods that do not decouple these (e.g., de Vilmarest and Wintenberger 2021) are excluded by design.
  • domain assumption Known hazard function for runlength transitions
    Equation (30) assumes p(rt|rt-1) is known via a hazard function H(r); experiments use constant kappa, encoding prior knowledge of change frequency.
  • ad hoc to paper Threshold epsilon for reset decisions
    Equations (19)-(20) introduce epsilon to choose between soft and hard resets; no principled selection method is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A unifying framework for generalised Bayesian online learning in non-stationary environments." pith.science (2026). https://pith.science/paper/6QKS2YPX

@misc{pith2026241110153,
  author       = {Pith},
  title        = {Pith review of: A unifying framework for generalised Bayesian online learning in non-stationary environments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6QKS2YPX}},
  note         = {Machine review of arXiv:2411.10153}
}
read the original abstract

We propose a unifying framework for methods that perform probabilistic online learning in non-stationary environments. We call the framework BONE, which stands for generalised (B)ayesian (O)nline learning in (N)on-stationary (E)nvironments. BONE provides a common structure to tackle a variety of problems, including online continual learning, prequential forecasting, and contextual bandits. The framework requires specifying three modelling choices: (i) a model for measurements (e.g., a neural network), (ii) an auxiliary process to model non-stationarity (e.g., the time since the last changepoint), and (iii) a conditional prior over model parameters (e.g., a multivariate Gaussian). The framework also requires two algorithmic choices, which we use to carry out approximate inference under this framework: (i) an algorithm to estimate beliefs (posterior distribution) about the model parameters given the auxiliary variable, and (ii) an algorithm to estimate beliefs about the auxiliary variable. We show how the modularity of our framework allows for many existing methods to be reinterpreted as instances of BONE, and it allows us to propose new methods. We compare experimentally existing methods with our proposed new method on several datasets, providing insights into the situations that make each method more suitable for a specific task. We provide a Jax open source library to facilitate the adoption of this framework.

Figures

Figures reproduced from arXiv: 2411.10153 by the authors.

Figure 1
Figure 1. Overview of BONE methods grouped by the qualitative nature of the auxiliary variable and the [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Two-levelled hierarchical state-space model (SSM) with known dynamics, motivating our BONE [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The top panel shows the target variable (electricity consumption) from March 1 2020 to March 12 2020. The bottom panel shows the twelve-hour rolling relative absolute error of predictions for the same time window. The dotted black line corresponds to March 7 2020, when Covid lockdown began. Next, [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: One day ahead electricity forecasting results for Figure [PITH_FULL_IMAGE:figures/full_fig_p019_4.png]
Figure 5
Figure 5. Figure 5: One day ahead electricity forecasting results for [PITH_FULL_IMAGE:figures/full_fig_p019_5.png]
Figure 6
Figure 6. Figure 6: Distribution of the 5-day mean absolute error (MAE) for each of the competing methods on [PITH_FULL_IMAGE:figures/full_fig_p020_6.png]
Figure 7
Figure 7. Figure 7: Misclassification rate of various methods on the online classification with periodic drift task. [PITH_FULL_IMAGE:figures/full_fig_p020_7.png]
Figure 8
Figure 8. Figure 8: Accuracy of predictions for RL[1]-PR as a function of the number of hypothesis and the prior probability of a changepoint κ. The black dotted line is the performance of RL[1]-OUPR* reported in [PITH_FULL_IMAGE:figures/full_fig_p021_8.png]
Figure 9
Figure 9. Figure 9: Misclassification rate of various methods on the online classification with drift and jumps task. [PITH_FULL_IMAGE:figures/full_fig_p021_9.png]
Figure 10
Figure 10. Figure 10: Accuracy of predictions for RL[K]-PR as a function of the number of hypotheses (K) and the probability of a changepoint κ. The black dotted line is the performance of RL[1]-OUPR* reported in [PITH_FULL_IMAGE:figures/full_fig_p022_10.png]
Figure 11
Figure 11. Figure 11: Regret of competing methods on the contextual bandits task. Confidence bands are computed [PITH_FULL_IMAGE:figures/full_fig_p022_11.png]
Figure 12
Figure 12. Figure 12: The left panel shows a sample run of the piecewise polynomial regression with dependence across segments. The x-axis is for the features, the (left) y-axis is for measurements together with the estimations made by RL[1]-PR, RL-MMPR, and RL[1]-OUPR*, the (right) y-axis…
Figure 13
Figure 13. Figure 13: Count of changepoints over an experiment for 100 trials. The orange line shows the true number [PITH_FULL_IMAGE:figures/full_fig_p024_13.png]
Figure 14
Figure 14. Figure 14: Sample run of the heavy-tailed-regression process. Each box corresponds to the samples within a [PITH_FULL_IMAGE:figures/full_fig_p024_14.png]
Figure 15
Figure 15. Figure 15: The left panel shows the rolling RMSE using a window of the 10 previous observations. The right panel shows the distribution of final RMSE over 30 runs. The vertical dotted line denotes a change in the true model parameters. The left panel of [PITH_FULL_IMAGE:figures…
Figure 16
Figure 16. Figure 16: Segmentation of the non-stationary linear regression problem. The left panel shows the [PITH_FULL_IMAGE:figures/full_fig_p025_16.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Online Learning of Modular Bayesian Deep Receivers: Single-Step Adaptation with Streaming Data

    eess.SP 2025-11 conditional novelty 5.0 of 10

    A Kalman-filter-style single-step update, applied module-by-module in a deep neural receiver, can track changing wireless channels with lower latency and error than multi-epoch stochastic gradient descent.

Reference graph

Works this paper leans on

79 extracted references · 63 canonical work pages · cited by 1 Pith paper

  1. [1]

    Adaptive time series forecasting with markovian variance switching, 2024

    Baptiste Abélès, Joseph de Vilmarest, and Olivier Wintemberger. Adaptive time series forecasting with markovian variance switching, 2024

    work page 2024

  2. [2]

    Ryan Prescott Adams and David J. C. MacKay. Bayesian online changepoint detection, 2007

    2007

  3. [3]

    Bayesian online prediction of change points

    Diego Agudelo-Espa \ n a, Sebastian Gomez-Gonzalez, Stefan Bauer, Bernhard Sch \"o lkopf, and Jan Peters. Bayesian online prediction of change points. In Conference on Uncertainty in Artificial Intelligence, pp.\ 320--329. PMLR, 2020

    work page 2020

  4. [4]

    Bayesian change-point detection for bandit feedback in non-stationary environments

    Reda Alami. Bayesian change-point detection for bandit feedback in non-stationary environments. In Asian Conference on Machine Learning, pp.\ 17--31. PMLR, 2023

    work page 2023

  5. [5]

    Restarted bayesian online change-point detector achieves optimal detection delay

    R \'e da Alami, Odalric Maillard, and Raphael F \'e raud. Restarted bayesian online change-point detector achieves optimal detection delay. In International conference on machine learning, pp.\ 211--221. PMLR, 2020

    work page 2020

  6. [6]

    Robust and scalable bayesian online changepoint detection, 2023

    Matias Altamirano, François-Xavier Briol, and Jeremias Knoblauch. Robust and scalable bayesian online changepoint detection, 2023

    work page 2023

  7. [7]

    A survey of methods for time series change point detection

    Samaneh Aminikhanghahi and Diane J Cook. A survey of methods for time series change point detection. Knowledge and information systems, 51 0 (2): 0 339--367, 2017

    work page 2017

  8. [8]

    On warm-starting neural network training

    Jordan T Ash and Ryan P Adams. On warm-starting neural network training. In NIPS, 2020. URL http://arxiv.org/abs/1910.08475

    arXiv 2020

Show all 79 references
  1. [9]

    Product partition models for change point problems

    Daniel Barry and John A Hartigan. Product partition models for change point problems. The Annals of Statistics, pp.\ 260--279, 1992

    1992

  2. [10]

    Detection of abrupt changes: theory and application, volume 104

    Michele Basseville, Igor V Nikiforov, et al. Detection of abrupt changes: theory and application, volume 104. Prentice hall Englewood Cliffs, 1993

    1993

  3. [11]

    Space guidance evolution-a personal narrative

    Richard H Battin. Space guidance evolution-a personal narrative. Journal of Guidance, Control, and Dynamics, 5 0 (2): 0 97--110, 1982

    1982

  4. [12]

    The infinite hidden markov model

    Matthew Beal, Zoubin Ghahramani, and Carl Rasmussen. The infinite hidden markov model. Advances in neural information processing systems, 14, 2001

    2001

  5. [13]

    Bencomo, Jake C

    Gianluca M. Bencomo, Jake C. Snell, and Thomas L. Griffiths. Implicit maximum a posteriori filtering via adaptive optimization, 2023

    2023

  6. [14]

    Bernardo and A

    J. Bernardo and A. Smith. Bayesian Theory. John Wiley, 1994

    1994

  7. [15]

    Weight uncertainty in neural network

    Charles Blundell, Julien Cornebise, Koray Kavukcuoglu, and Daan Wierstra. Weight uncertainty in neural network. In Francis Bach and David Blei (eds.), Proceedings of the 32nd International Conference on Machine Learning, volume 37 of Proceedings of Machine Learning Research, p...

    2015

  8. [16]

    JAX : composable transformations of P ython+ N um P y programs, 2018

    James Bradbury, Roy Frostig, Peter Hawkins, Matthew James Johnson, Chris Leary, Dougal Maclaurin, George Necula, Adam Paszke, Jake Vander P las, Skye Wanderman- M ilne, and Qiao Zhang. JAX : composable transformations of P ython+ N um P y programs, 2018. URL http://github.com/...

    2018

  9. [17]

    Online continual learning with natural distribution shifts: An empirical study with visual data

    Zhipeng Cai, Ozan Sener, and Vladlen Koltun. Online continual learning with natural distribution shifts: An empirical study with visual data. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 8281--8290, 2021

    2021

  10. [18]

    Nonlinear bayesian filtering with natural gradient gaussian approximation

    Wenhan Cao, Tianyi Zhang, Zeju Sun, Chang Liu, Stephen S-T Yau, and Shengbo Eben Li. Nonlinear bayesian filtering with natural gradient gaussian approximation. arXiv [eess.SY], October 2024. URL http://arxiv.org/abs/2410.15832

    2024

  11. [19]

    Bandits for algorithmic trading with signals

    \'A lvaro Cartea, Fay c al Drissi, and Pierre Osselin. Bandits for algorithmic trading with signals. Available at SSRN 4484004, 2023 a

    2023

  12. [20]

    Detecting toxic flow, 2023 b

    Álvaro Cartea, Gerardo Duran-Martin, and Leandro Sánchez-Betancourt. Detecting toxic flow, 2023 b

    2023

  13. [21]

    A mixture-of-experts framework for adaptive kalman filtering

    Wassim S Chaer, Robert H Bishop, and Joydeep Ghosh. A mixture-of-experts framework for adaptive kalman filtering. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), 27 0 (3): 0 452--464, 1997

    1997

  14. [22]

    State estimation for discrete systems with switching parameters

    Chaw-Bing Chang and Michael Athans. State estimation for discrete systems with switching parameters. IEEE Transactions on Aerospace and Electronic Systems, AES-14 0 (3): 0 418--425, 1978

    1978

  15. [23]

    On diagonal approximations to the extended kalman filter for online training of bayesian neural networks

    Peter G Chang, Kevin Patrick Murphy, and Matt Jones. On diagonal approximations to the extended kalman filter for online training of bayesian neural networks. In Continual Lifelong Learning Workshop at ACML 2022, 2022

    2022

  16. [24]

    Low-rank extended kalman filtering for online learning of neural networks from streaming data

    Peter G Chang, Gerardo Duran-Martin, Alex Shestopaloff, Matt Jones, and Kevin Patrick Murphy. Low-rank extended kalman filtering for online learning of neural networks from streaming data. In Conference on Lifelong Learning Agents, pp.\ 1025--1071. PMLR, 2023

    2023

  17. [25]

    Gee, and Arnaud Doucet

    Joao FG de Freitas, Mahesan Niranjan, Andrew H. Gee, and Arnaud Doucet. Sequential monte carlo methods to train neural network models. Neural computation, 12 0 (4): 0 955--993, 2000

    2000

  18. [26]

    Loss of plasticity in deep continual learning

    Shibhansh Dohare, J Fernando Hernandez-Garcia, Qingfeng Lan, Parash Rahman, A Rupam Mahmood, and Richard S Sutton. Loss of plasticity in deep continual learning. Nature, 632 0 (8026): 0 768--774, 2024

    2024

  19. [27]

    Efficient online bayesian inference for neural bandits

    Gerardo Duran-Martin, Aleyna Kara, and Kevin Murphy. Efficient online bayesian inference for neural bandits. In International Conference on Artificial Intelligence and Statistics, pp.\ 6002--6021. PMLR, 2022

    2022

  20. [28]

    Shestpaloff, Leandro S \'a nchez-Betancourt, Jeremias Knoblauch, Matt Jones, Briol Fran c ois-Xavier, and Kevin P

    Gerardo Duran-Martin, Matias Altamirano, Alexander Y. Shestpaloff, Leandro S \'a nchez-Betancourt, Jeremias Knoblauch, Matt Jones, Briol Fran c ois-Xavier, and Kevin P. Murphy. Outlier-robust kalman filtering through generalised bayes. In International Conference on Machine Le...

    2024

  21. [29]

    Sequential data assimilation with a nonlinear quasi-geostrophic model using monte carlo methods to forecast error statistics

    Geir Evensen. Sequential data assimilation with a nonlinear quasi-geostrophic model using monte carlo methods to forecast error statistics. Journal of Geophysical Research: Oceans, 99 0 (C5): 0 10143--10162, 1994

    1994

  22. [30]

    Econometric policy model construction: the post-bayesian approach

    Arnold M Faden and Gordon C Rausser. Econometric policy model construction: the post-bayesian approach. In Annals of Economic and Social Measurement, Volume 5, number 3, pp.\ 349--363. NBER, 1976

    1976

  23. [31]

    Day-ahead electricity demand forecasting competition: Post-covid paradigm

    Mostafa Farrokhabadi, Jethro Browell, Yi Wang, Stephen Makonin, Wencong Su, and Hamidreza Zareipour. Day-ahead electricity demand forecasting competition: Post-covid paradigm. IEEE Open Access Journal of Power and Energy, 9: 0 185--191, 2022. doi:10.1109/OAJPE.2022.3161101

    2022

  24. [32]

    On-line inference for multiple changepoint problems

    Paul Fearnhead and Zhen Liu. On-line inference for multiple changepoint problems. Journal of the Royal Statistical Society Series B: Statistical Methodology, 69 0 (4): 0 589--605, 2007

    2007

  25. [33]

    Efficient bayesian analysis of multiple changepoint models with dependence across segments

    Paul Fearnhead and Zhen Liu. Efficient bayesian analysis of multiple changepoint models with dependence across segments. Statistics and Computing, 21: 0 217--229, 2011

    2011

  26. [34]

    Changepoint detection in the presence of outliers

    Paul Fearnhead and Guillem Rigaill. Changepoint detection in the presence of outliers. Journal of the American Statistical Association, 114 0 (525): 0 169--183, 2019

    2019

  27. [35]

    The sticky hdp-hmm: Bayesian nonparametric hidden markov models with persistent states

    Emily B Fox, Erik B Sudderth, Michael I Jordan, and Alan S Willsky. The sticky hdp-hmm: Bayesian nonparametric hidden markov models with persistent states. Arxiv preprint, 2, 2007

    2007

  28. [36]

    Non-stationary learning of neural networks with automatic soft parameter reset

    Alexandre Galashov, Michalis K Titsias, András György, Clare Lyle, Razvan Pascanu, Teh Yee Whye, and Maneesh Sahani. Non-stationary learning of neural networks with automatic soft parameter reset. In NIPS, November 2024. URL https://arxiv.org/abs/2411.04034

    2024 arXiv

  29. [37]

    Knowledge discovery from data streams

    Joao Gama, Jesus Aguilar-Ruiz, and Ralf Klinkenberg. Knowledge discovery from data streams. Intelligent Data Analysis, 12 0 (3): 0 251--252, 2008

    2008

  30. [38]

    Variational learning for switching state-space models

    Zoubin Ghahramani and Geoffrey E Hinton. Variational learning for switching state-space models. Neural computation, 12 0 (4): 0 831--864, 2000

    2000

  31. [39]

    Comprehensive analysis of change-point dynamics detection in time series data: A review

    Muktesh Gupta, Rajesh Wadhvani, and Akhtar Rasool. Comprehensive analysis of change-point dynamics detection in time series data: A review. Expert Systems with Applications, pp.\ 123342, 2024

    2024

  32. [40]

    Kalman filtering and neural networks

    Simon Haykin. Kalman filtering and neural networks. John Wiley & Sons, 2004

    2004

  33. [41]

    Gradual changes versus abrupt changes

    M Hu s kov \'a . Gradual changes versus abrupt changes. Journal of Statistical Planning and Inference, 76 0 (1-2): 0 109--125, 1999

    1999

  34. [42]

    Improving predictions of bayesian neural nets via local linearization

    Alexander Immer, Maciej Korzepa, and Matthias Bauer. Improving predictions of bayesian neural nets via local linearization. In International conference on artificial intelligence and statistics, pp.\ 703--711. PMLR, 2021

    2021

  35. [43]

    Bayesian online natural gradient ( BONG )

    Matt Jones, Peter Chang, and Kevin Murphy. Bayesian online natural gradient ( BONG ). In Advances in Neural Information Processing Systems, May 2024. URL http://arxiv.org/abs/2405.19681

    2024 arXiv

  36. [44]

    A new approach to linear filtering and prediction problems

    Rudolph Emil Kalman. A new approach to linear filtering and prediction problems. Transactions of the ASME--Journal of Basic Engineering, 82 0 (Series D): 0 35--45, 1960

    1960

  37. [45]

    Spatio-temporal bayesian on-line changepoint detection with model selection

    Jeremias Knoblauch and Theodoros Damoulas. Spatio-temporal bayesian on-line changepoint detection with model selection. In International Conference on Machine Learning, pp.\ 2718--2727. PMLR, 2018

    2018

  38. [46]

    Doubly robust bayesian inference for non-stationary streaming data with -divergences

    Jeremias Knoblauch, Jack E Jewson, and Theodoros Damoulas. Doubly robust bayesian inference for non-stationary streaming data with -divergences. In S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett (eds.), Advances in Neural Information Processi...

    2018

  39. [47]

    An optimization-centric view on bayes' rule: Reviewing and generalizing variational inference

    Jeremias Knoblauch, Jack Jewson, and Theodoros Damoulas. An optimization-centric view on bayes' rule: Reviewing and generalizing variational inference. Journal of Machine Learning Research, 23 0 (132): 0 1--109, 2022. URL http://jmlr.org/papers/v23/19-1047.html

    2022

  40. [48]

    Ridge regression signal processing

    Mark R Kuhl. Ridge regression signal processing. NASA, Langley Research Center, Joint University Program for Air Transportation Research, 1989-1990, 1990

    1989

  41. [49]

    Continual learning with bayesian neural networks for non-stationary data

    Richard Kurle, Botond Cseke, Alexej Klushyn, Patrick Van Der Smagt, and Stephan G \"u nnemann. Continual learning with bayesian neural networks for non-stationary data. In International Conference on Learning Representations, 2019

    2019

  42. [50]

    The recursive variational gaussian approximation (r-vga)

    Marc Lambert, Silv \`e re Bonnabel, and Francis Bach. The recursive variational gaussian approximation (r-vga). Statistics and Computing, 32 0 (1): 0 10, 2022

    2022

  43. [51]

    The limited-memory recursive variational gaussian approximation (l-rvga)

    Marc Lambert, Silv \`e re Bonnabel, and Francis Bach. The limited-memory recursive variational gaussian approximation (l-rvga). Statistics and Computing, 33 0 (3): 0 70, 2023

    2023

  44. [52]

    Detecting and adapting to irregular distribution shifts in bayesian online learning, 2021

    Aodong Li, Alex Boyd, Padhraic Smyth, and Stephan Mandt. Detecting and adapting to irregular distribution shifts in bayesian online learning, 2021

    2021

  45. [53]

    A contextual-bandit approach to personalized news article recommendation

    Lihong Li, Wei Chu, John Langford, and Robert E Schapire. A contextual-bandit approach to personalized news article recommendation. In Proceedings of the 19th international conference on World wide web, pp.\ 661--670, 2010

    2010

  46. [54]

    Bayesian learning and inference in recurrent switching linear dynamical systems

    Scott Linderman, Matthew Johnson, Andrew Miller, Ryan Adams, David Blei, and Liam Paninski. Bayesian learning and inference in recurrent switching linear dynamical systems. In Artificial intelligence and statistics, pp.\ 914--922. PMLR, 2017

    2017

  47. [55]

    Robust sequential online prediction with dynamic ensemble of multiple models: A review

    Bin Liu. Robust sequential online prediction with dynamic ensemble of multiple models: A review. Neurocomputing, pp.\ 126553, 2023

    2023

  48. [56]

    Nonstationary bandit learning via predictive sampling

    Yueyang Liu, Benjamin Van Roy, and Kuang Xu. Nonstationary bandit learning via predictive sampling. In International Conference on Artificial Intelligence and Statistics, pp.\ 6215--6244. PMLR, 2023

    2023

  49. [57]

    Optimal adaptive estimation of sampled stochastic processes

    David Magill. Optimal adaptive estimation of sampled stochastic processes. IEEE Transactions on Automatic Control, 10 0 (4): 0 434--439, 1965

    1965

  50. [58]

    Thompson sampling in switching environments with bayesian online change point detection

    Joseph Mellor and Jonathan Shapiro. Thompson sampling in switching environments with bayesian online change point detection. arXiv preprint arXiv:1302.3721, 2013

    2013 arXiv

  51. [59]

    Slang: Fast structured covariance approximations for bayesian deep learning with natural gradient

    Aaron Mishkin, Frederik Kunstner, Didrik Nielsen, Mark Schmidt, and Mohammad Emtiyaz Khan. Slang: Fast structured covariance approximations for bayesian deep learning with natural gradient. Advances in neural information processing systems, 31, 2018

    2018

  52. [60]

    Bam: Bayes with adaptive memory

    Josue Nassar, Jennifer Brennan, Ben Evans, and Kendall Lowrey. Bam: Bayes with adaptive memory. arXiv preprint arXiv:2202.02405, 2022

    2022 arXiv

  53. [61]

    Variational continual learning

    Cuong V Nguyen, Yingzhen Li, Thang D Bui, and Richard E Turner. Variational continual learning. arXiv preprint arXiv:1710.10628, 2017

    2017 arXiv

  54. [62]

    Online natural gradient as a Kalman filter

    Yann Ollivier. Online natural gradient as a Kalman filter . Electronic Journal of Statistics, 12 0 (2): 0 2930 -- 2961, 2018. doi:10.1214/18-EJS1468. URL https://doi.org/10.1214/18-EJS1468

    2018 doi

  55. [63]

    From hmm's to segment models: A unified view of stochastic modeling for speech recognition

    Mari Ostendorf, Vassilios V Digalakis, and Owen A Kimball. From hmm's to segment models: A unified view of stochastic modeling for speech recognition. IEEE Transactions on speech and audio processing, 4 0 (5): 0 360--378, 1996

    1996

  56. [64]

    Bayesian approach to system identification

    V \'a clav Peterka. Bayesian approach to system identification. In Trends and Progress in System identification, pp.\ 239--304. Elsevier, 1981

    1981

  57. [65]

    Towards robust inference for bayesian filtering of linear gaussian dynamical systems subject to additive change

    Hans Reimann. Towards robust inference for bayesian filtering of linear gaussian dynamical systems subject to additive change. masterthesis, Universit \"a t Potsdam, 2024

    2024

  58. [66]

    Deep bayesian bandits showdown: An empirical comparison of bayesian deep networks for thompson sampling, 2018

    Carlos Riquelme, George Tucker, and Jasper Snoek. Deep bayesian bandits showdown: An empirical comparison of bayesian deep networks for thompson sampling, 2018

    2018

  59. [67]

    The Bayesian choice: from decision-theoretic foundations to computational implementation, volume 2

    Christian P Robert et al. The Bayesian choice: from decision-theoretic foundations to computational implementation, volume 2. Springer, 2007

    2007

  60. [68]

    The ensemble kalman filter: a signal processing perspective

    Michael Roth, Gustaf Hendeby, Carsten Fritsche, and Fredrik Gustafsson. The ensemble kalman filter: a signal processing perspective. EURASIP J. Adv. Signal Processing , 2017 0 (1): 0 56, 2017. URL https://doi.org/10.1186/s13634-017-0492-x

    2017 doi

  61. [69]

    Gaussian process change point models

    Yunus Saat c i, Ryan D Turner, and Carl E Rasmussen. Gaussian process change point models. In Proceedings of the 27th International Conference on Machine Learning (ICML-10), pp.\ 927--934, 2010

    2010

  62. [70]

    Bayesian filtering and smoothing, volume 17

    Simo S \"a rkk \"a and Lennart Svensson. Bayesian filtering and smoothing, volume 17. Cambridge university press, 2023

    2023

  63. [71]

    Test-time adaptation with state-space models

    Mona Schirmer, Dan Zhang, and Eric Nalisnick. Test-time adaptation with state-space models. arXiv preprint arXiv:2407.12492, 2024

    2024

  64. [72]

    Bayesian online change point detection with hilbert space approximate student-t process

    Jeremy Sellier and Petros Dellaportas. Bayesian online change point detection with hilbert space approximate student-t process. In International Conference on Machine Learning, pp.\ 30553--30569. PMLR, 2023

    2023

  65. [73]

    On the likelihood that one unknown probability exceeds another in view of the evidence of two samples

    William R Thompson. On the likelihood that one unknown probability exceeds another in view of the evidence of two samples. Biometrika, 25 0 (3-4): 0 285--294, 1933

    1933

  66. [74]

    Kalman filter for online classification of non-stationary data

    Michalis K Titsias, Alexandre Galashov, Amal Rannen-Triki, Razvan Pascanu, Yee Whye Teh, and Jorg Bornschein. Kalman filter for online classification of non-stationary data. In ICLR, 2024

    2024

  67. [75]

    An evaluation of change point detection algorithms

    Gerrit JJ Van den Burg and Christopher KI Williams. An evaluation of change point detection algorithms. arXiv preprint arXiv:2003.06222, 2020

    2003 arXiv

  68. [76]

    Beam sampling for the infinite hidden markov model

    Jurgen Van Gael, Yunus Saatci, Yee Whye Teh, and Zoubin Ghahramani. Beam sampling for the infinite hidden markov model. In Proceedings of the 25th international conference on Machine learning, pp.\ 1088--1095, 2008

    2008

  69. [77]

    Bayesian forecasting and dynamic models

    Mike West and Jeff Harrison. Bayesian forecasting and dynamic models. Springer, 1997

    1997

  70. [78]

    Bayesian online learning of the hazard rate in change-point problems

    Robert C Wilson, Matthew R Nassar, and Joshua I Gold. Bayesian online learning of the hazard rate in change-point problems. Neural computation, 22 0 (9): 0 2452--2476, 2010

    2010

  71. [79]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.