Pith. sign in

REVIEW 3 major objections 5 minor 24 references

Sequential dependence in LLM tasks widens reliability uncertainty, and ignoring it overstates confidence.

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

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

T0 review · deepseek-v4-flash

2026-08-01 04:00 UTC pith:MKBKBSKU

load-bearing objection A clean formal HMM extension of hierarchical LLM reliability, but the empirical case for temporal dependence is not yet made: the log-score differences are tiny and lack uncertainty quantification. the 3 major comments →

arxiv 2607.22951 v1 pith:MKBKBSKU submitted 2026-07-24 stat.ML cs.AIcs.LG

Modeling Memory-Dependent Reliability of LLMs: A Hidden Markov Model

Robab Aghazadeh Chakherlou , Siddartha Khastgir , Peter Popov , Xingyu Zhao This is my paper
classification stat.ML cs.AIcs.LG MSC 62M0562F15
keywords LLM reliabilityhidden Markov modeltemporal dependenceBayesian hierarchical modeluncertainty quantificationsequential interactionoperational profilefailure clustering
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper argues that LLM reliability estimates built on independent benchmark trials are overconfident when real use is sequential: context persists, errors propagate, and failures cluster. To capture this, it extends a hierarchical Bayesian reliability framework with a Hidden Markov Model in which an unobserved interaction state evolves over time and drives per-task success probabilities. The paper derives posterior distributions for reliability at subdomain, domain, and whole-model levels, and tests the model on two frontier LLMs across four datasets. Its central finding is that ignoring temporal dependence produces narrower credible intervals than the HMM—i.e., overly confident reliability claims—while the HMM matches the independent model when data are truly independent. The work matters because deployed LLM interactions are rarely a sequence of fresh, independent trials, so reliability claims from conventional benchmarks may not reflect sustained real-world performance.

Core claim

The paper claims that task outcomes in sequential LLM interactions are temporally dependent through an unobserved interaction state, and that reliability estimates that ignore this dependence are overconfident. It builds a hierarchical Bayesian Hidden Markov Model where a latent state evolves by a first-order Markov chain and each subdomain's success probability depends on that state; the posterior over parameters (Theorem 1) induces posterior distributions on n-task reliability at subdomain, domain, and LLM levels (Theorem 2). Empirically, on two frontier LLMs across four datasets, the HMM with K=3 latent states in one domain produces broader credible intervals than the independent-outcomes

What carries the argument

The central object is the domain-specific Hidden Markov Model with a latent interaction state Z_t evolving by a row-stochastic transition matrix A_i from a uniform initial distribution; each observed binary outcome is Bernoulli with success probability θ_ij,s that depends on both subdomain j and latent state s, and these θ values share a hierarchical Beta prior for partial pooling. The load-bearing recursion is R̄_i,s(n) = θ̄_i,s Σ_s' (A_i)_{s,s'} R̄_i,s'(n−1), which converts latent-state dynamics into the probability of n consecutive successes, and model selection uses out-of-sample predictive log scores to choose the number of latent states.

Load-bearing premise

The load-bearing premise is that binary task outcomes within a domain are generated by a time-homogeneous first-order Markov chain over a small fixed number (K=0 or 3) of latent states, with a uniform initial distribution and an operational profile that is independent of the latent state; if this model family is wrong, the temporal-dependence conclusions do not follow.

What would settle it

Reset the model's context before every task in the same benchmark sessions, so outcomes are genuinely independent, and rerun the HMM inference: if predictive log scores still favor K=3 and credible intervals still widen versus the i.i.d. model, then the dependence finding is an artifact of the HMM family rather than of real sequential interaction. Alternatively, fit a time-varying-transition HMM to the same 70/30 splits; if it clearly outperforms the time-homogeneous model, the stationarity assumption is the culprit.

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

If this is right

  • Reliability reported from independent benchmark trials is not a safe bound for sequential deployment: in the paper's experiments the HMM's credible intervals are wider than the i.i.d. model's, so claims made under independence can be overconfident.
  • Because the HMM reduces to the i.i.d. hierarchical model when K=1, practitioners can use predictive log scores to decide per domain whether temporal dependence actually matters for a given LLM and task mix.
  • n-step reliability decays with horizon, reaching posterior means around 0.13–0.15 at n=20 in the paper's experiments, so any 'sustained success' claim must state the horizon and account for state dynamics.
  • Task ordering changes the inferred dependence: shuffling the Reasoning-domain sequence changes credible-interval width, meaning session protocol design affects the uncertainty of reliability estimates.
  • The static operational profile captures task frequencies but not ordering; the paper explicitly notes this limitation, so ordering-sensitive usage profiles remain an open gap for deployment-representative claims.

Where Pith is reading between the lines

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

  • The inferred latent states are plausibly interpretable as persistent high- or low-performance interaction regimes; a natural extension is to correlate the inferred states with observable session features such as context length, topic, or accumulated error count, which the paper does not attempt.
  • The same latent-state machinery should transfer to other trajectory-based evaluations—agentic workflows, multi-turn tool use, or long-horizon coding sessions—where 'reliability' is naturally a function of the whole sequence, not a single fresh trial.
  • The paper's K-selection relies on a 70/30 chronological split; longer sessions or multiple sessions per domain would sharpen identifiability, and if the true process were non-stationary, a time-varying transition model would outperform the time-homogeneous HMM—a direct testable comparison.
  • The overconfidence claim rests on comparing posterior widths; a frequentist calibration check on the synthetic data—whether 95% intervals actually cover the true reliability 95% of the time—would quantify the real-world cost of ignoring dependence rather than just the difference in interval width.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper extends a hierarchical Bayesian framework for LLM reliability assessment by replacing the assumption of independent task outcomes with a domain-specific Hidden Markov Model (HMM). A latent interaction state evolves as a first-order Markov chain, and binary task outcomes are emitted conditionally on the state and subdomain. The authors derive posterior reliability distributions for subdomain, domain, and LLM levels (Theorem 2), select the number of latent states K by out-of-sample predictive log scores, and compare the HMM with a hierarchical i.i.d. model (HIP) on benchmark sessions built with Anthropic Claude and OpenAI models across four datasets. The paper claims that ignoring sequential dependence can lead to overconfident reliability estimates, and that the HMM better captures failure clustering.

Significance. If the central empirical claim were established, this would be a useful contribution: it identifies a realistic limitation of current i.i.d.-based hierarchical reliability models and provides a principled Bayesian extension with a verifiable recursion for n-step reliability. The theoretical derivation in Theorem 2 and Appendix A is sound under the stated assumptions, and the synthetic validation with a known data-generating process is a positive feature. The paper is also transparent about several limitations. However, the evidence for temporal dependence in real benchmark sessions is currently weak: K selection rests on single-split log-score differences without uncertainty quantification, and the paper's own independent-process simulation shows that the HMM's wider credible intervals can arise purely from added model flexibility. The headline claim is therefore plausible but not yet convincingly supported by the experiments.

major comments (3)
  1. [Sec. IV-B-a, Table III] The data-driven selection of K>1 is the load-bearing evidence for the central claim, but it is not supported with uncertainty quantification. The table reports only single 70/30 chronological-split log scores. For Claude Reasoning the K=1 vs K=3 difference is 0.212 nats and for OpenAI Coding it is 0.270 nats; the step from K=2 to K=3 is only 0.013 and 0.041 nats, respectively. Since each domain is one continuous session, the test segment is a single sequence and these differences are within the scale of a few binary outcomes. Without standard errors, repeated splits, block bootstrap, or a scoring-rule significance test, the selection of K>1 is not distinguishable from noise. Please report test-set lengths and provide uncertainty estimates for the log-score differences.
  2. [Sec. IV-B-f, Table V and Fig. 6] The wider HMM credible intervals are not themselves diagnostic of temporal dependence. The paper's own synthetic 'Independent HMM' row shows that a K=3 HMM fitted to i.i.d. data produces a CI width of 0.1457, versus 0.0257 for HIP, even though the data are independent. Thus the broader posteriors in Fig. 6 could simply reflect the extra flexibility of additional latent states rather than persistent sequential structure. The authors should provide a direct test that separates these explanations, such as a posterior predictive check on lag-1 autocorrelation or a comparison against a K=3 finite mixture model with no transition dynamics. The synthetic comparison also reports no uncertainty on the log scores.
  3. [Sec. IV-A(c) and Sec. III-A] The real-data analysis uses a single continuous session per domain, and no sequence lengths, effective sample sizes, or repeated sessions are reported. Identifiability of K and of the transition matrix from one observed sequence is questionable, especially when the session is constructed by concatenating benchmark tasks. With a single 70/30 split, temporal dependence is also confounded with non-stationarity (e.g., drift or a change in task mix). The paper should report the length of each session, assess posterior sensitivity to the particular session realization, and consider multiple sessions or a block-bootstrap procedure to support the claim that HMM structure is actually present.
minor comments (5)
  1. [Sec. I] Typo: 'In the LMM interaction setting' should be 'LLM interaction setting'.
  2. [References] Reference [6] has 'Reliabilty Engineering' in the title; should be 'Reliability Engineering'.
  3. [Sec. IV-B-h] The state-dependent success rates and self-transition probabilities are reported as point estimates without credible intervals or discussion of label switching. Since the states are only identifiable up to permutation, some comment on how label switching was handled in the MCMC would be useful.
  4. [Appendix A] Step 5 contains an incomplete sentence: 'denote the τ−th future task' should be 'Let x_τ denote the τ-th future task'.
  5. [Sec. IV-B-a] Table III should include the number of test observations and ideally a baseline column for the HIP model, so the reader can see the K=1 HMM and HIP are actually identical as claimed.

Circularity Check

0 steps flagged

No significant circularity: HMM posterior and reliability derivations are self-contained; only minor caveats are the transparent synthetic HMM DGP validation and a non-load-bearing self-citation.

full rationale

The paper derives the HMM posterior (Theorem 1) and reliability functionals (Theorem 2) from the model assumptions in Sec. III-A, with the recursion proof in Appendix A; these are direct consequences of the stated HMM rather than a renaming of inputs. The empirical comparison against HIP is a genuine baseline, not a fitted quantity presented as a prediction. The synthetic validation (Sec. IV-B(f)) intentionally generates data from an HMM, so the HMM's better log score and lower estimation error in that scenario are built in by design; however, the paper presents this as a controlled sanity check and explicitly conditions the conclusion on 'when the data-generating process contains persistent sequential structure,' so it is not a disguised circular derivation. The real-data selection of K=3 via out-of-sample predictive log scores (Table III) is a standard model-selection procedure; the absence of standard errors on the small log-score differences is a robustness/identifiability concern, not a formal circularity. Table V itself shows that a K=3 HMM on i.i.d. data still yields wider intervals (0.1457 vs 0.0257), which undercuts the interpretation in Sec. IV-B(d) that wider HMM intervals 'reflect sequential dependence' — but this is an inference-strength problem. The only self-citation to the authors' prior HIP paper [6] is used to motivate the independence assumption and as a baseline; the HMM derivation does not load on it. The paper also flags its own limitation that benchmark sessions 'should not be interpreted as direct evidence of real-world context accumulation' and that a complete proof is outsourced to GitHub; these are flagged concerns but not circularity.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 1 invented entities

The model has a moderate parameter count (theta_{ij,s}, transitions A_i, hyperparameters), but the load-bearing choices are the number of states K and the OP weights, both fitted/adjusted to the data, plus the assumption that a first-order time-homogeneous HMM is the right dependence structure. The latent state is the only invented entity and it has no independent evidence.

free parameters (4)
  • Number of latent states K = K=3 (Claude Reasoning, OpenAI Coding), K=1 otherwise, chosen by predictive log score
    K is selected per domain using the out-of-sample predictive log score rather than fixed by theory; the entire conclusion that sequential dependence matters depends on this selection. Table III.
  • Domain-level hyperparameters (mu_i, nu_i) priors = Beta(a_i,b_i) and Gamma(c_i,d_i), hyper-hyperparameters not specified in text
    These govern the hierarchical shrinkage of theta_{ij,s}; their values and the hyperpriors on a_i,b_i,c_i,d_i are not given, so part of the model is unstated.
  • Dirichlet transition priors lambda_{i,s,k} = Not specified numerically ('equal values yielding a symmetric uninformative prior')
    The strength of the prior on the transition matrix rows is not specified, which affects the posterior width and the reliability intervals.
  • Operational profile weights W_i and Omega_{ij} = Baseline: proportional to benchmark dataset sizes; sensitivity with uniform/skewed alternatives
    The OP is not measured from deployment but set by benchmark composition; the paper acknowledges this and tests sensitivity, but the central reliability numbers depend on these weights.
axioms (5)
  • ad hoc to paper Within each domain, binary task outcomes are generated by a time-homogeneous first-order Markov chain over K latent states (uniform initial distribution).
    Introduced in Sec. III-A (latent interaction state dynamics, Remark 4, Remark 5). This is the core modeling premise; no empirical check that a first-order time-homogeneous HMM fits the real benchmark sessions is provided.
  • domain assumption Subdomain labels j_t are drawn independently from the operational profile Omega_i and independently of the latent state Z_{i,t}.
    Stated in Theorem 2 and Remark 9. Needed for the domain-level recursion to match Definition 1; the paper admits a feedback loop between task selection and model performance is excluded.
  • domain assumption LLM-level reliability R_L is an OP-weighted sum of domain-level reliabilities rather than the joint reliability over cross-domain sequences.
    Remark 8 explicitly states this limitation; the paper does not model domain switching.
  • domain assumption The 70/30 chronological split and predictive log score correctly select K and detect temporal dependence in sequences of about 300–600 tasks.
    Used in Remark 2 and Table III; no statistical evidence that the split length is sufficient or that the predictive score has adequate power.
  • domain assumption The benchmark-constructed sessions (one prompt per task, fixed ordering within each domain) approximate sequential interaction sufficiently for the conclusions.
    Stated in Sec. IV-A and the Discussion; the authors acknowledge these are not naturally occurring interaction traces.
invented entities (1)
  • Latent interaction states Z_{i,t} no independent evidence
    purpose: Unobserved state that evolves as a Markov chain and modulates the success probability of every subdomain within a domain, capturing sequential dependence.
    The states are abstract statistical constructs without external interpretation; the paper provides only their learned success probabilities and self-transition probabilities. They are inferred from the same outcome sequences used to fit the model, and no external predictive handle (e.g., a falsifiable prediction about new subdomains) is provided.

pith-pipeline@v1.3.0-alltime-deepseek · 16801 in / 8511 out tokens · 61435 ms · 2026-08-01T04:00:49.414200+00:00 · methodology

0 comments
read the original abstract

Reliability assessment of large language models (LLMs) seeks to estimate the probability that a model produces correct responses under a specified operational profile. Conventional benchmark-based evaluation, often summarized by aggregate accuracy, provides a point estimate of performance but does not characterize the uncertainty associated with reliability claims. Currently, statistical inference methods for LLM reliability assessment are emerging. However, a key assumption underlying these models is that test outcomes can be treated as independent repeated trials. This assumption may be inappropriate in sequential settings, where later responses depend on earlier interactions through retained context, error propagation, or an evolving interaction state. We extend a hierarchical Bayesian framework for LLM reliability assessment by relaxing the assumption of independent task outcomes and introducing a Hidden Markov Model to capture sequential dependence in benchmark-constructed interaction sessions. In this formulation, outcomes are generated from a latent interaction state evolving according to a first-order Markov process, capturing changes in interaction context. Through experiments using Anthropic Claude and OpenAI on four datasets, we demonstrate the potential impact of sequential dependence on reliability assessment. The results suggest that ignoring sequential dependence may lead to overconfident reliability estimates.

Figures

Figures reproduced from arXiv: 2607.22951 by Peter Popov, Robab Aghazadeh Chakherlou, Siddartha Khastgir, Xingyu Zhao.

Figure 2
Figure 2. Figure 2: Hierarchical partial pooling in domain Di : subdomains share hyperparameters (µi , νi), so each is informed by its own data and others in the domain. While the hierarchical structure captures cross-subdomain dependence through partial pooling, it assumes that task out￾comes are independent over time. To capture temporal de￾pendence between successive tasks, we extend this framework using a HMM. In this for… view at source ↗
Figure 1
Figure 1. Figure 1: Hierarchical structure of the LLM evaluation. The [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Hidden Markov Model for domain Di. The interaction starts in a latent state Zi,1 ∼ δi (uniform). At each time t, the subdomain label jt is observed and the outcome Yi,t is generated with success probability θi,jt,Zi,t (θij,s with j = jt, s = Zi,t). The state then evolves to Zi,t+1 via transition matrix Ai. This process repeats, so outcomes depend on both the latent state and task type (initial state → obse… view at source ↗
Figure 6
Figure 6. Figure 6: Posterior reliability PDFs for Claude under HIP and [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of posterior reliability distributions at [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: HMM reliability decreases with n in both models. d) Comparison between HIP (independent outcomes) and HMM reliability distributions [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗

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

24 extracted references · 5 linked inside Pith

  1. [1]

    Pods revisited-a study of software failure behaviour,

    P. G. Bishop and F. Pullen, “Pods revisited-a study of software failure behaviour,” in[1988] The Eighteenth International Symposium on Fault- Tolerant Computing. Digest of Papers. IEEE, 1988, pp. 2–8

    1988

  2. [2]

    The variation of software survival time for different operational input profiles (or why you can wait a long time for a big bug to fail),

    P. G. Bishop, “The variation of software survival time for different operational input profiles (or why you can wait a long time for a big bug to fail),” inFTCS-23 The Twenty-Third International Symposium on Fault-Tolerant Computing. IEEE, 1993, pp. 98–107

    1993

  3. [3]

    On testing process control software for reliability assess- ment: the effects of correlation between successive failures,

    L. Strigini, “On testing process control software for reliability assess- ment: the effects of correlation between successive failures,”Software Testing, Verification and Reliability, vol. 6, no. 1, pp. 33–48, 1996

    1996

  4. [4]

    Adding error bars to evals: A statistical approach to language model evaluations,

    E. Miller, “Adding error bars to evals: A statistical approach to language model evaluations,”arXiv preprint arXiv:2411.00640, 2024

    Pith/arXiv arXiv 2024

  5. [5]

    Hibayes: A Hierarchical Bayesian Modeling Framework for AI Eval- uation Statistics,

    L. Luettgau, H. Coppock, M. Dubois, C. Summerfield, and C. Ududec, “Hibayes: A Hierarchical Bayesian Modeling Framework for AI Eval- uation Statistics,”arXiv preprint arXiv:2505.05602, 2025

    Pith/arXiv arXiv 2025

  6. [6]

    A hierarchical imprecise probability approach to reliabil- ity assessment of large language models,

    R. Aghazadeh-Chakherlou, Q. Guo, S. Khastgir, P. Popov, X. Zhang, and X. Zhao, “A hierarchical imprecise probability approach to reliabil- ity assessment of large language models,”Reliabilty Engineering and System Safety, 2026

    2026

  7. [7]

    Pre-trained large language models learn hidden markov models in-context,

    Y . Dai, Z. Gao, Y . Sattar, S. Dean, and J. J. Sun, “Pre-trained large language models learn hidden markov models in-context,”arXiv preprint arXiv:2506.07298, 2025

    Pith/arXiv arXiv 2025

  8. [8]

    Evaluating motivational interview quality using large language models and hidden markov models,

    K. Lim, Y .-C. Jung, and B.-H. Kim, “Evaluating motivational interview quality using large language models and hidden markov models,”BMC psychiatry, vol. 25, no. 1, p. 908, 2025

    2025

  9. [9]

    Regression models for nonstationary categorical time series: asymptotic estimation theory,

    H. Kaufmann, “Regression models for nonstationary categorical time series: asymptotic estimation theory,”The Annals of Statistics, pp. 79– 98, 1987

    1987

  10. [10]

    A bayesian analysis for change point problems,

    D. Barry and J. A. Hartigan, “A bayesian analysis for change point problems,”Journal of the American Statistical Association, vol. 88, no. 421, pp. 309–319, 1993

    1993

  11. [11]

    Some conservative stopping rules for the operational testing of safety critical software,

    B. Littlewood and D. Wright, “Some conservative stopping rules for the operational testing of safety critical software,”IEEE Trans. on software Engineering, vol. 23, no. 11, pp. 673–683, 1997

    1997

  12. [12]

    The unnecessity of assuming statistically independent tests in bayesian software reliability assessments,

    K. Salako and X. Zhao, “The unnecessity of assuming statistically independent tests in bayesian software reliability assessments,”IEEE Trans. on Software Engineering, vol. 49, no. 4, pp. 2829–2838, 2023

    2023

  13. [13]

    A binary Markov process model for random testing,

    S. Chen and S. Mills, “A binary Markov process model for random testing,”IEEE Trans. on Softw. Eng., vol. 22, no. 3, pp. 218–223, 1996

    1996

  14. [14]

    Failure correlation in software reliability models,

    K. Goseva-Popstojanova and K. S. Trivedi, “Failure correlation in software reliability models,”IEEE Trans. on Reliability, vol. 49, no. 1, pp. 37–48, 2000

    2000

  15. [15]

    Hidden markov models in reliability and maintenance,

    M. L. G ´amiz, N. Limnios, and M. del Carmen Segovia-Garc ´ıa, “Hidden markov models in reliability and maintenance,”European Journal of Operational Research, vol. 304, no. 3, pp. 1242–1255, 2023

    2023

  16. [16]

    Resilience evaluation of multi- feature system based on hidden markov model,

    J. Liu, J. Zhang, Q. Tian, and B. Wu, “Resilience evaluation of multi- feature system based on hidden markov model,”Reliability Engineering & System Safety, vol. 253, p. 110561, 2025

    2025

  17. [17]

    Large language models as markov chains,

    O. Zekri, A. Odonnat, A. Benechehab, L. Bleistein, N. Boull ´e, and I. Redko, “Large language models as markov chains,”arXiv preprint arXiv:2410.02724, 2024

    Pith/arXiv arXiv 2024

  18. [18]

    From self-attention to markov models: Unveiling the dynamics of generative transformers,

    M. E. Ildiz, Y . Huang, Y . Li, A. S. Rawat, and S. Oymak, “From self-attention to markov models: Unveiling the dynamics of generative transformers,”arXiv preprint arXiv:2402.13512, 2024

    Pith/arXiv arXiv 2024

  19. [19]

    A survey of confidence estimation and calibration in large language models,

    J. Geng, F. Cai, Y . Wang, H. Koeppl, P. Nakov, and I. Gurevych, “A survey of confidence estimation and calibration in large language models,” inProceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), 2024, pp. 6577–6595

    2024

  20. [20]

    Uncertainty quantification and confidence calibration in large language models: A survey,

    X. Liu, T. Chen, L. Da, C. Chen, Z. Lin, and H. Wei, “Uncertainty quantification and confidence calibration in large language models: A survey,” inProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 2, 2025, pp. 6107–6117

    2025

  21. [21]

    Strictly proper scoring rules, prediction, and estimation,

    T. Gneiting and A. E. Raftery, “Strictly proper scoring rules, prediction, and estimation,”Journal of the American statistical Association, vol. 102, no. 477, pp. 359–378, 2007

    2007

  22. [22]

    Capp ´e, E

    O. Capp ´e, E. Moulines, and T. Ryd ´en,Inference in hidden Markov models. Springer, 2005

    2005

  23. [23]

    A tutorial on hidden markov models and selected applications in speech recognition,

    L. R. Rabiner, “A tutorial on hidden markov models and selected applications in speech recognition,”Proceedings of the IEEE, vol. 77, no. 2, pp. 257–286, 2002

    2002

  24. [24]

    Dynamic safety assessment of autonomous vehicle based on multivariate bayesian inference (dyavsa),

    P. Popov, “Dynamic safety assessment of autonomous vehicle based on multivariate bayesian inference (dyavsa),”Journal of Reliable Intelligent Environments, vol. 11, no. 3, p. 14, 2025. APPENDIXA OUTLINE OFPROOFS OFTHEOREMS1AND2 Proof.The proof proceeds in five steps. Steps 1–3 derive the posterior distribution of the HMM parameters (Theorem 1). Steps 4–5 ...

    2025