Pith. sign in

REVIEW 2 major objections 3 minor 3 cited by

The paper claims that a multi-scale structured state-space forecaster, MS3M, can match Transformer-level accuracy for next-step RSRP prediction while running 3–10x faster, making it suitable for near-real-time O-RAN control loops.

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

2026-08-04 11:20 UTC pith:BAQVW6RH

load-bearing objection Solid engineering paper with a real stability proof and a careful data pipeline, but the headline 0.057 s 'per-inference' latency is actually a batched throughput number, and the Transformer baselines look undertuned. the 2 major comments →

arxiv 2510.05255 v2 pith:BAQVW6RH submitted 2025-10-06 cs.NI

Rivaling Transformers: Multi-Scale Structured State-Space Mixtures for Agentic 6G O-RAN

Farhad Rezazadeh , Hatim Chergui , Merouane Debbah , Houbing Song , Lingjia Liu This is my paper
classification cs.NI
keywords state-space modelstime-series forecastingKPI predictionO-RANRICHiPPO-LegSlatency-efficient inferenceRSRP
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.

MS3M is a forecaster built from multi-scale structured state-space kernels: it takes a window of multivariate radio KPIs, embeds them, applies per-channel causal convolutions whose taps come from bilinear-discretized HiPPO-LegS dynamics at several learned time scales, and mixes channels with squeeze-and-excitation gating and a gated linear unit. The paper's central claim is that this design gives Transformer-competitive accuracy for next-step RSRP prediction—RMSE 0.292 dB, MAE 0.170 dB, R2 0.993 on a 59,441-window O-RAN testbed dataset—while running in 0.057 s per inference with ~0.70M parameters, 3–10x faster than the tested Transformers on the same hardware. The reason this matters is that near-real-time RIC loops have tight latency and memory budgets, and attention-based models with quadratic cost and large footprints are hard to fit there; a stable, linear-time forecaster makes control-grade prediction practical inside an xApp. The paper also formalizes a leakage-safe training pipeline (past-only covariates, chronological splits, train-only standardization) that is reusable for fair benchmarking.

Core claim

On its own terms, MS3M is a strict-causal sequence model that approximates fading-memory radio dynamics by summing stable exponentials: the HiPPO-LegS continuous-time operator is discretized with the bilinear (Tustin) transform, which maps a Hurwitz A to a Schur-stable discrete transition for any positive step size, and the resulting impulse responses are used directly as depthwise causal convolution kernels. A mixture of M such kernels with learned per-component step sizes captures fast and slow dynamics within one receptive field; squeeze-and-excitation gating reweights channels, and a compact GLU layer models cross-feature nonlinearities. The paper reports that on the test tail this model

What carries the argument

The load-bearing object is the MS3M layer: a depthwise causal convolution whose finite impulse response is constructed from a Tustin-discretized HiPPO-LegS state-space model. HiPPO-LegS is a continuous-time memory operator that optimally projects the history of an input onto Legendre polynomials; bilinear (Tustin) discretization turns it into a discrete transition that is Schur-stable whenever the continuous operator is Hurwitz, so every kernel decays geometrically and finite truncation is bounded. Summing M such kernels with learned step sizes creates a multi-scale dictionary of fading-memory filters, giving O(L) sequence mixing per layer. Squeeze-and-excitation gating and a gated linear un

Load-bearing premise

The load-bearing premise is that the Transformer baselines' fixed, untuned hyperparameter configuration is a fair proxy for their realistic performance on this one-step-ahead task; the 'rivaling Transformers' accuracy comparison depends on that premise.

What would settle it

Run PatchTST and iTransformer on the same leakage-safe W=32, H=1 RSRP task with per-model hyperparameter sweeps, and check whether either drops below MS3M's 0.292 dB RMSE or closes the 0.057 s latency gap; then measure MS3M's latency on an actual near-RT RIC platform with E2-interface traffic in the loop, since the reported 0.057 s is a single forward pass without I/O.

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

If this is right

  • Near-real-time RIC xApps can run per-UE RSRP forecasts at 0.057 s per inference while leaving most of the near-RT latency budget for control decisions.
  • The accuracy–efficiency Pareto result implies that one-step KPI forecasting does not require self-attention; a linear-time state-space mixer suffices on this task class.
  • Stability is guaranteed by construction: bilinear discretization of a Hurwitz operator yields Schur-stable transitions, so bounded inputs produce bounded outputs and truncation error decays geometrically.
  • The leakage-safe pipeline (past-only covariates, chronological tail splits, train-only standardization) is a reusable protocol that prevents optimistic bias in O-RAN KPI benchmarks.
  • With ~0.70M parameters and O(L) inference, the model fits memory-constrained near-RT and edge deployments where Transformers' activation footprint is prohibitive.

Where Pith is reading between the lines

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

  • The 3–10x latency advantage is measured against Transformers run with one shared, untuned hyperparameter set; per-model tuning, especially for PatchTST and iTransformer on this specific W=32/H=1 task, could narrow the accuracy gap, though the linear-time complexity advantage would remain.
  • Since the architecture is KPI-agnostic, the same kernel-mixing recipe could be extended to multi-horizon and probabilistic forecasting (the paper tests only one-step point forecasts); a distributional head or conformal wrapper would be a direct next step.
  • The SSM core is channel-independent, so cross-UE and inter-cell couplings (handover graphs, interference, PRB contention) are not modeled; injecting such structure is the most plausible route to better extrapolation in dense mobility.
  • The paper stops at open-loop forecasting; the end-to-end claim that these predictions improve closed-loop RIC policies remains untested, and a real xApp integration on a near-RT RIC platform would be the decisive validation.

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

2 major / 3 minor

Summary. MS3M is a lightweight multi-scale structured state-space forecaster for one-step-ahead RSRP prediction in O-RAN. It mixes HiPPO-LegS kernels with Tustin discretization, per-channel depthwise convolutions, Squeeze-and-Excitation gating, and a GLU channel mixer. The paper contributes a stability proposition (Schur stability under bilinear discretization), a leakage-safe training protocol, a bespoke testbed dataset (59,441 windows, 13 KPIs), and a comparison against seven Transformer baselines. Headline claims are 0.057 s per-inference latency with ~0.70M parameters, 3.4–10.3× lower latency than baselines, and RMSE 0.292 dB, MAE 0.170 dB, R2 = 0.993 on the test tail.

Significance. If the empirical claims hold, MS3M would be a practical contribution for near-RT RIC analytics: it combines a provably stable SSM construction with a compact, causal architecture and a carefully specified leakage-safe evaluation protocol. The paper includes a self-contained proof (Prop. 1) and releases source code, which strengthens reproducibility; the diagnostics (bootstrap CIs, residual analysis, permutation importance) are unusually thorough. The skeptical concern about circularity does not, on reading, land: the stability proof is independent, and the leakage-safe pipeline is standard supervised learning. However, the central latency/speedup claim is not well-defined as written, and the baseline tuning protocol may understate Transformer performance; the significance therefore hinges on these points being resolved.

major comments (2)
  1. [Section IV-F vs. Abstract/Table III] The headline latency number is not well-defined. §IV-F states 'inference time is for a single forward pass over the contiguous test tail (data already in memory), after a short warm-up.' The Abstract and C3 report '0.057 s per-inference latency.' If the test tail is processed as one batch, 0.057 s is aggregate batch time, not per-inference latency; with a 15% test split of 59,441 windows (~8,900 windows), per-window latency would be ~6 µs, which is implausible for a 0.7M-parameter model. If 0.057 s is a single window, the stated protocol is wrong. Because every speedup ratio in Table III is computed from these values, the central O-RAN readiness claim (C3) cannot be verified. Please report the test-tail size, batching, hardware, and separate per-window/request latency.
  2. [Sections IV-C/IV-D/IV-G] Baseline tuning undermines the 'rivaling Transformers' claim. All baselines share one hyperparameter set (Adam LR=2e-3, batch 256, max 60 epochs, patience 20) with no per-model sweeps, while MS3M is selected via ablations (Section V). PatchTST and iTransformer achieve only +15.3% and +10.0% RMSE skill over persistence (RMSE 3.197 and 3.396 vs persistence ~3.78), suggesting these baselines are far from their achievable performance on this task. Since the accuracy comparison is a central claim, the paper should either tune each baseline (or use each method's published recommended hyperparameters) and report MS3M's validation-based selection; otherwise the claimed superiority may be an artifact of under-tuned baselines.
minor comments (3)
  1. [Table V vs Table III] RMSE/MAE values differ between Table V (0.290/0.169) and Table III (0.292/0.170). Please reconcile or clarify the source of the discrepancy.
  2. [Section III vs Section IV-B] Section III uses Nseq=28 for sequential samples, while Section IV-B and Table III state W=32 for all models. Clarify the actual window length used in the experiments.
  3. [Table IV] The MS3M row 'drop 0.12×10−3/20' is ambiguous; specify dropout, learning rate, and patience separately to match the other baseline rows.

Circularity Check

0 steps flagged

No significant circularity: the accuracy, latency, and stability claims rest on held-out evaluation, measured timing, and a self-contained algebraic proof rather than on definitions or load-bearing self-citation.

full rationale

The paper's main quantitative claims are: (i) MS3M accuracy on a held-out chronological test tail, (ii) measured inference latency and derived speedups, and (iii) Schur stability of the discretized SSM. None of these reduces to its own inputs by construction. The model is trained on the training split (Algorithms 1-2) and evaluated on a contiguous test tail after train-only standardization, with errors reported in inverse-standardized physical units; this is standard supervised learning, not a fitted parameter being renamed as a prediction. The stability claim is a conditional mathematical proof (Proposition 1, Eq. 4): if the HiPPO-LegS operator is Hurwitz, the bilinear transform is Schur-stable, with the proof given in the text. Lemma 1 bounds exponential kernel decay from that stability and is also self-contained. The empirical latency figure is potentially ambiguous (a single batched forward pass over the test tail versus a true per-inference latency), and the Transformer baselines are deliberately not per-model tuned; these are correctness/fairness concerns, not circularity, because neither makes an output equal to a fitted input by construction. The only self-citations are [30] for the testbed setup and [32] for the KPI dataset/statistics; they are data-provenance citations and are not used as the authority for the model's accuracy, stability, or any uniqueness/optimality claim. No circular step is established.

Axiom & Free-Parameter Ledger

12 free parameters · 6 axioms · 0 invented entities

The ledger captures the hand-chosen architecture hyperparameters, the learned SSM kernels/time scales, the data preprocessing choices, and the mathematical/domain assumptions underlying the empirical claims. No new physical entities are introduced; the model is a learned forecaster, so its trained weights are enumerated here only insofar as they constitute fitted quantities the central claim depends on.

free parameters (12)
  • lookback W (model) / Nseq (data) = 32 / 28
    Benchmark uses W=32 but data section builds sequences of length Nseq=28; chosen by hand with no sensitivity analysis.
  • embedding width d = 128
    Hand-chosen model width; Table IV.
  • SSM state size N = 64
    Hand-chosen state dimension; Table IV.
  • mixture components M = 4
    Hand-chosen number of time scales; Table IV.
  • number of layers L_ell = 4
    Hand-chosen depth; Table IV 'S6Mix x 4'.
  • dropout = 0.12
    Hand-chosen; Table IV.
  • kernel support length Lk = not reported
    Only mentioned as an ablation axis; the final value is not given in the configuration tables.
  • SE reduction ratio r = not reported
    Squeeze-and-Excitation gate uses a reduction ratio r but the final value is absent from the reported configuration.
  • learned time scales Delta t = softplus(tau) = trained
    Per-component step sizes are optimized on training data; they determine the multi-scale mixing and are not derived from first principles.
  • SSM kernel parameters (B,C,D) = trained
    The impulse-response taps used in depthwise convolutions are learned from data, which is standard for SSMs but means the kernels are fitted, not predicted.
  • IQR pruning quantiles and multiplier = Q1=0.10, Q3=0.90, 1.5*IQR
    Hand-set preprocessing rule that removes tail samples before training and evaluation; no sensitivity analysis is given.
  • training hyperparameters = LR=2e-3, batch=256, max 60 epochs, patience 20, fixed seed
    Shared with baselines; no per-model tuning was performed.
axioms (6)
  • standard math The HiPPO-LegS continuous-time operator A_ct is Hurwitz (all eigenvalues in the open left half-plane).
    Used in Prop. 1 and throughout the discretization section; well-known property of the LegS operator.
  • standard math Bilinear (Tustin) discretization maps Hurwitz-stable A_ct to Schur-stable A(Delta t) for any Delta t > 0.
    Proved in Prop. 1; standard control-theory result, self-contained in the paper.
  • domain assumption There exists an operator norm with ||A(Delta t)|| <= alpha < 1 after Schur stability.
    Lemma 1 relies on a norm bound; Schur stability implies such a norm exists in principle, but the paper does not identify the norm used in the actual model, so the geometric tail bound may not hold for the implemented kernel truncation in a standard norm.
  • domain assumption Sums of stable exponentials (from multi-scale SSM mixtures) can approximate causal fading-memory filters on compact domains.
    Invoked in Section II-I(e) as expressivity justification, citing Laguerre/Kautz theory; not proven for the exact MS3M architecture.
  • domain assumption The O-RAN testbed logs and the alignment/missingness preprocessing preserve the true KPI dynamics.
    Section III: windowed averaging, left-join, sentinel -1 for Delay, and dropping rows missing other KPIs are ad hoc decisions that shape the dataset and thus all empirical results.
  • domain assumption The test tail (15% chronological split) is representative of the deployment distribution, with no concept drift during the test period.
    Section IV-B uses contiguous tail splits; the paper's accuracy/latency claims assume the test tail reflects real operating conditions.

pith-pipeline@v1.3.0-alltime-deepseek · 20311 in / 17104 out tokens · 142245 ms · 2026-08-04T11:20:35.410312+00:00 · methodology

0 comments
read the original abstract

In sixth-generation (6G) Open Radio Access Networks (O-RAN), proactive control is preferable. A key open challenge is delivering control-grade predictions within Near-Real-Time (Near-RT) latency and computational constraints under multi-timescale dynamics. We therefore cast RAN Intelligent Controller (RIC) analytics as an agentic perceive-predict xApp that turns noisy, multivariate RAN telemetry into short-horizon per-User Equipment (UE) key performance indicator (KPI) forecasts to drive anticipatory control. In this regard, Transformers are powerful for sequence learning and time-series forecasting, but they are memory-intensive, which limits Near-RT RIC use. Therefore, we need models that maintain accuracy while reducing latency and data movement. To this end, we propose a lightweight Multi-Scale Structured State-Space Mixtures (MS3M) forecaster that mixes HiPPO-LegS kernels to capture multi-timescale radio dynamics. We develop stable discrete state-space models (SSMs) via bilinear (Tustin) discretization and apply their causal impulse responses as per-feature depthwise convolutions. Squeeze-and-Excitation gating dynamically reweights KPI channels as conditions change, and a compact gated channel-mixing layer models cross-feature nonlinearities without Transformer-level cost. The model is KPI-agnostic -- Reference Signal Received Power (RSRP) serves as a canonical use case -- and is trained on sliding windows to predict the immediate next step. Empirical evaluations conducted using our bespoke O-RAN testbed KPI time-series dataset (59,441 windows across 13 KPIs). Crucially for O-RAN constraints, MS3M achieves a 0.057 s per-inference latency with 0.70M parameters, yielding 3-10x lower latency than the Transformer baselines evaluated on the same hardware, while maintaining competitive accuracy.

Figures

Figures reproduced from arXiv: 2510.05255 by Farhad Rezazadeh, Hatim Chergui, Houbing Song, Lingjia Liu, Merouane Debbah.

Figure 1
Figure 1. Figure 1: Virginia Tech Innovation Campus O-RAN testbed setup [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comprehensive test-set diagnostics for the MS [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 3 Pith papers

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

  1. LiQSS: Post-Transformer Linear Quantum-Inspired State-Space Tensor Networks for Real-Time 6G

    cs.NI 2026-01 conditional novelty 5.0

    LiQSS, a linear-time state-space tensor-network forecaster, matches or beats Transformer-based KPI predictors on an O-RAN dataset with up to 155x fewer parameters and faster inference.

  2. Agentic World Modeling for 6G: Near-Real-Time Generative State-Space Reasoning

    cs.NI 2025-11 conditional novelty 5.0

    WM-MS3M adds a compact stochastic latent and dual decoders to a causal multi-scale SSM, improving KPI MAE by 1.69% over MS3M with 32% fewer parameters and enabling PRB what-if rollouts.

  3. A Tutorial on Cognitive Biases in Agentic AI-Driven 6G Autonomous Networks

    cs.NI 2025-10 unverdicted novelty 5.0

    Randomized Weibull anchors and debiased collective memory with decay and inflection bonuses let agentic AI in 6G cut anchoring, temporal, and confirmation biases, doubling energy savings to 25% and reducing latency by...

Reference graph

Works this paper leans on

34 extracted references · 11 linked inside Pith · cited by 3 Pith papers

  1. [1]

    ETSI TS 103 983, 01 2024

    A1 interface: General Aspects and Principles, ETSI (Produced by O-RAN ALLIANCE) Std. ETSI TS 103 983, 01 2024. [Online]. Available: https://www.etsi.org/deliver/etsi ts/103900 103999/103983/03. 01.00 60/ts 103983v030100p.pdf

    2024

  2. [2]

    ETSI TS 104 038, 10 2024

    E2 interface: General Aspects and Principles, ETSI (Produced by O-RAN ALLIANCE) Std. ETSI TS 104 038, 10 2024. [Online]. Available: https://www.etsi.org/deliver/etsi ts/104000 104099/104038/04. 01.00 60/ts 104038v040100p.pdf

    2024

  3. [3]

    2024, (O-RAN.WG10.O1-Interface-R003- v11.00)

    O-RAN Operations and Maintenance Interface Specification, ETSI (pro- duced from O-RAN ALLIANCE docs) Publicly Available Specification (PAS) ETSI TS 104 043, Jun. 2024, (O-RAN.WG10.O1-Interface-R003- v11.00). [Online]. Available: https://www.etsi.org/deliver/etsi ts/104000 104099/104043/11.00.00 60/ts 104043v110000p.pdf

    2024

  4. [4]

    2024, O-RAN.WG1.OAD-R003-v08.00

    O-RAN Architecture Description, ETSI (produced from O-RAN AL- LIANCE docs) Publicly Available Specification (PAS) ETSI TS 103 982, Jan. 2024, O-RAN.WG1.OAD-R003-v08.00. [Online]. Avail- able: https://www.etsi.org/deliver/etsi ts/103900 103999/103982/08.00. 00 60/ts 103982v080000p.pdf

    2024

  5. [5]

    Attention Is All You Need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention Is All You Need,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 30, 2017, pp. 5998–6008. [Online]. Available: https://papers.nips.cc/paper/ 7181-attention-is-all-you-need.pdf

    2017

  6. [6]

    Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting,

    H. Zhou, S. Zhang, J. Peng, S. Zhang, J. Li, H. Xiong, and W. Zhang, “Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 35, no. 12, 2021, pp. 11 106–11 115. [Online]. Available: https://doi.org/10.1609/aaai.v35i12.17325

    doi:10.1609/aaai.v35i12.17325 2021

  7. [7]

    Autoformer: Decomposition Transformers with Auto-Correlation for Long-term Series Forecasting,

    H. Wu, J. Xu, J. Wang, and M. Long, “Autoformer: Decomposition Transformers with Auto-Correlation for Long-term Series Forecasting,” in Proceedings of the 35th International Conference on Neural Information Processing Systems, ser. NIPS ’21. Red Hook, NY , USA: Curran Associates Inc., 2021

    2021

  8. [8]

    FEDformer: Frequency Enhanced Decomposed Transformer for Long-term Series Forecasting,

    T. Zhou, Z. Ma, Q. Wen, X. Wang, L. Sun, and R. Jin, “FEDformer: Frequency Enhanced Decomposed Transformer for Long-term Series Forecasting,” inProceedings of the 39th International Conference on Machine Learning, vol. 162. PMLR, 2022, pp. 27 268–27 286. [Online]. Available: https://proceedings.mlr.press/v162/zhou22g.html

    2022

  9. [9]

    A Time Series is Worth 64 Words: Long-term Forecasting with Transformers,

    Y . Nie, N. H. Nguyen, P. Sinthong, and J. Kalagnanam, “A Time Series is Worth 64 Words: Long-term Forecasting with Transformers,” inThe Eleventh International Conference on Learning Representations, 2023. [Online]. Available: https://openreview.net/forum?id=Jbdc0vTOcol

    2023

  10. [10]

    Temporal Fusion Transformers for Interpretable Multi-horizon Time Series Forecasting,

    B. Lim, S. ¨O. Arık, N. Loeff, and T. Pfister, “Temporal Fusion Transformers for Interpretable Multi-horizon Time Series Forecasting,” International Journal of Forecasting, vol. 37, no. 4, pp. 1748–1764,

  11. [11]

    ETSformer: Exponential Smoothing Transformers for Time-series Forecasting,

    G. Woo, C. Liu, D. Sahoo, A. Kumar, and S. Hoi, “ETSformer: Exponential Smoothing Transformers for Time-series Forecasting,”arXiv preprint arXiv:2202.01381, 2022. [Online]. Available: https://arxiv.org/ abs/2202.01381

    Pith/arXiv arXiv 2022

  12. [12]

    Crossformer: Transformer Utilizing Cross- Dimension Dependency for Multivariate Time Series Forecasting,

    Y . Zhang and J. Yan, “Crossformer: Transformer Utilizing Cross- Dimension Dependency for Multivariate Time Series Forecasting,” inThe Eleventh International Conference on Learning Representations, 2023. [Online]. Available: https://openreview.net/forum?id=vSVLM2j9eie

    2023

  13. [13]

    iTrans- former: Inverted Transformers Are Effective for Time Series Forecasting,

    Y . Liu, T. Hu, H. Zhang, H. Wu, S. Wang, L. Ma, and M. Long, “iTrans- former: Inverted Transformers Are Effective for Time Series Forecasting,” inThe Twelfth International Conference on Learning Representations,

  14. [14]

    Transformers in Time Series: A Survey,

    Q. Wen, T. Zhou, C. Zhang, W. Chen, Z. Ma, J. Yan, and L. Sun, “Transformers in Time Series: A Survey,” inProceedings of the Thirty- Second International Joint Conference on Artificial Intelligence, IJCAI-23, E. Elkind, Ed. International Joint Conferences on Artificial Intelligence Organization, 8 2023, pp. 6778–6786, survey Track. [Online]. Available: ht...

    doi:10.24963/ijcai.2023/759 2023

  15. [15]

    Efficiently Modeling Long Sequences with Structured State Spaces,

    A. Gu, K. Goel, and C. R ´e, “Efficiently Modeling Long Sequences with Structured State Spaces,” inInternational Conference on Learning Representations (ICLR), 2022. [Online]. Available: https://openreview. net/forum?id=uYLFoz1vlAC

    2022

  16. [16]

    It’s Raw! Audio Generation with State-Space Models,

    K. Goel, A. Gu, C. Donahue, and C. R ´e, “It’s Raw! Audio Generation with State-Space Models,” inProceedings of the 39th International Conference on Machine Learning (ICML), ser. Proceedings of Machine Learning Research, vol. 162. PMLR, 2022, introduces Hurwitz-based stabilization for SSMs and links CT stability to DT via bilinear transform. [Online]. Ava...

    2022

  17. [17]

    On the Parameterization and Initialization of Diagonal State Space Models,

    A. Gu, A. Gupta, D. Reich, and C. R ´e, “On the Parameterization and Initialization of Diagonal State Space Models,”arXiv preprint arXiv:2206.11893, 2022, analyzes stable parameterizations/initializations for diagonal SSMs (negative real-part spectra). [Online]. Available: https://arxiv.org/abs/2206.11893

    Pith/arXiv arXiv 2022

  18. [18]

    Diagonal State Spaces are as Effective as Structured State Spaces,

    A. Gupta, A. Gu, and J. Berant, “Diagonal State Spaces are as Effective as Structured State Spaces,” inAdvances in Neural Information Processing Systems (NeurIPS), 2022, neurIPS 2022, Main Conference Track. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/2022/ file/9156b0f6dfa9bbd18c79cc459ef5d61c-Paper-Conference.pdf

    2022

  19. [19]

    Simplified State Space Layers for Sequence Modeling,

    J. T. H. Smith, A. Warrington, and S. Linderman, “Simplified State Space Layers for Sequence Modeling,”arXiv preprint arXiv:2208.04933, 2022, s4/S5 variants with practical, stability-preserving initializations. [Online]. Available: https://arxiv.org/abs/2208.04933

    Pith/arXiv arXiv 2022

  20. [20]

    How to Train Your HiPPO: State Space Models with Generalized Orthogonal Basis Projections,

    A. Gu, T. Dao, S. Ermon, and C. R ´e, “How to Train Your HiPPO: State Space Models with Generalized Orthogonal Basis Projections,” arXiv preprint arXiv:2206.12037, 2022, hiPPO matrices used to initialize SSMs; foundations tied to stable continuous-time dynamics. [Online]. Available: https://arxiv.org/abs/2206.12037

    Pith/arXiv arXiv 2022

  21. [21]

    HiPPO: Recur- rent Memory with Optimal Polynomial Projections,

    A. Gu, T. Dao, S. Ermon, A. Rudra, and C. R ´e, “HiPPO: Recur- rent Memory with Optimal Polynomial Projections,”arXiv preprint arXiv:2008.07669, 2020. [Online]. Available: https://arxiv.org/abs/2008. 07669

    Pith/arXiv arXiv 2008

  22. [22]

    MobileNetV2: Inverted Residuals and Linear Bottlenecks,

    M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “MobileNetV2: Inverted Residuals and Linear Bottlenecks,”arXiv preprint arXiv:1801.04381, 2018. [Online]. Available: https://arxiv.org/abs/1801. 04381

    Pith/arXiv arXiv 2018

  23. [23]

    Squeeze-and-Excitation Networks,

    J. Hu, L. Shen, S. Albanie, G. Sun, and E. Wu, “Squeeze-and-Excitation Networks,”arXiv preprint arXiv:1709.01507, 2017. [Online]. Available: https://arxiv.org/abs/1709.01507

    Pith/arXiv arXiv 2017

  24. [24]

    Language Modeling with Gated Convolutional Networks,

    Y . N. Dauphin, A. Fan, M. Auli, and D. Grangier, “Language Modeling with Gated Convolutional Networks,” inProceedings of the 34th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, D. Precup and Y . W. Teh, Eds., vol. 70. PMLR, 2017, pp. 933–941. [Online]. Available: https://proceedings.mlr. press/v70/dauphin17a.html

    2017

  25. [25]

    Exploring Adversarial Robustness of Deep State Space Models,

    B. Qi, Y . Luo, J. Gao, P. Li, K. Tian, Z. Ma, and B. Zhou, “Exploring Adversarial Robustness of Deep State Space Models,” inAdvances in Neural Information Processing Systems (NeurIPS), 2024, paper uses bilinear (Tustin) discretization for S4. [On- line]. Available: https://proceedings.neurips.cc/paper files/paper/2024/ file/0c8bbdf2657b58fa0a620f650fbdd4...

    2024

  26. [27]

    P-SpikeSSM: Harnessing Probabilistic Spiking State Space Models for Long-Range Dependency Tasks,

    M. Bal and A. Sengupta, “P-SpikeSSM: Harnessing Probabilistic Spiking State Space Models for Long-Range Dependency Tasks,” inProceedings of the International Conference on Learning Representations (ICLR), 2025, iCLR 2025 version; arXiv:2406.02923. [Online]. Available: https: //arxiv.org/pdf/2406.02923

    Pith/arXiv arXiv 2025

  27. [29]

    An Introduction to Models Based on Laguerre, Kautz and Other Related Orthonormal Functions – Part II: Non-linear Models,

    G. H. C. Oliveira, A. da Rosa, R. J. G. B. Campello, J. B. Machado, and W. C. Amaral, “An Introduction to Models Based on Laguerre, Kautz and Other Related Orthonormal Functions – Part II: Non-linear Models,” International Journal of Modelling, Identification and Control, vol. 16, no. 1, pp. 1–14, 2012

    2012

  28. [30]

    Learning Low-Dimensional Representation for O-RAN Testing via Transformer- ESN,

    J. Dai, R. Zhao, F. Rezazadeh, L. Zheng, H. Wang, and L. Liu, “Learning Low-Dimensional Representation for O-RAN Testing via Transformer- ESN,” 2025, accepted for publication at IEEE MASS, 2025

    2025

  29. [31]

    O-RAN End-to-End Test Specification,

    “O-RAN End-to-End Test Specification,” O-RAN Alliance, Tech. Rep. O-RAN.TIFG.E2E-Test.0-R003-v06.00, 2024, release R003

    2024

  30. [32]

    Available: https://arxiv.org/abs/2406.05815

    [Online]. Available: https://arxiv.org/abs/2406.05815

    Pith/arXiv arXiv

  31. [36]

    An Experimental Reservoir-Augmented Foundation Model: 6G O-RAN Case Study,

    F. Rezazadeh, R. Zhao, J. Dai, A. Ashtari Gargari, H. Chergui, and L. Liu, “An Experimental Reservoir-Augmented Foundation Model: 6G O-RAN Case Study,” 2025, arXiv:2508.07778. [Online]. Available: https://arxiv.org/abs/2508.07778

    Pith/arXiv arXiv 2025

  32. [2021]

    Available: https://www.sciencedirect.com/science/article/ pii/S0169207021000637

    [Online]. Available: https://www.sciencedirect.com/science/article/ pii/S0169207021000637

  33. [2024]

    Available: https://openreview.net/forum?id=JePfAI8fah

    [Online]. Available: https://openreview.net/forum?id=JePfAI8fah

  34. [2025]

    Available: https://arxiv.org/abs/2502.13729

    [Online]. Available: https://arxiv.org/abs/2502.13729

    Pith/arXiv arXiv