LogSig-SSM: Time-Series Modelling with Multi-Scale Log-Signature Compression for State-Space Models

Felix Oury; Nicolas Calvo Peiro; Reiko J. Tanaka

cs.LGOct 4, 2026

LogSig-SSM: Time-Series Modelling with Multi-Scale Log-Signature Compression for State-Space Models

Authors: Felix Oury, Nicolas Calvo Peiro, Reiko J. Tanaka

Organizations: Department of Computing Imperial College London · Department of Bioengineering Imperial College London

Abstract

Time-series data are often sampled irregularly at high frequencies and exhibit long-range dependencies, which makes long-horizon modelling difficult. Continuous-time models such as neural controlled differential equations (NCDEs) and neural rough differential equations (NRDEs) can handle irregular sampling, but they scale poorly to long sequences. Selective state-space models (SSMs) such as Mamba scale linearly with sequence length, but they provide limited recurrent mixing across hidden dimensions within a single block. We propose LogSig-SSM (Log-Signature Compression for State-Space Models), which first compresses long multivariate time series into a shorter sequence of tokens using multi-scale windowed log-signatures, and then processes these tokens with a selective SSM backbone. LogSig-SSM is scalable and robust to irregular sampling, combining log-signature tokens that capture higher-order cross-channel interactions with a selective SSM that models long-range dependencies. The model also admits a continuous-time interpretation as an NCDE/NRDE-style system driven by a log-signature-based input, in which selectivity induces an input-dependent rescaling of the latent dynamics. Across four benchmarks, namely long-sequence classification on UEA, high-frequency physiological regression on PPG-DaLiA, multivariate weather forecasting, and irregularly sampled clinical prediction on PhysioNet Sepsis, LogSig-SSM outperforms or matches strong SSM and continuous-time baselines while training up to $30\times$ faster and using up to $37\times$ less GPU memory than Mamba on the longest sequences.

Figures & tables

Figure 1 : Discrete-time LogSig-SSM pipeline. A multivariate time series $x\in\mathbb{R}^{d_{x}\times T}$ is summarised at strided anchors $\tilde{t}_{\tau}:=t_{\tau s}$ . For each scale $k$ , a windowed segment $\mathcal{W}_{k}(x)_{\tau}$ is extracted, and its truncated log-signature coordinate vector, $u_{\tau,k}:=\phi_{m_{k}}\!\Big(\mathrm{LogSig}_{m_{k}}\!\big(\mathcal{W}_{k}(x)_{\tau}\big)\Big)\in\mathbb{R}^{D_{k}},$ is computed. These are concatenated across scales to form the token $U_{\tau}:=\mathrm{concat}\big(u_{\tau,1},\dots,u_{\tau,K}\big)\in\mathbb{R}^{D_{\mathrm{tot}}}$ . Finally, $U_{\tau}$ is projected to the scan width $d$ via a learned linear embedding to form $\widetilde{U}_{\tau}=W_{\mathrm{in}}U_{\tau}$ , which serves as the input to the Mamba block.

Continuous-time models
Model	EigenWorms	SCP1	SCP2	Ethanol	Heartbeat	Motor	Avg	Rank
Seq. length	17,984	896	1,152	1,751	405	3,000
# classes	5	2	2	4	2	2
NRDE [ 25 ]	$83.9{\scriptstyle\pm 7.3}$	$80.9{\scriptstyle\pm 2.5}$	$53.7{\scriptstyle\pm 6.9}$	$25.3{\scriptstyle\pm 1.8}$	$72.9{\scriptstyle\pm 4.8}$	$47.0{\scriptstyle\pm 5.7}$	$60.6$	$11.5$
NCDE [ 19 ]	$75.0{\scriptstyle\pm 3.9}$	$79.8{\scriptstyle\pm 5.6}$	$53.0{\scriptstyle\pm 2.8}$	$29.9{\scriptstyle\pm 6.5}$	$73.9{\scriptstyle\pm 2.6}$	$49.5{\scriptstyle\pm 2.8}$	$60.2$	$11.3$
Log-NCDE [ 38 ]	$85.6{\scriptstyle\pm 5.1}$	$83.1{\scriptstyle\pm 2.8}$	$53.7{\scriptstyle\pm 4.1}$	$34.4{\scriptstyle\pm 6.4}$	$75.2{\scriptstyle\pm 4.6}$	$53.7{\scriptstyle\pm 5.3}$	$64.3$	$7.7$

Table 1 : UEA long-sequence classification accuracy (%

\uparrow

), mean

\pm

std over five random splits, and average rank over the six datasets (

\downarrow

, competition ranking, missing entries ranked last). Bold = best, underline = second best, italic = third best per column (ties share a mark). † Transformer OOM on EigenWorms ; TSPulse not evaluated on EigenWorms ; ∗ Model evaluated on 1 seed.

	NCDE	NRDE	Log-NCDE	Mamba	S5	LinOSS-IM	D-LinOSS	LogSig-SSM
Mem. (MB)	2484	2506	2510	13486	6646	3488	3488	362
Time (s/1k steps)	24595	5386	1956	122	31	14	14	4
Accuracy (%)	75.0	83.9	85.6	70.9	81.1	95.0	93.9	95.3

Table 2 : Efficiency on EigenWorms (

T=17{,}984

): GPU memory (MB), wall-clock time per 1000 training steps (s) on one RTX 4090, and test accuracy (%). Full results in Appendix B .

Regression (PPG-DaLiA)		Forecasting (Weather)		Irregular Sampling (PhysioNet Sepsis)
Model	MSE $\!\times\!10^{-2}$	Model	MAE	Method	OI	No OI
NRDE	$9.90{\scriptstyle\pm 0.97}$	Informer	$0.731$	GRU- $\Delta t$	$87.8{\scriptstyle\pm 0.6}$	$84.0{\scriptstyle\pm 0.7}$
NCDE	$13.54{\scriptstyle\pm 0.69}$	Informer †	$0.741$	GRU-D	$87.1{\scriptstyle\pm 2.2}$	$85.0{\scriptstyle\pm 1.3}$
Log-NCDE	$9.56{\scriptstyle\pm 0.59}$	LogTrans	$0.773$	GRU-ODE	$85.2{\scriptstyle\pm 1.0}$	$77.1{\scriptstyle\pm 2.4}$
LRU	$12.17{\scriptstyle\pm 0.49}$	Reformer	$1.575$	ODE-RNN	$87.4{\scriptstyle\pm 1.6}$	$83.3{\scriptstyle\pm 2.0}$
S5	$12.63{\scriptstyle\pm 1.25}$	LSTMa	$1.109$	Neural CDE	$88.0{\scriptstyle\pm 0.6}$	$77.6{\scriptstyle\pm 0.9}$

Table 3 : Results on regression, forecasting, and irregular-sampling benchmarks. Left: PPG-DaLiA heart-rate regression, MSE

\times 10^{-2}

(

\downarrow

), mean

\pm

std over five runs [ 28 ] . Centre: weather forecasting MAE (

\downarrow

), predict 30 days of hourly data from the preceding 30 days; Informer † uses canonical self-attention. Right: PhysioNet Sepsis AUROC (

\uparrow

10.3\%

of values observed; OI = with observation indicators [ 30 ] . Bold = best, underline = second best per sub-table.

Configuration	Avg. acc. (%)	$\Delta$ (pp)	What it adds
Window descriptor (all rows compress the sequence), Mamba backbone fixed
Mamba (raw input)	58.6	–	no compression, no channel mixing
Catch24 statistics [ 22 ]	58.9	$+0.3$	path discarded, summary statistics only
Depthwise convolution	60.1	$+1.5$	channels kept separate
Standard convolution	62.9	$+4.3$	first-order (linear) channel mixing
Wavelet scattering	67.6	$+9.0$	multi-scale spectral content, no cross-channel geometry

Table 4 : UEA ablations, average accuracy over the six datasets (%

\uparrow

). All variants share the 200-trial Optuna protocol and budget of LogSig-SSM, and its window positions and token dimension where applicable;

\Delta

is relative to raw Mamba. Per-dataset results in Appendix C , Table 7 .

Appendix figures & tables12 assets

Supplementary material from the paper’s appendix.

Appendix

		NRDE	NCDE	Log-NCDE	LRU	S5	Mamba	S6	LinOSS-IMEX	LinOSS-IM	D-LinOSS	LogSig-SSM
EigenWorms	$\|\theta\|$	105110	166789	37977	101129	22007	27381	15045	26119	134279	134279	52933
	mem.	2506	2484	2510	10716	6646	13486	7922	6556	3488	3488	362
	time	5386	24595	1956	94	31	122	68	37	14	14	4
SCP1	$\|\theta\|$	117187	166274	91557	25892	226328	184194	24898	447944	991240	992776	310786
	mem.	716	694	724	960	1798	1110	904	4768	4772	4790	202
	time	1014	973	635	9	17	7	3	42	38	38	3

Appendix

Table 5 : Compute requirements for the UEA classification tasks of Section 5 . Bold indicates the lowest value per dataset for parameter count (

|\theta|

), GPU memory usage (MB), and time (seconds) per 1000 steps. All statistics were recorded on an NVIDIA RTX 4090.

Dataset	$d_{x}$	$T$	Tokenisation (s)	% of one training run
EigenWorms	6	17,984	$1.587\pm 0.004$	1.2
SCP1	6	896	$0.887\pm 0.051$	0.8
SCP2	7	1,152	$0.052\pm 0.026$	0.1
Ethanol	3	1,751	$0.113\pm 0.031$	0.1
Heartbeat	61	405	$0.959\pm 0.026$	0.9
Motor	64	3,000	$2.469\pm 0.103$	1.5

Appendix

Table 6 : Log-signature tokenisation cost for LogSig-SSM on the UEA datasets, using iisignature [ 29 ] on CPU: wall-clock seconds for the whole dataset (mean

\pm

std over five seeds) and the same time as a percentage of one training run.

Tokeniser variants (Mamba backbone)
Model	EigenWorms	SCP1	SCP2	Ethanol	Heartbeat	Motor	Avg
Mamba (raw input)	$70.9{\scriptstyle\pm 15.8}$	$80.7{\scriptstyle\pm 1.4}$	$48.2{\scriptstyle\pm 3.9}$	$27.9{\scriptstyle\pm 4.5}$	$76.2{\scriptstyle\pm 3.8}$	$47.7{\scriptstyle\pm 4.5}$	$58.6$
ConvDepthwise-Mamba	$78.4{\scriptstyle\pm 3.0}$	$83.9{\scriptstyle\pm 5.8}$	$45.2{\scriptstyle\pm 3.8}$	$35.8{\scriptstyle\pm 12.5}$	$68.7{\scriptstyle\pm 4.3}$	$48.6{\scriptstyle\pm 2.4}$	$60.1$
ConvMixing-Mamba	$85.9{\scriptstyle\pm 11.8}$	$81.4{\scriptstyle\pm 3.4}$	$51.0{\scriptstyle\pm 3.1}$	$33.6{\scriptstyle\pm 5.6}$	$74.9{\scriptstyle\pm 10.3}$	$50.6{\scriptstyle\pm 3.8}$	$62.9$
Catch24-Mamba	$72.5{\scriptstyle\pm 11.0}$	$82.4{\scriptstyle\pm 4.7}$	$44.5{\scriptstyle\pm 4.2}$	$34.5{\scriptstyle\pm 5.7}$	$66.8{\scriptstyle\pm 6.1}$	$52.8{\scriptstyle\pm 7.0}$	$58.9$
Wst-Mamba	$89.1{\scriptstyle\pm 12.5}$	$85.9{\scriptstyle\pm 1.2}$	$53.1{\scriptstyle\pm 5.6}$	$\mathbf{53.5{\scriptstyle\pm 4.0}}$	$72.9{\scriptstyle\pm 6.3}$	$51.0{\scriptstyle\pm 9.5}$	$67.6$

Appendix

Table 7 : UEA ablations (% accuracy

\uparrow

, mean

\pm

std over five splits). Top: tokeniser variants with the Mamba backbone fixed. Bottom: backbone variants with the log-signature tokeniser fixed. All variants share the same window positions, token dimensions where applicable, and Optuna budget (Appendix F ). Bold = best per column.

	Accuracy (%)		Tokenisation (s)
Dataset	Log-signature	Signature	Log-signature	Signature
EigenWorms	$\mathbf{95.3}$	$94.4$	$1.56$	$3.73$
SCP1	$\mathbf{88.4}$	$85.7$	$0.85$	$1.34$
SCP2	$\mathbf{58.4}$	$52.9$	$0.04$	$0.05$
Average / total	$\mathbf{80.7}$	$77.6$	$2.45$	$5.12$

Appendix

Table 8: Signature against log-signature tokens at the same depth and configuration (Table 11 ): test accuracy (%, mean over five seeds) and whole-dataset tokenisation time with iisignature (s, best of three).

Dataset	$d_{x}$	$m_{k}=1$	$m_{k}=2$	$\Delta$
EigenWorms	6	$88.1{\scriptstyle\pm 6.4}$	$95.3{\scriptstyle\pm 3.6}$	$+7.2$
SCP1	6	$84.1{\scriptstyle\pm 1.4}$	$88.4{\scriptstyle\pm 1.9}$	$+4.3$
SCP2	7	$55.6{\scriptstyle\pm 4.9}$	$58.4{\scriptstyle\pm 2.9}$	$+2.8$
Ethanol	3	$51.0{\scriptstyle\pm 4.2}$	$51.4{\scriptstyle\pm 6.9}$	$+0.4$
Heartbeat	61	$78.3{\scriptstyle\pm 9.4}$	$76.9{\scriptstyle\pm 9.7}$	$-1.4$
Motor	64	$59.6{\scriptstyle\pm 7.5}$	$59.9{\scriptstyle\pm 3.4}$	$+0.3$

Appendix

Table 9 : Matched-budget comparison of truncation depth

m_{k}=1

against

m_{k}=2

on every window (UEA accuracy %; mean

\pm

std over five seeds).

\Delta

is the gain of depth two. The depth-two column is the configuration reported in Table 1 .

Figure 2 : Representative time-series from the six UEA datasets. EthanolConcentration exhibits smooth, low-frequency structure with global shape variation, while the other datasets are dominated by high-frequency or noisy dynamics.

Model	EigenWorms	SCP1	SCP2	Ethanol	Heartbeat	Motor	Avg
LogSig-SSM (searched)	$95.3{\scriptstyle\pm 3.6}$	$88.4{\scriptstyle\pm 1.9}$	$58.4{\scriptstyle\pm 2.9}$	$51.4{\scriptstyle\pm 6.9}$	$76.9{\scriptstyle\pm 9.7}$	$59.9{\scriptstyle\pm 3.4}$	$71.7$
LogSig-SSM (rule, no search)	$92.5{\scriptstyle\pm 6.8}$	$85.5{\scriptstyle\pm 3.5}$	$51.9{\scriptstyle\pm 8.9}$	$32.3{\scriptstyle\pm 2.1}$	$72.2{\scriptstyle\pm 7.3}$	$49.7{\scriptstyle\pm 4.8}$	$64.0$
S5	$81.1{\scriptstyle\pm 3.7}$	$89.9{\scriptstyle\pm 4.6}$	$50.5{\scriptstyle\pm 2.6}$	$24.1{\scriptstyle\pm 4.3}$	$77.7{\scriptstyle\pm 5.5}$	$47.7{\scriptstyle\pm 5.5}$	$61.8$
LRU	$87.8{\scriptstyle\pm 2.8}$	$82.6{\scriptstyle\pm 3.4}$	$51.2{\scriptstyle\pm 3.6}$	$21.5{\scriptstyle\pm 2.1}$	$78.4{\scriptstyle\pm 6.7}$	$48.4{\scriptstyle\pm 5.0}$	$61.7$
Mamba	$70.9{\scriptstyle\pm 15.8}$	$80.7{\scriptstyle\pm 1.4}$	$48.2{\scriptstyle\pm 3.9}$	$27.9{\scriptstyle\pm 4.5}$	$76.2{\scriptstyle\pm 3.8}$	$47.7{\scriptstyle\pm 4.5}$	$58.6$

Appendix

Table 10: Search-free rule on UEA (accuracy %, mean

\pm

std over five seeds): tokeniser hyperparameters set from the sequence length alone, Mamba backbone as in Table 11 , against the searched configuration and the raw SSM baselines of Table 1 .

Dataset	LR	Hidden	State	Blocks	Time	Conv	Exp	Norm	Lengths / Depths	Stride	Global
EigenWorms	1e-3	64	64	2	F	2	1	T	[281] / [2]	21	–
Ethanol	1e-3	16	256	6	T	2	2	T	[18, 132] / [2, 2]	66	–
Heartbeat	1e-3	128	64	2	F	4	1	T	[143] / [2]	26	–
Motor	1e-4	16	16	2	T	3	1	F	[51, 16] / [2, 2]	50	–
SCP1	1e-3	128	64	2	F	3	2	T	[56] / [2]	5	1
SCP2	1e-4	128	16	2	T	2	3	T	[36] / [2]	46	–

Appendix

Table 11 : Selected hyperparameters for LogSig-SSM . LR: learning rate; Hidden: model dimension

d

; State: state size

N

; Blocks: number of Mamba blocks; Time: time channel appended to the input; Conv: convolution width; Exp: expansion factor; Norm: per-feature normalisation; Lengths/Depths: window lengths

L_{k}

and truncation depths

m_{k}

; Global: depth of the optional global branch (– = none).

Dataset	LR	Hidden	State	Blocks	Time	Norm	Lengths	Depths	Stride	Global
EigenWorms	1e-3	64	16	4 $\times$ 4	F	F	[562, 281]	[2, 2]	57	F
Ethanol	1e-3	16	64	2 $\times$ 2	F	T	[109, 43]	[2, 1]	70	F
Heartbeat	1e-5	64	16	6 $\times$ 2	T	T	[14]	[1]	3	F
Motor	1e-4	64	16	4 $\times$ 4	T	F	[93, 250, 125]	[2,1,2]	35	T
SCP1	1e-3	16	256	6 $\times$ 4	F	F	[112, 18]	[1, 1]	30	F
SCP2	1e-4	16	64	4 $\times$ 2	F	T	[18, 144, 192]	[4,2,3]	13	T

Appendix

Table 12 : Selected hyperparameters for LogSig-S5 (log-signature tokens + S5). Blocks are reported as layers

\times

blocks.

Dataset	LR	Hidden	Heads	FFN	State	Blocks	Time	Norm	Lengths	Depths	Stride	Global
EigenWorms	1e-3	128	4	256	64	2	T	T	[374, 899]	[2, 1]	55	–
Ethanol	1e-3	16	8	128	64	6	T	F	[22, 48, 114]	[1, 1, 1]	74	–
Heartbeat	1e-5	16	8	256	64	4	T	T	[16, 25]	[1, 1]	14	–
Motor	1e-4	16	8	512	64	4	T	T	[250, 46]	[1, 1]	46	–
SCP1	1e-4	16	8	512	64	4	T	T	[74]	[1]	14	1
SCP2	1e-3	128	8	512	64	2	T	T	[68]	[2]	76	–

Appendix

Table 13 : Selected hyperparameters for LogSig-Transformer (log-signature tokens + Transformer). FFN: feed-forward dimension; Global: depth of the optional global branch (– = none).

Dataset	LR	Hidden	State	Blocks	Time	Conv	Exp	Norm	Lengths	Stride
EigenWorms	1e-3	64	256	6	T	2	4	T	[70, 749, 499, 70, 35]	135
Ethanol	1e-3	128	256	4	T	3	3	T	[109, 350, 13, 109, 350]	79
Heartbeat	1e-5	128	64	2	T	3	2	F	[11, 16, 50, 8, 3]	38
Motor	1e-3	64	16	6	F	3	3	T	[187, 11]	163
SCP1	1e-4	64	16	2	T	4	2	T	[179, 9, 179, 32, 224]	98
SCP2	1e-3	64	256	6	T	4	3	T	[192, 48, 115, 192]	67

Appendix

Table 14 : Selected hyperparameters for Catch24-Mamba (Catch24 tokens).

Dataset	LR	Hidden	State	Blocks	Conv	Exp	WST Lengths	Stride
EigenWorms	1e-3	128	64	6	2	1	[749, 2997]	48
Ethanol	1e-4	128	64	2	3	4	[18, 291, 62, 48]	66
Heartbeat	1e-4	64	64	2	3	4	[67, 67, 40]	75
Motor	1e-5	16	16	2	2	1	[375, 125, 150, 187]	125
SCP1	1e-5	128	256	2	4	2	[179, 28]	122
SCP2	1e-3	16	16	2	3	1	[48, 4, 96, 48, 192]	141

Appendix

Table 15 : Selected hyperparameters for Wst-Mamba (wavelet-scattering tokens, Kymatio Scattering1D with Morlet wavelets).

Explore similar work

May 10, 2026cs.LG

TIDES: Implicit Time-Awareness in Selective State Space Models

Selective state space models (SSMs), such as Mamba, achieve strong per-token expressivity by making the time discretization step

\TildeΔ

a learned function of the input. However, in doing so,

\TildeΔ

no longer equals the physical time gap

Δ

between consecutive observations, limiting the ability of these models to handle irregular time series. Continuous time SSMs, such as S5, keep

\TildeΔ\equivΔ

and therefore handle irregular timestamps natively, but their dynamics remain linear time invariant (LTI), limiting per token expressivity. We propose \textbf{TIDES}, a selective SSM variant that reconciles selective and continuous architectures by moving input dependence off the step size and onto the diagonal state matrix. As a result,

\TildeΔ\equivΔ

as in S5, allowing the model to handle irregular timestamps natively without sacrificing the per-token expressivity that makes selective SSMs effective. We show this on a novel \emph{Fading Flash} experimental benchmark, a compact controlled diagnostic for sequence models that jointly tests input dependence and extrapolation to out-of-distribution

Δ

values, and isolates the distinct failure modes of current state-of-the-art architectures that TIDES avoids by construction. On large-scale benchmarks, TIDES sets the new best average rank on UEA time series classification and the Physiome ODE regression benchmark, and matches or exceeds the reference baseline model on 6 of 8 natively irregular datasets from astronomy, agriculture, neuromorphic sensing, and climate events. Code available at: https://github.com/TaylanSoydan/TIDES.

Taylan Soydan, Miguel A. Bessa, Dirk Mohr +1

AIMM, ETH Zürich Zurich, Switzerland

State Space Models Sequence Modeling Time Series Time Discretization

Ask PDF

Mar 19, 2026stat.ML

Extending SSMs with the Exponentially Weighted Signature

We introduce the exponentially weighted signature (EWS), a continuous-time model that computes iterated integrals of a path, where each increment is weighted by the matrix exponential of a learnable generator over elapsed clock time. We prove that it solves a linear controlled differential equation, keeps the group-like structure and the universality of the signature, and satisfies a modified Chen identity, enabling a parallel scan. At depth one the EWS is a state-space model (SSM), and we map linear time-invariant SSMs, Mamba channels and Mamba-

2

heads to it in closed form. The EWS extends SSMs through an arbitrary matrix generator, a clock that generalises the step size to causal functionals of the input, and higher truncation depths that are non-linear in the path within a single layer. Empirically, the EWS achieves the highest average accuracy and rank on six long time-series classification datasets, where depth generally helps. Learned clocks prove necessary for state tracking on formal language tasks, and at depth one, the EWS matches or exceeds competing SSMs on regression and forecasting with far fewer parameters.

Alexandre Bloch, Benjamin Walker, Joël Mouterde +3

Mathematical Institute, University of Oxford · SKF · Department of Mathematics, Imperial College London

Distributional Signatures Signatures State Space Models Time Series

Ask PDF

May 7, 2026cs.LG

A Simple State Space Model Excels at Multivariate Time Series Classification

Structured state space models (SSMs) have recently emerged as a promising foundation for sequence modeling, with Mamba-based architectures demonstrating strong performance through input-dependent state transitions, albeit at considerable complexity. However, their application to time-series classification (TSC) has been largely limited to Mamba-style architectures, leaving the broader SSM design space underexplored. We present the first systematic study spanning diagonal SSMs (S4D) and input-dependent SSMs (Mamba family) on large-scale TSC benchmarks, asking whether such complexity is necessary for top performance. Our results reveal a surprising finding: S4D consistently outperforms Mamba-based variants in both accuracy and efficiency, challenging the assumption that increased complexity translates to meaningful gains in TSC. Building on this, we introduce MS4, lightweight modifications to S4D via a linear input projection and channel-mixing mechanism, and MS4N, a normalized variant that stabilizes state dynamics with negligible overhead. Evaluated on 59 datasets across MONSTER (up to 60 million samples, 50K timesteps, 82 classes) and the UEA benchmark, against 15 baselines, MS4 and MS4N consistently outperform Mamba-based models while remaining more efficient, and MS4N matches or surpasses competing deep learning models that are roughly 2x and 10x larger in parameters. These results position lightweight structured SSMs as a compelling alternative to scaling complexity for TSC.

Hassan Saadatmand, Geoffrey I. Webb, Hamid Rezatofighi +1

Monash University, Melbourne, Victoria, Australia

Time-Series Classification State Space Models Mamba-Based Models Sequence Modeling Time Series

Ask PDF