Any-Scale Laplace Estimator (ASLE)

• Any-Scale Laplace Estimator (ASLE) is a framework for constructing Laplace-based estimators adaptable to various scales, applied in inertial localization, linear state estimation, and privacy mechanisms.
• It employs dual-task neural modules, randomized Kalman filter ensembles, or unbiased postprocessing operators to integrate measurements and accommodate variable scales.
• Empirical results show notable accuracy improvements, including up to a 40% reduction in MAE, confirming ASLE's robustness and versatility across different domains.

Searching arXiv for the cited papers to ground the article. I’m retrieving the relevant arXiv records for the ASLE-related papers. Any-Scale Laplace Estimator (ASLE) denotes a family of Laplace-based estimation constructions rather than a single canonical algorithm. In the arXiv literature considered here, the term is used explicitly for the dual-task neural module inside ReNiL, where it regresses relative displacement and aleatoric uncertainty from variable-length IMU sequences between Inertial Positioning Demand Points (IPDPs) (Wu et al., 8 Aug 2025). The same expression is also used as a descriptive label in technical syntheses of two other Laplace-centered estimators: a randomized ensemble of Kalman filters for linear state estimation with Laplace-corrupted measurements (Farokhi et al., 2016), and an unbiased estimator for functions of data released through the discrete Laplace mechanism together with exact postprocessing to continuous Laplace or Staircase outputs (Hillebrand et al., 7 May 2026). This suggests a shared organizing idea—estimation under Laplace models across arbitrary scales—while the concrete estimator, objective, and notion of “scale” differ by domain.

1. Terminological scope and domain-specific meanings

The phrase “Any-Scale Laplace Estimator” is explicit in ReNiL, where ASLE is the displacement-and-uncertainty regression module coupled to a motion-aware orientation filter and a Bayesian chaining process (Wu et al., 8 Aug 2025). By contrast, the 2016 state-estimation paper does not explicitly name its randomized estimator “ASLE,” and the 2026 privacy paper likewise does not explicitly use the term; in both cases, “ASLE” is an apt descriptive label rather than the paper’s original nomenclature (Farokhi et al., 2016, Hillebrand et al., 7 May 2026).

Context	Estimator role	Meaning of “any-scale”
ReNiL	Dual-task network for 2D displacement and Laplace scale regression	Any temporal length between successive IPDPs
Laplace-noisy linear systems	Randomized Kalman-filter ensemble using latent scale mixtures	Any Laplace scale $b$
Discrete Laplace privacy	Unbiased estimator and postprocessing framework	Any privacy scale $(\epsilon, S)$ and mechanism family

A common misconception is to treat ASLE as a standardized estimator with a single mathematical form. The literature instead supports three non-equivalent constructions. In ReNiL, ASLE is a learned regressor over IMU streams; in linear state estimation, it is a Monte Carlo approximation of the conditional mean via parallel Kalman filters; in privacy, it is an unbiased postprocessing operator built from discrete second differences. The term is therefore best understood as a cross-domain label for estimators that exploit Laplace structure while remaining operational across a tunable notion of scale.

2. ASLE in ReNiL: demand-driven inertial localization

In ReNiL, ASLE is the dual-task deep neural module that regresses relative displacement and aleatoric uncertainty from IMU sequences of any scale, meaning any temporal length between successive IPDPs $T_d = \{t_0,t_1,\ldots,t_n\}$ (Wu et al., 8 Aug 2025). For the IMU segment from $t_{n-1}$ to $t_n$, ASLE consumes an orientation-aligned accelerometer and gyroscope sequence of variable length $T$ and produces a 2D displacement $\Delta \hat p_{t_n}$ and a 2D Laplace scale vector $\hat b_{t_n}$. The aligned input is obtained by rotating raw IMU data into the navigation frame with the motion-aware orientation filter:

$$X^g = q^{-1} \otimes X \otimes q,\qquad (A^g,G^g,M^g)=q^{-1}\otimes(A,G,M)\otimes q.$$

The probabilistic output is modeled as

$$\Delta p_{t_n} \sim \mathrm{Laplace}(\Delta \hat p_{t_n}, \hat b_{t_n}),\qquad \Delta \hat p_{t_n}, \hat b_{t_n}\in\mathbb{R}^2,$$

with

$(\epsilon, S)$0

The paper emphasizes a Laplace parameterization for Euclidean-homogeneous uncertainty, so that the uncertainty has the same units as displacement and can be fused with other Euclidean sources such as GNSS.

The “any-scale” property is implemented through patching, contextual feature construction, and scale pooling. Aligned accelerations and angular rates are concatenated into $(\epsilon, S)$1, then patched as $(\epsilon, S)$2 with

$(\epsilon, S)$3

using zero-padding on the last patch when needed. In the reported implementation, $(\epsilon, S)$4 at $(\epsilon, S)$5 Hz, corresponding to $(\epsilon, S)$6 s per patch. This design allows inference cadence such as $(\epsilon, S)$7 s, $(\epsilon, S)$8 s, $(\epsilon, S)$9 s, or $T_d = \{t_0,t_1,\ldots,t_n\}$0 s to match application needs rather than a fixed sliding-window schedule.

ASLE is explicitly dual-task. Its supervised component is Bayesian regression under a Laplace likelihood, while its self-supervised component enforces context consistency between clean and augmented inputs at the feature-map level. A plausible implication is that ReNiL treats temporal-scale flexibility and uncertainty estimation as coupled design constraints rather than separate post hoc additions.

3. Probabilistic formulation, architecture, and training mechanics

ReNiL trains ASLE by maximum likelihood over a Laplace displacement model, but stabilizes learning across varying durations by re-parameterizing displacement through average speed $T_d = \{t_0,t_1,\ldots,t_n\}$1 over duration $T_d = \{t_0,t_1,\ldots,t_n\}$2:

$T_d = \{t_0,t_1,\ldots,t_n\}$3

ASLE therefore regresses the average-speed distribution $T_d = \{t_0,t_1,\ldots,t_n\}$4, from which displacement parameters are reconstructed. The paper gives the Laplace negative log-likelihood and a simplified speed-domain form:

$T_d = \{t_0,t_1,\ldots,t_n\}$5

and

$T_d = \{t_0,t_1,\ldots,t_n\}$6

The denominator equals $T_d = \{t_0,t_1,\ldots,t_n\}$7, and positivity is enforced by predicting $T_d = \{t_0,t_1,\ldots,t_n\}$8 and exponentiating it. The full training objective is

$T_d = \{t_0,t_1,\ldots,t_n\}$9

The architecture comprises a 2-layer TCN compressor with ReLU and group normalization, a 1D-ResNet18-like per-patch feature extractor in which GN replaces BN, a 2D residual contextual builder, and a PCO scale-pooling module. The reported implementation uses: 200 Hz IMU; patch length $t_{n-1}$0; embedding channels $t_{n-1}$1 with kernel size $t_{n-1}$2, stride $t_{n-1}$3, GN size $t_{n-1}$4; four 1D residual layers with output channels $t_{n-1}$5, $t_{n-1}$6, $t_{n-1}$7, $t_{n-1}$8; two 2D residual layers with channels $t_{n-1}$9 and kernel $t_n$0; and an output head FC-1024 $t_n$1 ReLU $t_n$2 dropout $t_n$3 vector of length $t_n$4, namely $t_n$5 (Wu et al., 8 Aug 2025).

The patch-based self-supervision augments the input with partial masking, quaternion constant bias interference, Gaussian noise, heading rotation, and abnormal protrusions. The mean-square error between contextual feature maps of clean and augmented inputs enforces robust context learning. Training uses Adam with initial learning rate $t_n$6, ReduceLROnPlateau with factor $t_n$7 and patience $t_n$8, minimum learning rate $t_n$9, batch size $T$0, and $T$1 epochs in PyTorch 2.5.1.

The motion-aware orientation filter is part of the same pipeline. It uses adaptive complementary weighting,

$T$2

with accelerometer and magnetometer weights defined over sensor windows. The accelerometer window uses one walking cycle, $T$3, and the magnetometer window starts only when $T$4, with $T$5 m in experiments. The reported hyperparameters are $T$6, $T$7, and $T$8.

4. Bayesian chaining, uncertainty propagation, and empirical behavior in ReNiL

ReNiL does not stop at per-segment regression. It links successive IPDPs with a Bayesian inference chain whose state transition is

$T$9

where $\Delta \hat p_{t_n}$0 is the ASLE displacement output and $\Delta \hat p_{t_n}$1 is Laplace process noise in 2D (Wu et al., 8 Aug 2025). For pure inertial chaining, Laplace noise is treated as a Gaussian-exponential mixture:

$\Delta \hat p_{t_n}$2

and conditionally

$\Delta \hat p_{t_n}$3

With external observations $\Delta \hat p_{t_n}$4, the observation model is

$\Delta \hat p_{t_n}$5

and the paper uses a Rao-Blackwellized Kalman/Gibbs procedure with

$\Delta \hat p_{t_n}$6

The reported evaluations use RoNIN-ds and WUDataset, with baselines PDR, RoNIN-ResNet18, TLIO, CTIN, iMoT, and fixed-window ASLE-ns variants. On RoNIN-ds with seen subjects, ASLE-10s attains MAE $\Delta \hat p_{t_n}$7, ADE $\Delta \hat p_{t_n}$8, and HE $\Delta \hat p_{t_n}$9; on unseen subjects, ASLE-20s attains MAE $\hat b_{t_n}$0, ADE $\hat b_{t_n}$1, and HE $\hat b_{t_n}$2. On WUDataset with seen subjects, ASLE-20s attains MAE $\hat b_{t_n}$3 and ADE $\hat b_{t_n}$4, while ASLE-5s attains HE $\hat b_{t_n}$5; on unseen subjects, ASLE-10s attains MAE $\hat b_{t_n}$6 and HE $\hat b_{t_n}$7, while ASLE-20s attains ADE $\hat b_{t_n}$8. The paper further reports that Laplace and Gaussian parameterizations are similar at short scales, but Laplace yields lower MAE at longer scales, and that any-scale support surpasses fixed-window models as the time scale lengthens, with MAE reductions up to approximately $\hat b_{t_n}$9 on WUDataset.

Uncertainty calibration is described as reasonably calibrated but not exact: confidence interval coverage is $$X^g = q^{-1} \otimes X \otimes q,\qquad (A^g,G^g,M^g)=q^{-1}\otimes(A,G,M)\otimes q.$$0, $$X^g = q^{-1} \otimes X \otimes q,\qquad (A^g,G^g,M^g)=q^{-1}\otimes(A,G,M)\otimes q.$$1, and $$X^g = q^{-1} \otimes X \otimes q,\qquad (A^g,G^g,M^g)=q^{-1}\otimes(A,G,M)\otimes q.$$2, with conservative behavior at small time scales and deviations for rare large-error regimes attributed to data sparsity at long scales. The paper also states that it does not report analytic gradients or a formal proof of optimality for the Laplace mixture filter; behavior is empirically validated.

5. Randomized Kalman-filter ASLE for Laplace-corrupted measurements

For linear discrete-time systems with Gaussian process noise and Laplace measurement noise,

$$X^g = q^{-1} \otimes X \otimes q,\qquad (A^g,G^g,M^g)=q^{-1}\otimes(A,G,M)\otimes q.$$3

the 2016 estimator approximates the optimal least mean square error estimate

$$X^g = q^{-1} \otimes X \otimes q,\qquad (A^g,G^g,M^g)=q^{-1}\otimes(A,G,M)\otimes q.$$4

by exploiting an exact scale-mixture representation of Laplace noise (Farokhi et al., 2016). Each measurement component satisfies

$$X^g = q^{-1} \otimes X \otimes q,\qquad (A^g,G^g,M^g)=q^{-1}\otimes(A,G,M)\otimes q.$$5

with variance $$X^g = q^{-1} \otimes X \otimes q,\qquad (A^g,G^g,M^g)=q^{-1}\otimes(A,G,M)\otimes q.$$6. The key identity is that if $$X^g = q^{-1} \otimes X \otimes q,\qquad (A^g,G^g,M^g)=q^{-1}\otimes(A,G,M)\otimes q.$$7 and $$X^g = q^{-1} \otimes X \otimes q,\qquad (A^g,G^g,M^g)=q^{-1}\otimes(A,G,M)\otimes q.$$8, then marginally $$X^g = q^{-1} \otimes X \otimes q,\qquad (A^g,G^g,M^g)=q^{-1}\otimes(A,G,M)\otimes q.$$9.

Conditioning on the latent scales converts the problem to Gaussian state estimation with random measurement covariance. For each realization $$\Delta p_{t_n} \sim \mathrm{Laplace}(\Delta \hat p_{t_n}, \hat b_{t_n}),\qquad \Delta \hat p_{t_n}, \hat b_{t_n}\in\mathbb{R}^2,$$0, the method samples

$$\Delta p_{t_n} \sim \mathrm{Laplace}(\Delta \hat p_{t_n}, \hat b_{t_n}),\qquad \Delta \hat p_{t_n}, \hat b_{t_n}\in\mathbb{R}^2,$$1

runs a standard Kalman filter with measurement covariance $$\Delta p_{t_n} \sim \mathrm{Laplace}(\Delta \hat p_{t_n}, \hat b_{t_n}),\qquad \Delta \hat p_{t_n}, \hat b_{t_n}\in\mathbb{R}^2,$$2, and aggregates the realization-specific estimates by simple averaging:

$$\Delta p_{t_n} \sim \mathrm{Laplace}(\Delta \hat p_{t_n}, \hat b_{t_n}),\qquad \Delta \hat p_{t_n}, \hat b_{t_n}\in\mathbb{R}^2,$$3

The paper also proposes heuristic schemes for generating samples from $$\Delta p_{t_n} \sim \mathrm{Laplace}(\Delta \hat p_{t_n}, \hat b_{t_n}),\qquad \Delta \hat p_{t_n}, \hat b_{t_n}\in\mathbb{R}^2,$$4, including a memory-less approximation, moving-horizon aggregation, and a Gaussian approximation of $$\Delta p_{t_n} \sim \mathrm{Laplace}(\Delta \hat p_{t_n}, \hat b_{t_n}),\qquad \Delta \hat p_{t_n}, \hat b_{t_n}\in\mathbb{R}^2,$$5.

In this setting, the “any-scale” property refers to the fact that the method is valid for any Laplace scale $$\Delta p_{t_n} \sim \mathrm{Laplace}(\Delta \hat p_{t_n}, \hat b_{t_n}),\qquad \Delta \hat p_{t_n}, \hat b_{t_n}\in\mathbb{R}^2,$$6, because $$\Delta p_{t_n} \sim \mathrm{Laplace}(\Delta \hat p_{t_n}, \hat b_{t_n}),\qquad \Delta \hat p_{t_n}, \hat b_{t_n}\in\mathbb{R}^2,$$7 only enters through the Rayleigh sampling. The paper proves a Chebyshev-based probability bound: there exists $$\Delta p_{t_n} \sim \mathrm{Laplace}(\Delta \hat p_{t_n}, \hat b_{t_n}),\qquad \Delta \hat p_{t_n}, \hat b_{t_n}\in\mathbb{R}^2,$$8 such that

$$\Delta p_{t_n} \sim \mathrm{Laplace}(\Delta \hat p_{t_n}, \hat b_{t_n}),\qquad \Delta \hat p_{t_n}, \hat b_{t_n}\in\mathbb{R}^2,$$9

provided

$(\epsilon, S)$00

Under stability, a uniform finite constant exists for all $(\epsilon, S)$01. In the numerical example reported in the paper, ASLE with $(\epsilon, S)$02 Kalman filters achieved the lowest MSE among the tested methods, outperforming the optimal linear estimator, the MAP estimator, and a particle filter with $(\epsilon, S)$03 particles.

6. Discrete Laplace unbiased estimation, exact postprocessing, and limitations

In the privacy setting, the discrete Laplace mechanism releases $(\epsilon, S)$04 with independent coordinate noise

$(\epsilon, S)$05

for unit or bounded $(\epsilon, S)$06-sensitivity (Hillebrand et al., 7 May 2026). For any function $(\epsilon, S)$07 satisfying the paper’s subexponential-growth condition, the estimator

$(\epsilon, S)$08

with

$(\epsilon, S)$09

is unbiased:

$(\epsilon, S)$10

Equivalently,

$(\epsilon, S)$11

where $(\epsilon, S)$12 is the discrete second difference in coordinate $(\epsilon, S)$13.

The same paper gives exact input-independent postprocessing from discrete Laplace outputs to continuous Laplace or Staircase distributions with the same privacy parameters. For continuous Laplace, the added random variable $(\epsilon, S)$14 supported on $(\epsilon, S)$15 has density

$(\epsilon, S)$16

yielding

$(\epsilon, S)$17

For the Staircase mechanism with shape $(\epsilon, S)$18, the postprocessing density is

$(\epsilon, S)$19

The paper therefore frames an “any-scale” estimator as one calibrated by $(\epsilon, S)$20 and $(\epsilon, S)$21, usable with discrete Laplace directly and transferable to continuous Laplace or Staircase by postprocessing.

The limitations are explicit. Worst-case runtime for a black-box $(\epsilon, S)$22 is $(\epsilon, S)$23, though structured classes admit polynomial-time computation. Unbiased estimators can have larger variance than naive plug-in estimators and, for some monotone indicator functions, variance can grow exponentially in $(\epsilon, S)$24. In ReNiL, broader applicability beyond pedestrian localization is implied but not explicitly evaluated; in the linear-systems estimator, the randomized approximation accuracy depends on the number of parallel filters; and in the privacy estimator, calibration errors in sensitivity $(\epsilon, S)$25 alter $(\epsilon, S)$26 and thus the operator coefficients. Taken together, these results indicate that ASLE is best regarded as a Laplace-structured estimation paradigm whose operational meaning depends on whether “scale” denotes temporal extent, distribution scale, or privacy calibration.