-
Computational philosophy of science
Authors:
Michał J. Gajda
Abstract:
Philosophy of science attempts to describe all parts of the scientific process in a general way in order to facilitate the description, execution and improvements of this process.
So far, all proposed philosophies have only covered existing processes and disciplines partially and imperfectly. In particular logical approaches have always received a lot of attention due to attempts to fundamentall…
▽ More
Philosophy of science attempts to describe all parts of the scientific process in a general way in order to facilitate the description, execution and improvements of this process.
So far, all proposed philosophies have only covered existing processes and disciplines partially and imperfectly. In particular logical approaches have always received a lot of attention due to attempts to fundamentally address issues with the definition of science as a discipline with reductionist theories.
We propose a new way to approach the problem from the perspective of computational complexity and argue why this approach may be better than previous propositions based on pure logic and mathematics.
△ Less
Submitted 4 February, 2023;
originally announced February 2023.
-
Data accounting and error counting
Authors:
Michał J. Gajda
Abstract:
Can we infer sources of errors from outputs of the complex data analytics software?
Bidirectional programming promises that we can reverse flow of software, and translate corrections of output into corrections of either input or data analysis.
This allows us to achieve holy grail of automated approaches to debugging, risk reporting and large scale distributed error tracking.
Since processing…
▽ More
Can we infer sources of errors from outputs of the complex data analytics software?
Bidirectional programming promises that we can reverse flow of software, and translate corrections of output into corrections of either input or data analysis.
This allows us to achieve holy grail of automated approaches to debugging, risk reporting and large scale distributed error tracking.
Since processing of risk reports and data analysis pipelines can be frequently expressed using a
sequence relational algebra operations, we propose a replacement of this traditional approach with a data
summarization algebra that helps to determine an impact of errors. It works by defining data analysis of
a necessarily complete summarization of a dataset, possibly in multiple ways along multiple dimensions.
We also present a description to better communicate how the complete summarizations of the input
data may facilitates easier debugging and more efficient development of analysis pipelines.
This approach can also be described as an generalization of axiomatic theories of accounting into
data analytics, thus dubbed data accounting.
We also propose formal properties that allow for transparent assertions about impact of individual
records on the aggregated data and ease debugging by allowing to find minimal changes that change
behaviour of data analysis on per-record basis.
△ Less
Submitted 29 January, 2023;
originally announced January 2023.
-
Non-Gaussian Measures in Infinite Dimensional Spaces: the Gamma-Grey Noise
Authors:
Luisa Beghin,
Lorenzo Cristofaro,
Janusz Gajda
Abstract:
In the context of non-Gaussian analysis, Schneider [27] introduced grey noise measures, built upon Mittag-Leffler functions; analogously, grey Brownian motion and its generalizations were constructed (see, for example, [25], [6], [7], [8]). In this paper, we construct and study a new non-Gaussian measure, by means of the incomplete-gamma function (exploiting its complete monotonicity). We label th…
▽ More
In the context of non-Gaussian analysis, Schneider [27] introduced grey noise measures, built upon Mittag-Leffler functions; analogously, grey Brownian motion and its generalizations were constructed (see, for example, [25], [6], [7], [8]). In this paper, we construct and study a new non-Gaussian measure, by means of the incomplete-gamma function (exploiting its complete monotonicity). We label this measure Gamma-grey noise and we prove, for it, the existence of Appell system. The related generalized processes, in the infinite dimensional setting, are also defined and, through the use of the Riemann-Liouville fractional operators, the (possibly tempered) Gamma-grey Brownian motion is consequently introduced. A number of different characterizations of these processes are also provided, together with the integro-differential equation satisfied by their transition densities. They allow to model anomalous diffusions, mimicking the procedures of classical stochastic calculus.
△ Less
Submitted 27 July, 2022;
originally announced July 2022.
-
Tempered relaxation equation and related generalized stable processes
Authors:
Luisa Beghin,
Janusz Gajda
Abstract:
Fractional relaxation equations, as well as relaxation functions time-changed by independent stochastic processes have been widely studied (see, for example, \cite{MAI}, \cite{STAW} and \cite{GAR}). We start here by proving that the upper-incomplete Gamma function satisfies the tempered-relaxation equation (of index $ρ\in (0,1)$); thanks to this explicit form of the solution, we can then derive it…
▽ More
Fractional relaxation equations, as well as relaxation functions time-changed by independent stochastic processes have been widely studied (see, for example, \cite{MAI}, \cite{STAW} and \cite{GAR}). We start here by proving that the upper-incomplete Gamma function satisfies the tempered-relaxation equation (of index $ρ\in (0,1)$); thanks to this explicit form of the solution, we can then derive its spectral distribution, which extends the stable law. Accordingly, we define a new class of selfsimilar processes (by means of the $n$-times Laplace transform of its density) which is indexed by the parameter $ρ$: in the special case where $ρ=1$, it reduces to the stable subordinator. Therefore the parameter $ρ$ can be seen as a measure of the local deviation from the temporal dependence structure displayed in the standard stable case.
△ Less
Submitted 1 September, 2020; v1 submitted 27 December, 2019;
originally announced December 2019.
-
Integro-differential equations linked to compound birth processes with infinitely divisible addends
Authors:
L. Beghin,
J. Gajda,
A. Maheshwari
Abstract:
Stochastic modelling of fatigue (and other material's deterioration), as well as of cumulative damage in risk theory, are often based on compound sums of independent random variables, where the number of addends is represented by an independent counting process. We consider here a cumulative model where, instead of a renewal process (as in the Poisson case), a linear birth (or Yule) process is use…
▽ More
Stochastic modelling of fatigue (and other material's deterioration), as well as of cumulative damage in risk theory, are often based on compound sums of independent random variables, where the number of addends is represented by an independent counting process. We consider here a cumulative model where, instead of a renewal process (as in the Poisson case), a linear birth (or Yule) process is used. This corresponds to the assumption that the frequency of \textquotedblleft damage" increments accelerates according to the increasing number of \textquotedblleft damages". We start from the partial differential equation satisfied by its transition density, in the case of exponentially distributed addends, and then we generalize it by introducing a space-derivative of convolution type (i.e. defined in terms of the Laplace exponent of a subordinator). Then we are concerned with the solution of integro-differential equations, which, in particular cases, reduce to fractional ones. Correspondingly, we analyze the related cumulative jump processes under a general infinitely divisible distribution of the (positive) jumps. Some special cases (such as the stable, tempered stable, gamma and Poisson) are presented.
△ Less
Submitted 28 November, 2019;
originally announced November 2019.
-
Large deviations of time-averaged statistics for Gaussian processes
Authors:
J. Gajda,
A. Wylomanska,
H. Kantz,
A. V. Chechkin,
G. Sikora
Abstract:
In this paper we study the large deviations of time averaged mean square displacement (TAMSD) for Gaussian processes. The theory of large deviations is related to the exponential decay of probabilities of large fluctuations in random systems. From the mathematical point of view a given statistics satisfies the large deviation principle, if the probability that it belongs to a certain range decreas…
▽ More
In this paper we study the large deviations of time averaged mean square displacement (TAMSD) for Gaussian processes. The theory of large deviations is related to the exponential decay of probabilities of large fluctuations in random systems. From the mathematical point of view a given statistics satisfies the large deviation principle, if the probability that it belongs to a certain range decreases exponentially. The TAMSD is one of the main statistics used in the problem of anomalous diffusion detection. Applying the theory of generalized chi-squared distribution and sub-gamma random variables we prove the upper bound for large deviations of TAMSD for Gaussian processes. As a special case we consider fractional Brownian motion, one of the most popular models of anomalous diffusion. Moreover, we derive the upper bound for large deviations of the estimator for the anomalous diffusion exponent.
△ Less
Submitted 27 November, 2018;
originally announced November 2018.
-
Probabilistic properties of detrended fluctuation analysis for Gaussian processes
Authors:
G. Sikora,
M. Hoell,
A. Wylomanska,
J. Gajda,
A. V. Chechkin,
H. Kantz
Abstract:
The detrended fluctuation analysis (DFA) is one of the most widely used tools for the detection of long-range correlations in time series. Although DFA has found many interesting applications and has been shown as one of the best performing detrending methods, its probabilistic foundations are still unclear. In this paper we study probabilistic properties of DFA for Gaussian processes. The main at…
▽ More
The detrended fluctuation analysis (DFA) is one of the most widely used tools for the detection of long-range correlations in time series. Although DFA has found many interesting applications and has been shown as one of the best performing detrending methods, its probabilistic foundations are still unclear. In this paper we study probabilistic properties of DFA for Gaussian processes. The main attention is paid to the distribution of the squared error sum of the detrended process. This allows us to find the expected value and the variance of the fluctuation function of DFA for a Gaussian process of general form. The results obtained can serve as a starting point for analyzing the statistical properties of the DFA-based estimators for the fluctuation and correlation parameters. The obtained theoretical formulas are supported by numerical simulations of particular Gaussian processes possessing short-and long-memory behaviour.
△ Less
Submitted 27 November, 2018;
originally announced November 2018.
