-
Using Gaussian Processes to Design Dynamic Experiments for Black-Box Model Discrimination under Uncertainty
Authors:
Simon Olofsson,
Eduardo S. Schultz,
Adel Mhamdi,
Alexander Mitsos,
Marc Peter Deisenroth,
Ruth Misener
Abstract:
Diverse domains of science and engineering use parameterised mechanistic models. Engineers and scientists can often hypothesise several rival models to explain a specific process or phenomenon. Consider a model discrimination setting where we wish to find the best mechanistic, dynamic model candidate and the best model parameter estimates. Typically, several rival mechanistic models can explain th…
▽ More
Diverse domains of science and engineering use parameterised mechanistic models. Engineers and scientists can often hypothesise several rival models to explain a specific process or phenomenon. Consider a model discrimination setting where we wish to find the best mechanistic, dynamic model candidate and the best model parameter estimates. Typically, several rival mechanistic models can explain the available data, so design of dynamic experiments for model discrimination helps optimally collect additional data by finding experimental settings that maximise model prediction divergence. We argue there are two main approaches in the literature for solving the optimal design problem: (i) the analytical approach, using linear and Gaussian approximations to find closed-form expressions for the design objective, and (ii) the data-driven approach, which often relies on computationally intensive Monte Carlo techniques. Olofsson et al. (ICML 35, 2018) introduced Gaussian process (GP) surrogate models to hybridise the analytical and data-driven approaches, which allowed for computationally efficient design of experiments for discriminating between black-box models. In this study, we demonstrate that we can extend existing methods for optimal design of dynamic experiments to incorporate a wider range of problem uncertainty. We also extend the Olofsson et al. (2018) method of using GP surrogate models for discriminating between dynamic black-box models. We evaluate our approach on a well-known case study from literature, and explore the consequences of using GP surrogates to approximate gradient-based methods.
△ Less
Submitted 31 October, 2021; v1 submitted 7 February, 2021;
originally announced February 2021.
-
GPdoemd: a Python package for design of experiments for model discrimination
Authors:
Simon Olofsson,
Lukas Hebing,
Sebastian Niedenführ,
Marc Peter Deisenroth,
Ruth Misener
Abstract:
Model discrimination identifies a mathematical model that usefully explains and predicts a given system's behaviour. Researchers will often have several models, i.e. hypotheses, about an underlying system mechanism, but insufficient experimental data to discriminate between the models, i.e. discard inaccurate models. Given rival mathematical models and an initial experimental data set, optimal des…
▽ More
Model discrimination identifies a mathematical model that usefully explains and predicts a given system's behaviour. Researchers will often have several models, i.e. hypotheses, about an underlying system mechanism, but insufficient experimental data to discriminate between the models, i.e. discard inaccurate models. Given rival mathematical models and an initial experimental data set, optimal design of experiments suggests maximally informative experimental observations that maximise a design criterion weighted by prediction uncertainty. The model uncertainty requires gradients, which may not be readily available for black-box models. This paper (i) proposes a new design criterion using the Jensen-Rényi divergence, and (ii) develops a novel method replacing black-box models with Gaussian process surrogates. Using the surrogates, we marginalise out the model parameters with approximate inference. Results show these contributions working well for both classical and new test instances. We also (iii) introduce and discuss GPdoemd, the open-source implementation of the Gaussian process surrogate method.
△ Less
Submitted 8 March, 2019; v1 submitted 5 October, 2018;
originally announced October 2018.
-
Design of Experiments for Model Discrimination Hybridising Analytical and Data-Driven Approaches
Authors:
Simon Olofsson,
Marc Peter Deisenroth,
Ruth Misener
Abstract:
Healthcare companies must submit pharmaceutical drugs or medical devices to regulatory bodies before marketing new technology. Regulatory bodies frequently require transparent and interpretable computational modelling to justify a new healthcare technology, but researchers may have several competing models for a biological system and too little data to discriminate between the models. In design of…
▽ More
Healthcare companies must submit pharmaceutical drugs or medical devices to regulatory bodies before marketing new technology. Regulatory bodies frequently require transparent and interpretable computational modelling to justify a new healthcare technology, but researchers may have several competing models for a biological system and too little data to discriminate between the models. In design of experiments for model discrimination, the goal is to design maximally informative physical experiments in order to discriminate between rival predictive models. Prior work has focused either on analytical approaches, which cannot manage all functions, or on data-driven approaches, which may have computational difficulties or lack interpretable marginal predictive distributions. We develop a methodology introducing Gaussian process surrogates in lieu of the original mechanistic models. We thereby extend existing design and model discrimination methods developed for analytical models to cases of non-analytical models in a computationally efficient manner.
△ Less
Submitted 31 May, 2018; v1 submitted 12 February, 2018;
originally announced February 2018.
-
The mid-infrared extinction in molecular clouds Case study of B 335
Authors:
Sven Olofsson,
Göran Olofsson
Abstract:
The purpose of the investigation is to probe the dust properties inside a molecular cloud, how particle grow and how the presence of ice coatings may change the overall shape of the extinction curve.
Field stars can be used to probe the cloud extinction. By combining multi-colour photometry and IR spectroscopy the spectral class of the star can be determined as can the extinction curve.
We det…
▽ More
The purpose of the investigation is to probe the dust properties inside a molecular cloud, how particle grow and how the presence of ice coatings may change the overall shape of the extinction curve.
Field stars can be used to probe the cloud extinction. By combining multi-colour photometry and IR spectroscopy the spectral class of the star can be determined as can the extinction curve.
We determine the reddening curve from 0.35 to 24 μm. The water ice band at 3.1 μm is weaker (τ(3.1) = 0.4) than expected from the cloud extinction (AV \approx 10 for the sightline to the most obscured star). On the other hand, the CO ice band at 4.7 μm is strong (τ(4.67) = 0.7) and indicates, that the mass column density of frozen CO is about the same as that of water ice. We show that the reddening curves for the two background stars, for which the silicate band has been measured, can be accurately modelled from the UV to 24 μm. These models only include graphite and silicate grains. No need for any additional major grain component to explain the slow decline of the reddening curve beyond the K band. The dust model for the dense part of the cloud has more large grains than for the rim. We propose that the well established shallow reddening curve beyond the K band has two different explanations: larger graphite grains in dense regions and relatively small grains in the diffuse ISM, giving rise to substantially less extinction beyond the K band than previously thought.
For the sight line towards the most obscured star, we derive the relation AKs = 0.97 \cdot E(J - Ks), and assuming that all silicon is bound in silicates, N(2 H2 +H) \approx 1.5 \times 10^{21} \cdot AV \approx 9 \times 10^{21} \cdot AKs. For the rim of the cloud we get AKs = 0.51 \cdot E(J - Ks), which is close to recent determinations for the diffuse ISM. The corresponding gas column density is N(2 H2 +H) \approx 2.3 \times 10^{21} \cdot AV \approx 3 \times 10^{22} \cdot AKs.
△ Less
Submitted 1 December, 2011;
originally announced December 2011.
-
The extinction law for molecular clouds. Case study of B 335
Authors:
Sven Olofsson,
Göran Olofsson
Abstract:
We determine the extinction curve from the UV to the near-IR for molecular clouds and investigate whether current models can adequately explain this wavelength dependence of the extinction. The aim is also to interpret the extinction in terms of H2 column density.
We applied five different methods, including a new method for simultaneously determining the reddening law and the classification of…
▽ More
We determine the extinction curve from the UV to the near-IR for molecular clouds and investigate whether current models can adequately explain this wavelength dependence of the extinction. The aim is also to interpret the extinction in terms of H2 column density.
We applied five different methods, including a new method for simultaneously determining the reddening law and the classification of the background stars. Our method is based on multicolour observations and a grid of model atmospheres.
We confirm that the extinction law can be adequately described by a single parameter, RV (the selective to absolute extinction), in accordance with earlier findings. The RV value for B 335 is RV = 4.8. The reddening curve can be accurately reproduced by model calculations. By assuming that all the silicon is bound in silicate grains, we can interpret the reddening in terms of column density, NH = 4.4 (\pm0.5) \times 1021 EI-Ks cm-2, corresponding to NH = 2.3 (\pm0.2) \times 1021 \cdot AV cm-2, close to that of the diffuse ISM, (1.8-2.2) \times 1021 cm-2 . We show that the density of the B 335 globule outer shells can be modelled as an evolved Ebert-Bonnor gas sphere with ρ \propto r-2, and estimate the mass of this globule to 2.5 Msun
△ Less
Submitted 1 December, 2011;
originally announced December 2011.
-
A new method of determining distances to dark globules. The distance to B 335
Authors:
Sven Olofsson,
Göran Olofsson
Abstract:
We observed a test globule, B 335 in U, B, g, r, and I, and together with the 2MASS survey, this data set gives a well-defined spectral energy distribution (SED) of a large number of stars. The SED of each star depends on the interstellar extinction, the distance to the star, and its intrinsic SED. The method is based on the use of stellar atmospheric models to represent the intrinsic SEDs of the…
▽ More
We observed a test globule, B 335 in U, B, g, r, and I, and together with the 2MASS survey, this data set gives a well-defined spectral energy distribution (SED) of a large number of stars. The SED of each star depends on the interstellar extinction, the distance to the star, and its intrinsic SED. The method is based on the use of stellar atmospheric models to represent the intrinsic SEDs of the stars. Formally, it is then possible to determine the spectral class of each star and thereby its distance. For some of the stars we have optical spectra, allowing us to compare the photometric classification to the spectrometric.
We can identify one star at the front side of the globule. It has a photometric distance of 90 pc. The closest star behind the B 335 globule has a distance of only \approx 120 pc and we therefore determine the distance to B 335 as 90-120 pc. Our deep U image shows a relatively bright south-western rim of the globule, and we investigate whether it might be due to a local enhancement of the radiation field. A candidate source, located 1.5 arcminutes outside our field, would be the field star, HD 184982. This star has an entry in the Hipparcos Catalogue and its distance is 140-200 pc. However, we come to the conclusion that the bright SW rim is more likely due to the wing of the point-spread-function (PSF) of this star.
△ Less
Submitted 1 December, 2011;
originally announced December 2011.
-
ISOCAM observations of the L1551 star formation region
Authors:
M. Gålfalk,
G. Olofsson,
A. A. Kaas,
S. Olofsson,
S. Bontemps,
L. Nordh,
A. Abergel,
P. Andre,
F. Boulanger,
M. Burgdorf,
M. M. Casali,
C. J. Cesarsky,
J. Davies,
E. Falgarone,
T. Montmerle,
M. Perault,
P. Persi,
T. Prusti,
J. L. Puget,
F. Sibille
Abstract:
The results of a deep mid-IR ISOCAM survey of the L1551 dark molecular cloud are presented. The aim of this survey is a search for new YSO (Young Stellar Object) candidates, using two broad-band filters centred at 6.7 and 14.3 micron. Although two regions close to the centre of L1551 had to be avoided due to saturation problems, 96 sources were detected in total (76 sources at 6.7 micron and 44…
▽ More
The results of a deep mid-IR ISOCAM survey of the L1551 dark molecular cloud are presented. The aim of this survey is a search for new YSO (Young Stellar Object) candidates, using two broad-band filters centred at 6.7 and 14.3 micron. Although two regions close to the centre of L1551 had to be avoided due to saturation problems, 96 sources were detected in total (76 sources at 6.7 micron and 44 sources at 14.3 micron). Using the 24 sources detected in both filters, 14 were found to have intrinsic mid-IR excess at 14.3 micron and were therefore classified as YSO candidates. Using additional observations in B, V, I, J, H and K obtained from the ground, most candidates detected at these wavelengths were confirmed to have mid-IR excess at 6.7 micron as well, and three additional YSO candidates were found. Prior to this survey only three YSOs were known in the observed region (avoiding L1551 IRS5/NE and HL/XZ Tau). This survey reveals 15 new YSO candidates, although several of these are uncertain due to their extended nature either in the mid-IR or in the optical/near-IR observations. Two of the sources with mid-IR excess are previously known YSOs, one is a brown dwarf MHO 5 and the other is the well known T Tauri star HH30, consisting of an outflow and an optically thick disk seen edge on.
△ Less
Submitted 4 May, 2004;
originally announced May 2004.
