-
Variance Minimization in the Wasserstein Space for Invariant Causal Prediction
Authors:
Guillaume Martinet,
Alexander Strzalkowski,
Barbara E. Engelhardt
Abstract:
Selecting powerful predictors for an outcome is a cornerstone task for machine learning. However, some types of questions can only be answered by identifying the predictors that causally affect the outcome. A recent approach to this causal inference problem leverages the invariance property of a causal mechanism across differing experimental environments (Peters et al., 2016; Heinze-Deml et al., 2…
▽ More
Selecting powerful predictors for an outcome is a cornerstone task for machine learning. However, some types of questions can only be answered by identifying the predictors that causally affect the outcome. A recent approach to this causal inference problem leverages the invariance property of a causal mechanism across differing experimental environments (Peters et al., 2016; Heinze-Deml et al., 2018). This method, invariant causal prediction (ICP), has a substantial computational defect -- the runtime scales exponentially with the number of possible causal variables. In this work, we show that the approach taken in ICP may be reformulated as a series of nonparametric tests that scales linearly in the number of predictors. Each of these tests relies on the minimization of a novel loss function -- the Wasserstein variance -- that is derived from tools in optimal transport theory and is used to quantify distributional variability across environments. We prove under mild assumptions that our method is able to recover the set of identifiable direct causes, and we demonstrate in our experiments that it is competitive with other benchmark causal discovery algorithms.
△ Less
Submitted 27 February, 2022; v1 submitted 13 October, 2021;
originally announced October 2021.
-
A Balancing Weight Framework for Estimating the Causal Effect of General Treatments
Authors:
Guillaume Martinet
Abstract:
In observational studies, weighting methods that directly optimize the balance between treatment and covariates have received much attention lately; however these have mainly focused on binary treatments. Inspired by domain adaptation, we show that such methods can be actually reformulated as specific implementations of a discrepancy minimization problem aimed at tackling a shift of distribution f…
▽ More
In observational studies, weighting methods that directly optimize the balance between treatment and covariates have received much attention lately; however these have mainly focused on binary treatments. Inspired by domain adaptation, we show that such methods can be actually reformulated as specific implementations of a discrepancy minimization problem aimed at tackling a shift of distribution from observational to interventional data. More precisely, we introduce a new framework, Covariate Balance via Discrepancy Minimization (CBDM), that provably encompasses most of the existing balancing weight methods and formally extends them to treatments of arbitrary types (e.g., continuous or multivariate). We establish theoretical guarantees for our framework that both offer generalizations of properties known when the treatment is binary, and give a better grasp on what hyperparameters to choose in non-binary settings. Based on such insights, we propose a particular implementation of CBDM for estimating dose-response curves and demonstrate through experiments its competitive performance relative to other existing approaches for continuous treatments.
△ Less
Submitted 25 February, 2020;
originally announced February 2020.
-
Marginal Singularity, and the Benefits of Labels in Covariate-Shift
Authors:
Samory Kpotufe,
Guillaume Martinet
Abstract:
We present new minimax results that concisely capture the relative benefits of source and target labeled data, under covariate-shift. Namely, we show that the benefits of target labels are controlled by a transfer-exponent $γ$ that encodes how singular Q is locally w.r.t. P, and interestingly allows situations where transfer did not seem possible under previous insights. In fact, our new minimax a…
▽ More
We present new minimax results that concisely capture the relative benefits of source and target labeled data, under covariate-shift. Namely, we show that the benefits of target labels are controlled by a transfer-exponent $γ$ that encodes how singular Q is locally w.r.t. P, and interestingly allows situations where transfer did not seem possible under previous insights. In fact, our new minimax analysis - in terms of $γ$ - reveals a continuum of regimes ranging from situations where target labels have little benefit, to regimes where target labels dramatically improve classification. We then show that a recently proposed semi-supervised procedure can be extended to adapt to unknown $γ$, and therefore requests labels only when beneficial, while achieving minimax transfer rates.
△ Less
Submitted 16 June, 2020; v1 submitted 5 March, 2018;
originally announced March 2018.
-
Statistical universal branching ratios for cosmic ray dissociation, photodissociation, and dissociative recombination of the C(n=2-10), C(n=2-4)H and C3H2 neutral and cationic species
Authors:
M. Chabot,
T. Tuna,
K. Beroff,
T. Pino,
A. Le Padellec,
P. Desequelles,
G. Martinet,
V. O. Nguyen-Thi,
Y. Carpentier,
F. Le Petit,
E. Roueff,
V. Wakelam
Abstract:
Fragmentation branching ratios of electronically excited molecular species are of first importance for the modeling of gas phase interstellar chemistry. Despite experimental and theoretical efforts that have been done during the last two decades there is still a strong lack of detailed information on those quantities for many molecules such as Cn, CnH or C3H2. Our aim is to provide astrochemical d…
▽ More
Fragmentation branching ratios of electronically excited molecular species are of first importance for the modeling of gas phase interstellar chemistry. Despite experimental and theoretical efforts that have been done during the last two decades there is still a strong lack of detailed information on those quantities for many molecules such as Cn, CnH or C3H2. Our aim is to provide astrochemical databases with more realistic branching ratios for Cn (n=2 to 10), CnH (n=2 to 4), and C3H2 molecules that are electronically excited either by dissociative recombination, photodissociation, or cosmic ray processes, when no detailed calculations or measurements exist in literature. High velocity collision in an inverse kinematics scheme was used to measure the complete fragmentation pattern of electronically excited Cn (n=2 to 10), CnH (n=2 to 4), and C3H2 molecules. Branching ratios of dissociation where deduced from those experiments. The full set of branching ratios was used as a new input in chemical models and branching ratio modification effects observed in astrochemical networks that describe the dense cold Taurus Molecular Cloud-1 and the photon dominated Horse Head region. The comparison between the branching ratios obtained in this work and other types of experiments showed a good agreement. It was interpreted as the signature of a statistical behavior of the fragmentation. The branching ratios we obtained lead to an increase of the C3 production together with a larger dispersion of the daughter fragments. The introduction of these new values in the photon dominated region model of the Horse Head nebula increases the abundance of C3 and C3H, but reduces the abundances of the larger Cn and hydrocarbons at a visual extinction Av smaller than 4.
△ Less
Submitted 1 October, 2010;
originally announced October 2010.
-
Fragmentation branching ratios of highly excited hydrocarbon molecules CnH and their cations CnH+ (n<4)
Authors:
T. Tuna,
M. Chabot,
T. Pino,
P. Désesquelles,
A. Lepadellec,
G. . Martinet,
M. Barat,
B. Lucas,
F. Mezdari,
L. Montagnon,
N. T. Van-Oanh,
L. Lavergne,
A. Lachaize,
Y. Carpentier,
K. Béroff
Abstract:
We have measured fragmentation branching ratios of neutral CnH and CnH+ cations produced in high velocity (4.5 a.u) collisions between incident CnH+ cations and helium atoms. Electron capture gives rise to excited neutral species CnH and electronic excitation to excited cations CnH+. Thanks to a dedicated set-up, based on coincident detection of all fragments, the dissociation of the neutral and…
▽ More
We have measured fragmentation branching ratios of neutral CnH and CnH+ cations produced in high velocity (4.5 a.u) collisions between incident CnH+ cations and helium atoms. Electron capture gives rise to excited neutral species CnH and electronic excitation to excited cations CnH+. Thanks to a dedicated set-up, based on coincident detection of all fragments, the dissociation of the neutral and cationic parents were recorded separately and in a complete way. For the fragmentation of CnH, the H-loss channel is found to be dominant, as already observed by other authors. By contrast, the H-loss and C-loss channels equally dominate the two-fragment break up of CnH+ species. For these cations, we provide the first fragmentation data (n > 2). Results are also discussed in the context of astrochemistry.
△ Less
Submitted 31 March, 2008;
originally announced March 2008.
