scholarly journals Lifted Hybrid Variational Inference

2020 ◽  
Author(s):  
Yuqiao Chen ◽  
Yibo Yang ◽  
Sriraam Natarajan ◽  
Nicholas Ruozzi
Keyword(s):  
Message Passing ◽  
Computational Cost ◽  
Approximate Model ◽  
Variational Approximation ◽  
Lifted Inference ◽  
Inference Algorithms ◽  
Continuous Domains ◽  
Variational Approaches ◽  
Discrete Domains ◽  
Trivial Estimate

Lifted inference algorithms exploit model symmetry to reduce computational cost in probabilistic inference. However, most existing lifted inference algorithms operate only over discrete domains or continuous domains with restricted potential functions. We investigate two approximate lifted variational approaches that apply to domains with general hybrid potentials, and are expressive enough to capture multi-modality. We demonstrate that the proposed variational methods are highly scalable and can exploit approximate model symmetries even in the presence of a large amount of continuous evidence, outperforming existing message-passing-based approaches in a variety of settings. Additionally, we present a sufficient condition for the Bethe variational approximation to yield a non-trivial estimate over the marginal polytope.

scholarly journals Lifted Message Passing for Hybrid Probabilistic Inference

2019 ◽  
Author(s):  
Yuqiao Chen ◽  
Nicholas Ruozzi ◽  
Sriraam Natarajan
Keyword(s):  
Message Passing ◽  
Computational Cost ◽  
Markov Logic Networks ◽  
Markov Logic ◽  
Lifted Inference ◽  
Particle Sampling ◽  
Variable Elimination ◽  
Inference Algorithms ◽  
Arbitrary Potential ◽  
Inference Methods

Lifted inference algorithms for first-order logic models, e.g., Markov logic networks (MLNs), have been of significant interest in recent years.  Lifted inference methods exploit model symmetries in order to reduce the size of the model and, consequently, the computational cost of inference.  In this work, we consider the problem of lifted inference in MLNs with continuous or both discrete and continuous groundings. Existing work on lifting with continuous groundings has mostly been limited to special classes of models, e.g., Gaussian models, for which variable elimination or message-passing updates can be computed exactly.  Here, we develop approximate lifted inference schemes based on particle sampling.  We demonstrate empirically that our approximate lifting schemes perform comparably to existing state-of-the-art for models for Gaussian MLNs, while having the flexibility to be applied to models with arbitrary potential functions.

Efficient variable elimination for semi-structured simple temporal networks with continuous domains

2010 ◽  
Vol 25 (3) ◽  
pp. 337-351 ◽  
Author(s):  
Neil Yorke-Smith ◽  
Hung H. Bui
Keyword(s):  
Message Passing ◽  
Minimal Network ◽  
Hierarchical Task Network ◽  
Variable Elimination ◽  
Constraint Graph ◽  
Decomposition Structure ◽  
Tree Width ◽  
Continuous Domains ◽  
Discrete Domains ◽  
Simple Temporal Networks

AbstractThe Simple Temporal Network (STN) is a widely used framework for reasoning about quantitative temporal constraints over variables with continuous or discrete domains. The inference tasks of determining consistency and deriving the minimal network are traditionally achieved by graph algorithms (e.g. Floyd-Warshall, Johnson) or by iteration of narrowing operators (e.g. ΔSTP). None of these methods exploits effectively the tree-decomposition structure of the constraint graph of an STN. Methods based on variable elimination (e.g. adaptive consistency) can exploit this structure, but have not been applied to STNs as far as they could, in part because it is unclear how to efficiently pass the ‘messages’ over continuous domains. We show that for an STN, these messages can be represented compactly as sub-STNs. We then present an efficient message-passing scheme for computing the minimal constraints of an STN. Analysis of this algorithm,Prop-STP, brings formal explanation of the performance of the existing STN solvers ΔSTP and SR-PC. Empirical results validate the efficiency of Prop-STP, demonstrating performance comparable to ΔSTP, in cases where the constraint graph is known to have small tree width, such as those that arise during Hierarchical Task Network planning.

Message Passing Neural Networks for Partial Charge Assignment to Metal-Organic Frameworks

2020 ◽  
Author(s):  
Ali Raza ◽  
Arni Sturluson ◽  
Cory Simon ◽  
Xiaoli Fern
Keyword(s):  
Electronic Structure ◽  
Message Passing ◽  
Gas Adsorption ◽  
Metal Organic Frameworks ◽  
Computational Cost ◽  
Electronic Structure Calculations ◽  
High Fidelity ◽  
Partial Charges ◽  
Metal Organic ◽  
Structure Calculations

Virtual screenings can accelerate and reduce the cost of discovering metal-organic frameworks (MOFs) for their applications in gas storage, separation, and sensing. In molecular simulations of gas adsorption/diffusion in MOFs, the adsorbate-MOF electrostatic interaction is typically modeled by placing partial point charges on the atoms of the MOF. For the virtual screening of large libraries of MOFs, it is critical to develop computationally inexpensive methods to assign atomic partial charges to MOFs that accurately reproduce the electrostatic potential in their pores. Herein, we design and train a message passing neural network (MPNN) to predict the atomic partial charges on MOFs under a charge neutral constraint. A set of ca. 2,250 MOFs labeled with high-fidelity partial charges, derived from periodic electronic structure calculations, serves as training examples. In an end-to-end manner, from charge-labeled crystal graphs representing MOFs, our MPNN machine-learns features of the local bonding environments of the atoms and learns to predict partial atomic charges from these features. Our trained MPNN assigns high-fidelity partial point charges to MOFs with orders of magnitude lower computational cost than electronic structure calculations. To enhance the accuracy of virtual screenings of large libraries of MOFs for their adsorption-based applications, we make our trained MPNN model and MPNN-charge-assigned computation-ready, experimental MOF structures publicly available.<br>

scholarly journals Multi-fidelity prediction of molecular optical peaks with deep learning

2021 ◽  
Author(s):  
Kevin Greenman ◽  
William Green ◽  
Rafael Gómez-Bombarelli
Keyword(s):  
Optical Properties ◽  
Statistical Methods ◽  
Message Passing ◽  
Density Functional ◽  
Molecular Design ◽  
Chemical Space ◽  
Epistemic Uncertainty ◽  
Computational Cost ◽  
Regression Tree ◽  
Td Dft

Optical properties are central to molecular design for many applications, including solar cells and biomedical imaging. A variety of ab initio and statistical methods have been developed for their prediction, each with a trade-off between accuracy, generality, and cost. Existing theoretical methods such as time-dependent density functional theory (TD-DFT) are generalizable across chemical space because of their robust physics-based foundations but still exhibit random and systematic errors with respect to experiment despite their high computational cost. Statistical methods can achieve high accuracy at a lower cost, but data sparsity and unoptimized molecule and solvent representations often limit their ability to generalize. Here, we utilize directed message passing neural networks (D-MPNNs) to represent both dye molecules and solvents for predictions of molecular absorption peaks in solution. Additionally, we demonstrate a multi-fidelity approach based on an auxiliary model trained on over 28,000 TD-DFT calculations that further improves accuracy and generalizability, as shown through rigorous splitting strategies. Combining several openly-available experimental datasets, we benchmark these methods against a state-of-the-art regression tree algorithm and compare the D-MPNN solvent representation to several alternatives. Finally, we explore the interpretability of the learned representations using dimensionality reduction and evaluate the use of ensemble variance as an estimator of the epistemic uncertainty in our predictions of molecular peak absorption in solution. The prediction methods proposed herein can be integrated with active learning, generative modeling, and experimental workflows to enable the more rapid design of molecules with targeted optical properties.

scholarly journals HPC simulations of brownout: A noninteracting particles dynamic model

2020 ◽  
Vol 34 (3) ◽  
pp. 267-281
Author(s):  
Roberto Porcù ◽  
Edie Miglio ◽  
Nicola Parolini ◽  
Mattia Penati ◽  
Noemi Vergopolan
Keyword(s):  
Message Passing ◽  
High Performance ◽  
Message Passing Interface ◽  
Graphics Processing Unit ◽  
Time Integration ◽  
Computational Cost ◽  
Aircraft Design ◽  
Euler Method ◽  
Processing Unit ◽  
Integration Algorithm

Helicopters can experience brownout when flying close to a dusty surface. The uplifting of dust in the air can remarkably restrict the pilot’s visibility area. Consequently, a brownout can disorient the pilot and lead to the helicopter collision against the ground. Given its risks, brownout has become a high-priority problem for civil and military operations. Proper helicopter design is thus critical, as it has a strong influence over the shape and density of the cloud of dust that forms when brownout occurs. A way forward to improve aircraft design against brownout is the use of particle simulations. For simulations to be accurate and comparable to the real phenomenon, billions of particles are required. However, using a large number of particles, serial simulations can be slow and too computationally expensive to be performed. In this work, we investigate an message passing interface (MPI) + graphics processing unit (multi-GPU) approach to simulate brownout. In specific, we use a semi-implicit Euler method to consider the particle dynamics in a Lagrangian way, and we adopt a precomputed aerodynamic field. Here, we do not include particle–particle collisions in the model; this allows for independent trajectories and effective model parallelization. To support our methodology, we provide a speedup analysis of the parallelization concerning the serial and pure-MPI simulations. The results show (i) very high speedups of the MPI + multi-GPU implementation with respect to the serial and pure-MPI ones, (ii) excellent weak and strong scalability properties of the implemented time-integration algorithm, and (iii) the possibility to run realistic simulations of brownout with billions of particles at a relatively small computational cost. This work paves the way toward more realistic brownout simulations, and it highlights the potential of high-performance computing for aiding and advancing aircraft design for brownout mitigation.

scholarly journals Accurate and efficient representation of intramolecular energy inab initiogeneration of crystal structures. I. Adaptive local approximate models

2016 ◽  
Vol 72 (6) ◽  
pp. 864-874 ◽  
Author(s):  
Isaac Sugden ◽  
Claire S. Adjiman ◽  
Constantinos C. Pantelides
Keyword(s):  
Structure Prediction ◽  
Computational Cost ◽  
Search Space ◽  
Approximate Model ◽  
High Energy ◽  
Computational Effort ◽  
Crystal Structure Prediction ◽  
Blind Test ◽  
Acta Cryst ◽  
Simplifying Assumptions

The global search stage of crystal structure prediction (CSP) methods requires a fine balance between accuracy and computational cost, particularly for the study of large flexible molecules. A major improvement in the accuracy and cost of the intramolecular energy function used in theCrystalPredictor II[Habgoodet al.(2015).J. Chem. Theory Comput.11, 1957–1969] program is presented, where the most efficient use of computational effort is ensuredviathe use of adaptive local approximate model (LAM) placement. The entire search space of the relevant molecule's conformations is initially evaluated using a coarse, low accuracy grid. Additional LAM points are then placed at appropriate points determinedviaan automated process, aiming to minimize the computational effort expended in high-energy regions whilst maximizing the accuracy in low-energy regions. As the size, complexity and flexibility of molecules increase, the reduction in computational cost becomes marked. This improvement is illustrated with energy calculations for benzoic acid and the ROY molecule, and a CSP study of molecule (XXVI) from the sixth blind test [Reillyet al.(2016).Acta Cryst.B72, 439–459], which is challenging due to its size and flexibility. Its known experimental form is successfully predicted as the global minimum. The computational cost of the study is tractable without the need to make unphysical simplifying assumptions.

Optimized parallel simulations of analytic bond-order potentials on hybrid shared/distributed memory with MPI and OpenMP

2017 ◽  
Vol 33 (2) ◽  
pp. 227-241 ◽  
Author(s):  
Carlos Teijeiro ◽  
Thomas Hammerschmidt ◽  
Ralf Drautz ◽  
Godehard Sutmann
Keyword(s):  
Message Passing ◽  
Bond Order ◽  
Message Passing Interface ◽  
Parallel Implementation ◽  
Computational Cost ◽  
Parallel Simulations ◽  
Decomposition Scheme ◽  
Significant Performance ◽  
Restricted Volume ◽  
Bond Order Potentials

Analytic bond-order potentials (BOPs) allow to obtain a highly accurate description of interatomic interactions at a reasonable computational cost. However, for simulations with very large systems, the high memory demands require the use of a parallel implementation, which at the same time also optimizes the use of computational resources. The calculations of analytic BOPs are performed for a restricted volume around every atom and therefore have shown to be well suited for a message passing interface (MPI)-based parallelization based on a domain decomposition scheme, in which one process manages one big domain using the entire memory of a compute node. On the basis of this approach, the present work focuses on the analysis and enhancement of its performance on shared memory by using OpenMP threads on each MPI process, in order to use many cores per node to speed up computations and minimize memory bottlenecks. Different algorithms are described and their corresponding performance results are presented, showing significant performance gains for highly parallel systems with hybrid MPI/OpenMP simulations up to several thousands of threads.

scholarly journals Efficient Inference for Untied MLNs

2017 ◽  
Author(s):  
Somdeb Sarkhel ◽  
Deepak Venugopal ◽  
Nicholas Ruozzi ◽  
Vibhav Gogate
Keyword(s):  
Graphical Models ◽  
Scaling Up ◽  
Partition Functions ◽  
Markov Logic ◽  
Practical Applications ◽  
Lifted Inference ◽  
Hard Time ◽  
Inference Algorithms ◽  
Order Of Magnitude ◽  
Symmetric Approximation

We address the problem of scaling up local-search or sampling-based inference in Markov logic networks (MLNs) that have large shared sub-structures but no (or few) tied weights. Such untied MLNs are ubiquitous in practical applications. However, they have very few symmetries, and as a result lifted inference algorithms--the dominant approach for scaling up inference--perform poorly on them. The key idea in our approach is to reduce the hard, time-consuming sub-task in sampling algorithms, computing the sum of weights of features that satisfy a full assignment, to the problem of computing a set of partition functions of graphical models, each defined over the logical variables in a first-order formula. The importance of this reduction is that when the treewidth of all the graphical models is small, it yields an order of magnitude speedup. When the treewidth is large, we propose an over-symmetric approximation and experimentally demonstrate that it is both fast and accurate.

Δ-Quantum machine learning for medicinal chemistry

2021 ◽  
Author(s):  
Kenneth Atz ◽  
Clemens Isert ◽  
Markus N. A. Böcker ◽  
José Jiménez-Luna ◽  
Gisbert Schneider
Keyword(s):  
Machine Learning ◽  
Message Passing ◽  
Density Functional ◽  
Large Scale ◽  
Molecular Design ◽  
Low Cost ◽  
Computational Cost ◽  
Three Dimensional ◽  
Biomolecular Systems ◽  
Quantum Observables

Many molecular design tasks benefit from fast and accurate calculations of quantum-mechanical (QM) properties. However, the computational cost of QM methods applied to drug-like molecules currently renders large-scale applications of quantum chemistry challenging. Aiming to mitigate this problem, we developed DelFTa, an open-source toolbox for the prediction of electronic properties of drug-like molecules at the density functional (DFT) level of theory, using Δ-machine-learning. Δ-Learning corrects the prediction error (Δ) of a fast but inaccurate property calculation. DelFTa employs state-of-the-art three-dimensional message-passing neural networks trained on a large dataset of QM properties. It provides access to a wide array of quantum observables on the molecular, atomic and bond levels by predicting approximations to DFT values from a low-cost semiempirical baseline. Δ-Learning outperformed its direct-learning counterpart for most of the considered QM endpoints. The results suggest that predictions for non-covalent intra- and intermolecular interactions can be extrapolated to larger biomolecular systems. The software is fully open-sourced and features documented command-line and Python APIs.

Sign in / Sign up

Export Citation Format

Share Document