Monte-Carlo Search Research Articles

DrugSynthMC: An Atom-Based Generation of Drug-like Molecules with Monte Carlo Search.

A growing number of deep learning (DL) methodologies have recently been developed to design novel compounds and expand the chemical space within virtual libraries. Most of these neural network approaches design molecules to specifically bind a target based on its structural information and/or knowledge of previously identified binders. Fewer attempts have been made to develop approaches for de novo design of virtual libraries, as synthesizability of generated molecules remains a challenge. In this work, we developed a new Monte Carlo Search (MCS) algorithm, DrugSynthMC (Drug Synthesis using Monte Carlo), in conjunction with DL and statistical-based priors to generate thousands of interpretable chemical structures and novel drug-like molecules per second. DrugSynthMC produces drug-like compounds using an atom-based search model that builds molecules as SMILES, character by character. Designed molecules follow Lipinski's "rule of 5″, show a high proportion of highly water-soluble nontoxic predicted-to-be synthesizable compounds, and efficiently expand the chemical space within the libraries, without reliance on training data sets, synthesizability metrics, or enforcing during SMILES generation. Our approach can function with or without an underlying neural network and is thus easily explainable and versatile. This ease in drug-like molecule generation allows for future integration of score functions aimed at different target- or job-oriented goals. Thus, DrugSynthMC is expected to enable the functional assessment of large compound libraries covering an extensive novel chemical space, overcoming the limitations of existing drug collections. The software is available at https://github.com/RoucairolMilo/DrugSynthMC.

Bounds on galaxy stochasticity from halo occupation distribution modeling

The joint probability distribution of matter overdensity and galaxy counts in cells is a powerful probe of cosmology, and the extent to which variance in galaxy counts at fixed matter density deviates from Poisson shot noise is not fully understood. The lack of informed bounds on this stochasticity is currently the limiting factor in constraining cosmology with the galaxy–matter probability distribution function (PDF). We investigate stochasticity in the conditional distribution of galaxy counts along lines of sight with fixed matter density, and we present a halo occupation distribution (HOD)-based approach for obtaining plausible ranges for stochasticity parameters. To probe the high-dimensional space of possible galaxy–matter connections, we derive a set of HODs that conserve the galaxies’ linear bias and number density to produce REDMAGIC-like galaxy catalogs within the ABACUSSUMMIT suite of N-body simulations. We study the impact of individual HOD parameters and cosmology on stochasticity and perform a Monte Carlo search in HOD parameter space subject to the constraints on bias and density. In mock catalogs generated by the selected HODs, shot noise in galaxy counts spans both sub-Poisson and super-Poisson values, ranging from 80% to 133% of Poisson variance for cells with mean matter density. Nearly all of the derived HODs show a positive relationship between local matter density and stochasticity. For galaxy catalogs with higher stochasticity, modeling galaxy bias to second order is required for an accurate description of the conditional PDF of galaxy counts at fixed matter density. The presence of galaxy assembly bias also substantially extends the range of stochasticity in the super-Poisson direction. This HOD-based approach leverages degrees of freedom in the galaxy–halo connection to obtain informed bounds on nuisance model parameters and can be adapted to study other parametrizations of shot noise in galaxy counts, in particular to motivate prior ranges on stochasticity for cosmological analyses.

Design of parallel 𝛽-sheet nanofibrils using Monte Carlo search, coarse-grained simulations, and experimental testing.

Peptide self-assembly into amyloid fibrils provides numerous applications in drug delivery and biomedical engineering applications. We augment our previously-established computational screening technique along with experimental biophysical characterization to discover 7-mer peptides that self-assemble into "parallel β-sheets", that is, β-sheets with N-terminus-to-C-terminus 𝛽-strand vectors oriented in parallel. To accomplish the desired β-strand organization, we applied the PepAD amino acid sequence design software to the Class-1 cross-β spine defined by Sawaya et al. This molecular configuration includes two layers of parallel β-sheets stacked such that N-terminus-to-C-terminus vectors are oriented antiparallel for molecules on adjacent β-sheets. The first cohort of PepAD identified peptides were examined for their fibrillation behavior in DMD/PRIME20 simulations, and the top performing sequence was selected as a prototype for a subsequent round of sequence refinement. The two rounds of design resulted in a library of eight 7-mer peptides. In DMD/PRIME20 simulations, five of these peptides spontaneously formed fibril-like structures with a predominantly parallel 𝛽-sheet arrangement, two formed fibril-like structure with <50% in parallel 𝛽-sheet arrangement and one remained a random coil. Among the eight candidate peptides produced by PepAD and DMD/PRIME20, five were synthesized and purified. All five assembled into amyloid fibrils composed of parallel β-sheets based on Fourier transform infrared spectroscopy, circular dichroism, electron microscopy, and thioflavin-T fluorescence spectroscopy measurements.

Comparing search algorithms on the retrosynthesis problem.

In this article we try different algorithms, namely Nested Monte Carlo Search and Greedy Best First Search, on AstraZeneca's open source retrosynthetic tool : AiZynthFinder. We compare these algorithms to AiZynthFinder's base Monte Carlo Tree Search on a benchmark selected from the PubChem database and by Bayer's chemists. We show that both Nested Monte Carlo Search and Greedy Best First Search outperform AstraZeneca's Monte Carlo Tree Search, with a slight advantage for Nested Monte Carlo Search while experimenting on a playout heuristic. We also show how the search algorithms are bounded by the quality of the policy network, in order to improve our results the next step is to improve the policy network.

Multi-step carbon emissions forecasting model for industrial process based on a new strategy and machine learning methods

The rising industrial sector capacity increases carbon emissions, necessitating a low-carbon transformation for global sustainability. Accurate multi-step forecasting of industrial carbon emissions is crucial for optimizing production, reducing energy consumption, improving efficiency, and achieving decarbonization. However, current forecasting faces challenges like model uncertainty and low accuracy with multi forecasting steps. To address these issues, this study proposed a multi-step carbon emissions forecasting model for industrial processes. Firstly, a new multi-step forecasting strategy framework was proposed. Secondly, SHapley Additive exPlanations (SHAP) and lagged autocorrelation analysis were used to extract key features affecting industrial carbon emissions. On this basis, the grey model, autoregressive moving average (ARMA) model, and least squares support vector machine (LSSVM) method were used to build forecasting models separately. The weights of the three models were solved using the induced ordered weighted average operator and the Monte Carlo search tree (MCST) method. The final forecasting results were obtained by using the weighted summation method. The data collected from the real industry were used to validate the proposed model. The results showed that the proposed model could accurately forecast industrial carbon emissions within 12 steps and its MAPE is less than 10%. The proposed multi-step industrial carbon emissions forecasting model can provide data support for industrial energy saving, carbon reduction, and green and sustainable development.

TIR predictor and optimizer: Web-tools for accurate prediction of translation initiation rate and precision gene design in Saccharomyces cerevisiae.

Translation initiation is the primary determinant of the rate of protein production. The variation in the rate with which this step occurs can cause up to three orders of magnitude differences in cellular protein levels. Several mRNA features, including mRNA stability in proximity to the start codon, coding sequence length, and presence of specific motifs in the mRNA molecule, have been shown to influence the translation initiation rate. These molecular factors acting at different strengths allow precise control of in vivo translation initiation rate and thus the rate of protein synthesis. However, despite the paramount importance of translation initiation rate in protein synthesis, accurate prediction of the absolute values of initiation rate remains a challenge. In fact, as of now, there is no available model for predicting the initiation rate in Saccharomyces cerevisiae. To address this, we train a machine learning model for predicting the in vivo initiation rate in S. cerevisiae transcripts. The model is trained using a diverse set of mRNA transcripts, enabling the comparison of initiation rates across different transcripts. Our model exhibited excellent accuracy in predicting the translation initiation rate and demonstrated its effectiveness with both endogenous and exogenous transcripts. Then, by combining the machine learning model with the Monte-Carlo search algorithm, we have also devised a method to optimize the nucleotide sequence of any gene to achieve a specific target initiation rate. The machine learning model we've developed for predicting translation initiation rates, along with the gene optimization method, are deployed as a web server. Both web servers are accessible for free at the following link: ajeetsharmalab.com/TIRPredictor. Thus, this research advances our fundamental understanding of translation initiation processes, with direct applications in biotechnology.

Downhole Track Detection via Multi-dimensional Conditional Generative Adversarial Nets

Frequent mine disasters cause a large number of casualties and property losses. Autonomous driving is a fundamental measure for solving this problem, and track detection is one of the key technologies for computer vision to achieve downhole automatic driving. The track detection result based on the existing models lacks the semantic detail of the tracks and relies too much on visual postprocessing technology. Therefore, this paper proposes a track detection model based on the multi-dimensional conditional generative adversarial network. First, the generator is decomposed into global and local parts using a multi-granularity structure. Second, a multi-scale shared convolution structure is introduced into the discriminator to further guide the generator. In addition, this paper proposes a penalty mechanism based on Monte Carlo search to enhance the semantic constraints in the image generation process. Compared with the state-of-the-art semantic segmentation algorithms, extensive experiments on the downhole scene dataset demonstrate proposed model achieved the best results in terms of pixel accuracy, intersection-over-union (IOU) and the track detection accuracy. This paper provides a new idea for track line detection. In the future, the model can also be applied to other segmentation problems as well. Code and data will be shared.

Data Augmentation for Bayesian Deep Learning

Deep Learning (DL) methods have emerged as one of the most powerful tools for functional approximation and prediction. While the representation properties of DL have been well studied, uncertainty quantification remains challenging and largely unexplored. Data augmentation techniques are a natural approach to provide uncertainty quantification and to incorporate stochastic Monte Carlo search into stochastic gradient descent (SGD) methods. The purpose of our paper is to show that training DL architectures with data augmentation leads to efficiency gains. We use the theory of scale mixtures of normals to derive data augmentation strategies for deep learning. This allows variants of the expectation-maximization and MCMC algorithms to be brought to bear on these high dimensional nonlinear deep learning models. To demonstrate our methodology, we develop data augmentation algorithms for a variety of commonly used activation functions: logit, ReLU, leaky ReLU and SVM. Our methodology is compared to traditional stochastic gradient descent with back-propagation. Our optimization procedure leads to a version of iteratively re-weighted least squares and can be implemented at scale with accelerated linear algebra methods providing substantial improvement in speed. We illustrate our methodology on a number of standard datasets. Finally, we conclude with directions for future research.

Gene Expression Machine Learning Models Classify Pediatric AML Subtypes with High Performance

Pediatric acute myeloid leukemia (pAML) encompasses over 20 molecular subtypes driven by unique genetic alterations, including hallmark chromosomal rearrangements and less frequently, point mutations or tandem duplications. Collectively, many of these pAML molecular categories are enriched in pediatric populations and are not represented by current classification systems, including the recently updated WHO or ICC. While fusion detection from RNA-Seq-based approaches is robust, many fusion negative subtypes would need to be defined by expression-based approaches as mutation calling from RNA-Seq data is less developed. Nevertheless, this could be challenging for subtypes with similar transcriptional profiles, such as those with shared HOX expression patterns, including NPM1, NUP98r, UBTF, DEK::NUP214, and KMT2A-PTD. To aid in the appropriate molecular classification of pAML, which is crucial for prognosis, we developed and compared three gene expression-based classifiers. A total of 1707 pAML gene expression profiles were mapped and analyzed from three distinct sources (St. Jude = 659; TARGET = 168; AAML1031 = 880). Raw read count data was normalized and scaled to obtain a relative expression value in transcripts per million (TPM), which served as input for feature selection, model training, and testing. Ground truth labels for all 1707 samples were obtained through multi-omics analysis, including whole genome sequencing, to identify fusions and mutations. For validation purposes, the data was stratified by subtypes and split 70/30 into training (n=1187) and testing (n=520). Three machine learning models were selected: random forest, XGboost, and linear support vector machine (SVM). Each sample had gene expression TPM data for 60,754 transcripts, out of which 20,004 transcripts related to protein-coding genes were incorporated for feature selection. Feature selection was performed using a median absolute deviation (MAD) algorithm to select the 3000 transcripts with the highest variability. Top variable 3000 genes were selected to allow for adequate tuning of the number of predictors. Each model was independently trained using stratified cross-validation and Monte-Carlo search for hyperparameter tuning. The best model was selected based on the Matthews Correlation Coefficient (MCC). Each model was tested on the hold-out TPM set with z-score normalization. On the hold-out testing set (n=520), the linear SVM model outperformed the random forest and XGboost models on five performance metrics across all subtypes (sensitivity=0.9577; precision=0.9577; specificity=0.9978; F1=0.96; accuracy=0.9958). The random forest (sensitivity=0.9154; precision=0.9154; specificity=0.9955; F1=0.92; accuracy=0.9915) and XGboost models (sensitivity=0.9231; precision=0.9231; specificity=0.9960; F1=0.92; accuracy=0.9923) also performed well across all subtypes. Although feature selection was shared across all three models, performance within each subtype varied between models. The linear SVM model demonstrated strong performance overall, driven by high specificity in classifying the KMT2Ar subgroup (n=127) and equal sensitivity across the GLIS-rearranged (n=14), GATA1 (n=10), BCL11B (n=7), CBFB::MYH11 (n=60), CEBPA (n=33), and RUNX1::RUNX1T1 (n=78) subtypes. The primary difference in performance between models is the high false positive rate for KMT2Ar and NPM1 (n=55) in the random forest and XGboost models. A preliminary hypothesis for this might be due to the large representation of KMT2Ar and NPM1 in the training data (24.85% and 10.78%, respectively). Synthetic upsampling (SMOTE) for the training dataset (n=1369) counteracts bias towards the majority classes and increases performance for the random forest (sensitivity=0.9327; precision=0.9327; specificity=0.9965; F1=0.93; accuracy=0.9933), XGboost (sensitivity=0.9404; precision=0.9404; specificity=0.9969; F1=0.94; accuracy=0.9940) and linear SVM (sensitivity=0.9615; precision=0.9615; specificity=0.9980; F1=0.96; accuracy=0.9962) models. Conjointly, these models demonstrate the utility and effectiveness of a machine learning approach for classifying pAML samples from transcriptome sequencing data, which may have broad clinical and research utility, especially for fusion negative subtypes.

Data-Driven Discovery of Lithium-Ion Battery State of Charge Dynamics

Abstract We present a physics-inspired input/output predictor of lithium-ion batteries (LiBs) for online state-of-charge (SOC) prediction. The complex electrochemical behavior of batteries results in nonlinear and high-dimensional dynamics. Accurate SOC prediction is paramount for increased performance, improved operational safety, and extended longevity of LiBs. The battery's internal parameters are cell-dependent and change with operating conditions and battery health variations. We present a data-driven solution to discover governing equations pertaining to SOC dynamics from battery operando measurements. Our approach relaxes the need for detailed knowledge of the battery's composition while maintaining prediction fidelity. The predictor consists of a library of candidate terms and a set of coefficients found via a sparsity-promoting algorithm. The library was enhanced with explicit physics-inspired terms to improve the predictor's interpretability and generalizability. Further, we developed a Monte Carlo search of additional nonlinear terms to efficiently explore the high-dimensional search space and improve the characterization of highly nonlinear behaviors. Also, we developed a hyperparameter autotuning approach for identifying optimal coefficients that balance accuracy and complexity. The resulting SOC predictor achieved high predictive performance scores (RMSE) of 2.2×10−6 and 4.8×10−4, respectively, for training and validation on experimental results corresponding to a stochastic drive cycle. Furthermore, the predictor achieved an RMSE of 8.5×10−4 on unseen battery measurements corresponding to the standard US06 drive cycle, further showcasing the adaptability of the predictor and the enhanced modeling approach to new conditions.

Finding the best opening in chess with multi-armed bandit algorithm

With many gaming AI being developed and being able to defeat top human players in recent years, AI has once again become the hottest topic in research and even in our daily life. This paper also researches on gaming AI and chess. Instead of using deep learning and the Monte Carlo Search algorithm, this paper focuses on the opening only with multi-armed bandit algorithms to find the best moves and opening. Specifically, the method used in this paper is epsilon greedy and Thompson sampling. The dataset used in this paper is from Kaggle. This paper considers each move as a set of choices one needs to make and considers the big picture as a multi-armed bandit problem. This paper aims to develop a relative best strategy to counter those opening or make changes to a disadvantaged situation.

The LISA Data Challenge Radler analysis and time-dependent ultra-compact binary catalogues

Context. Galactic binaries account for the loudest combined continuous gravitational wave signal in the Laser Interferometer Space Antenna (LISA) band, which spans a frequency range of 0.1 mHz–1 Hz. Aims. A superposition of low frequency Galactic and extragalactic signals and instrument noise comprise the LISA data stream. Resolving as many Galactic binary signals as possible and characterising the unresolved Galactic foreground noise after their subtraction from the data are a necessary step towards a global fit solution to the LISA data. Methods. We analysed a simulated gravitational wave time series of tens of millions of ultra-compact Galactic binaries hundreds of thousands of years from merger. This dataset is called the Radler galaxy and is part of the LISA Data Challenges. We used a Markov chain Monte Carlo search pipeline specifically designed to perform a global fit to the Galactic binaries and detector noise. Our analysis was performed for increasingly larger observation times of 1.5, 3, 6 and 12 months. Results. We show that after one year of observing, as many as ten thousand ultra-compact binary signals are individually resolvable. Ultra-compact binary catalogues corresponding to each observation time are presented. The Radler galaxy is a training dataset, with binary parameters for every signal in the data stream included. We compare our derived catalogues to the LISA Data Challenge Radler catalogue to quantify the detection efficiency of the search pipeline. Included in the appendix is a more detailed analysis of two corner cases that provide insight into future improvements to our search pipeline.

Design and analysis of a 2-DOF compliant serial micropositioner based on "S-shaped" flexure hinge

A serial 2D planar compliant mechanism has been proposed based on novel "S-shaped" flexure hinges for atomic force microscopy applications. The bridge-lever type mechanism driven by piezoelectric actuators serves as a displacement amplifier. The optimal geometrical parameters were found to maximize the workspace and natural frequency through the Monte-Carlo search algorithm. Also, the performance of the developed system was investigated by Matrix-based Compliance Modeling (MCM), Finite Element Analysis (FEA), and experiments. All models indicate that the mechanism has an approximately linear force-deflection relationship, high safety factor, and a reasonable amplification ratio of about 4.5 and 5.5 for the inner and outer stages in the analytical approach, 4.4 and 5.3 in FEA. Finally, experiments on a fabricated open-loop controlled prototype revealed 4.2 and 5.1 amplification ratio for stages that resulted in a 94 × 124 μm rectangular workspace and a natural frequency of 224.6 Hz, which is about 5% lower than the results predicted using FEA.

Review on the application of deep learning algorithms in video game AI agent

AI algorithms have been applied in video games more than before. Since the early 2000s, the benchmark of the application of AI agent in video games has been put forward by researchers and engineers. This paper introduces the application of four different deep learning algorithms in video games, namely Component-based Hierarchical State Machine (CBHSM) algorithm, Monte Carlo Search Tree (MCTS) algorithm, Generative Adversarial Networks (GAN) algorithm, and the combination of A* algorithm and Q-learning algorithm. It can be summarized that both CBHSM and MCTS algorithms can modify and optimize the traditional Finite State Machine (FSM) algorithm applied in NPC (Non Player Characters) system. Additionally, A* algorithm can be combined with Q-learning algorithm to solve the problem of the difficulty on huge state space and limited time, and GAN algorithm, a model of unsupervised learning, can generate results by fewer training sets.

UAV Aerial Image Generation of Crucial Components of High-Voltage Transmission Lines Based on Multi-Level Generative Adversarial Network

With the aim of improving the image quality of the crucial components of transmission lines taken by unmanned aerial vehicles (UAV), a priori work on the defective fault location of high-voltage transmission lines has attracted great attention from researchers in the UAV field. In recent years, generative adversarial nets (GAN) have achieved good results in image generation tasks. However, the generation of high-resolution images with rich semantic details from complex backgrounds is still challenging. Therefore, we propose a novel GANs-based image generation model to be used for the critical components of power lines. However, to solve the problems related to image backgrounds in public data sets, considering that the image background of the common data set CPLID (Chinese Power Line Insulator Dataset) is simple. However, it cannot fully reflect the complex environments of transmission line images; therefore, we established an image data set named “KCIGD” (The Key Component Image Generation Dataset), which can be used for model training. CFM-GAN (GAN networks based on coarse–fine-grained generators and multiscale discriminators) can generate the images of the critical components of transmission lines with rich semantic details and high resolutions. CFM-GAN can provide high-quality image inputs for transmission line fault detection and line inspection models to guarantee the safe operation of power systems. Additionally, we can use these high-quality images to expand the data set. In addition, CFM-GAN consists of two generators and multiple discriminators, which can be flexibly applied to image generation tasks in other scenarios. We introduce a penalty mechanism-related Monte Carlo search (MCS) approach in the CFM-GAN model to introduce more semantic details in the generated images. Moreover, we presented a multiscale discriminator structure according to the multitask learning mechanisms to effectively enhance the quality of the generated images. Eventually, the experiments using the CFM-GAN model on the KCIGD dataset and the publicly available CPLID indicated that the model used in this work outperformed existing mainstream models in improving image resolution and quality.

Generative design of texture for sliding surface based on machine learning

Textured surface is of fundamental and practical importance in numerous emerging applications due to its beneficial effects on the tribological performance. In this work, a machine learning based universal generative design framework is proposed for surface texturing designing by combining specific convolutional neural network with improved Monte Carlo search. The optimal patterns of surface texture generated by machine learning are systematically studied under different conditions. Our results show that the machine generated wavy and chevron-like textures have the potential to dramatically improve the tribological performance of sliding surface with infinite design domain. Compared with the reported optimal texture, the friction coefficient of machine generated texture is reduced to 27.3∼49.7%, and the load carrying capacity is increased to 126.1∼144.4%.

Optimization design for slip/no-slip configuration of hydrophobic sliding bearings using Monte Carlo search

This paper aims to globally search the optimum slip/no-slip configuration of hydrophobic surface for improving tribological performance of the sliding bearing, such as load carrying capacity and friction loss. The nonlinear limiting shear stress model instead of conventional slip length model is employed to accurately simulate the boundary slippage of the hydrophobic surface. The efficient Monte Carlo search is developed for the optimization design of slip/no-slip configuration in hydrophobic surface without relying on initial configuration and sensitivity. Compared to genetic algorithm, only one numerical reanalysis is required for each Monte Carlo search in slip/no-slip configuration designing. Effect of aspect ratios of the hydrophobic sliding bearing on optimal slip/no-slip configuration is studied. Our results demonstrate the validity of the proposed method.

The taxicab sampler: MCMC for discrete spaces with application to tree models

Motivated by the problem of exploring discrete but very complex state spaces in Bayesian models, we propose a novel Markov Chain Monte Carlo search algorithm: the taxicab sampler. We describe the construction of this sampler and discuss how its interpretation and usage differs from that of standard Metropolis-Hastings as well as the related Hamming ball sampler. The proposed sampling algorithm is then shown to demonstrate substantial improvement in computation time without any loss of efficiency relative to a naïve Metropolis–Hastings search in a motivating Bayesian regression tree count model, in which we leverage the discrete state space assumption to construct a novel likelihood function that allows for flexibly describing different mean-variance relationships while preserving parameter interpretability compared to existing likelihood functions for count data.

Air combat manoeuvre strategy algorithm based on two-layer game decision-making and the distributed MCTS method with double game trees

ABSTRACT In view of the huge strategy space and high real-time requirement for multi-fighter air combat maneouvre decisions, the target allocation and the manoeuvre decision model are established, respectively and the air combat strategy solving algorithm is proposed based on two-layer game decision-making and the distributed Monte Carlo strategy search method with double game trees. Moreover, the two-layer game decision-making method can precut the huge game tree strategy space, which improves the efficiency of strategy search. The distributed Monte Carlo strategy search method with double game trees can quickly search out the optimal decision scheme of an air combat game based on the opponent’s strategy. The experiment results show that the designed algorithm is effective and improves the efficiency of the decision compared with the Monte Carlo search algorithm of single-layer decision-making.

Air Combat Maneuver Strategy Algorithm Based on Two-Layer Game Decision-Making and Distributed Double Game Trees MCTS under Uncertain Information

In this paper, a model for maneuver decisions in air combat is established based on position situation information, the performance of the fighter, the threat of combat intention, and the multi-fighter collaboration effect. Additionally, a two-layer game decision algorithm based on the double game tree distributed Monte Carlo search strategy is proposed, and the operational rules of interval numbers and possibility degree comparison rules are adopted to solve the designed method. The experiment results show that the model and algorithm are effective in their intended purpose. The two-layer game decision-making and distributed double game tree MCTS can precut the huge game tree strategy space and quickly identify the optimal air combat game decision scheme, which improves the efficiency of strategy searches. By compared experiments, it was found that the proposed algorithm can improve the performance of air combat.