Session Q12: Data Science in Physics

Various Examples of Using Freely Available, Downloadable Data to Teach Fourier and Wavelet Analysis

Fourier and Wavelet Analysis are important techniques for data analysis of big data. This presentation provides an example of a foundation for teaching Fourier and Wavelet Analysis to undergraduate STEM majors and provides several examples of using Fourier and Wavelet analysis on various time series such as long term climate data and light spectra. Also discussed is the importance of providing undergraduate students with an understanding of these techniques for a more complete background in undergraduate physics foundation, preparing the undergraduate students for graduate school or industry.   

Withdrawn

We present the ParaMonte software, a suite of parallel Monte Carlo optimization, sampling, integration, and Machine Learning algorithms for probabilistic and scientific inference. Born out of Data-Driven research needs in Physics, the library's primary goal is to streamline and automate scientific inference via a language-agnostic platform-agnostic environment while emphasizing and ensuring high performance, parallelism, and scalability.   

Machine Learning Enabled Intelligent Self-Powered Sensors for Next-Generation Electronic Devices

The quantum leap in artificial intelligence from the last decade makes it pervasive in almost all the domains of our daily life activities from economic activities to planning smart cities and from healthcare monitoring to smart agriculture. Machine learning is a way to provide intelligence to systems with the help of specialized algorithms. The integration of the machine learning algorithms with the devices/systems makes them independent, autonomously working which of course helps to reduce a lot of economic and human resources spending to manually carry out the work. Furthermore, triboelectric nanogenerators are one of the premier choices for fabricating self-powered sensors, referring to their remarkable features like a wide choice of materials, easy fabrication techniques, and the nonrequirement of any energy storage unit to operate. Here, we have proposed an effective strategy to integrate the machine learning algorithms with the triboelectric nanogenerators for fabricating the self-powered intelligent sensors, further employed for numerous next-generation applications such as early detection of diseases, predicting the properties of materials, and detecting abnormalities in the regular activities of the human body to a name of few.

    

Calibrated Predictive Distributions

Uncertainty quantification is crucial for assessing the predictive ability of AI algorithms. A large body of work (including normalizing flows and Bayesian neural networks) has been devoted to describing the entire predictive distribution (PD) of a target variable Y given input features X. However, off-the-shelf PDs are usually far from being conditionally calibrated; i.e., the probability of occurrence of an event given input X can be significantly different from the predicted probability. Most current research on predictive inference (such as conformal prediction) concerns constructing prediction sets, that do not only provide correct uncertainties on average over the entire population (that is, averaging over X), but are also approximately conditionally calibrated with accurate uncertainties for individual instances. It is often believed that the problem of obtaining and assessing entire conditionally calibrated PDs is too challenging to approach. In this work, we show that recalibration, as well as validation, are indeed attainable goals in practice. Our proposed method relies on the idea of regressing probability integral transform (PIT) scores against X. This regression gives full diagnostics of conditional coverage across the entire feature space and can be used to recalibrate misspecified PDs. We benchmark our corrected prediction bands against oracle bands and state-of-the-art predictive inference algorithms for synthetic data, including settings with distributional shift and dependent high-dimensional sequence data. Finally, we demonstrate an application to the physical sciences in which we assess and produce calibrated PDs for measurements of galaxy distances using imaging data (i.e., photometric redshifts).   

Machine Learning Classification of Time-varying Astrophysical Sources from the Zwicky Transient Facility

The Zwicky Transient Facility (ZTF) surveys the full Northern sky every two days, providing plentiful observations of time-varying astrophysical sources. The data from this photometric survey can reveal the intrinsic nature of sources, but the sheer amount of data prohibits a fully manual classification process. In this talk, I describe the background and current progress of the ZTF Source Classification Project (SCoPe), which employs a combination of human input and machine learning (ML) algorithms to classify sources. The project trains ML algorithms using an active learning approach, in which subsets of automated classifications are evaluated and revised by human experts before being added to a new iteration of training. By progressively improving our algorithms in this way, we advance toward our goal of providing a reliable public catalog of source classifications for the full ZTF dataset, enabling further research along multiple avenues.   

Sparse Convolutional Neural Network for NOvA

The NOvA (NuMI Off-axis v_e Appearance) experiment measures neutrino oscillations in a nearly pure muon neutrino beam over a 810 km baseline. NOvA has successfully used a Convolutional Neural Network as its main event selector and particle identifier for oscillation analysis since 2016. Despite having over 90% efficiency in classifying all neutrino types, our current network requires significant GPU resources during training and can only be applied to a window of activity around the beginning of each event. In order to reduce the computational cost, we have implemented a Sparse Convolutional Neural Network which performs convolutions only when the center pixel of the receptive field is non-zero. This reduces the size of training data and improves throughput. We have implemented a sparse FishNet architecture with PyTorch Lightning, a Python library providing high-level interface, which resulted in training accuracy of ~90%. We plan to utilize true neutrino energy information in the training in order to weight the accuracy at different energy regions to increase. In the future, we hope to incorporate semantic and instance segmentation simultaneously into the FishNet architecture for full end-to-end reconstruction.   

Evolutionary Search and Theoretical Study of Silicene Grain Boundaries' Mechanical Properties

One of the biggest obstacles in building devices using 2D materials is that defects such as grain boundaries (GBs) are almost inevitable during the synthesis process, such defects also provide a fabrication method for tuning properties for different applications. Therefore, to fully realize the potential of 2D materials, such as silicene, studying energetically stable GBs and understanding their impact on material properties is crucial. In this work, using an evolutionary algorithm search with an in-house Tersoff potential surrogate, we predicted multiple phases for the GBs on silicene, including those that were previously unreported to our best knowledge. The correlation between the topological structures, the formation energy, and mechanical properties was studied for these structures, and the research results shed light on various possible atomic structures for silicene GBs and give useful information on their impact on mechanical properties. More importantly, in this work, we demonstrate a generalizable workflow for investigating 2D GB or interfaces based on an evolutionary algorithm.   

Effects of the variation length, catastrophe and rescue on the dynamics of microtubules: machine learning approach

In this work, we investigate the effects of the variable length of the microtubule during the growth and shrink regimes, the cap length variations and the catastrophe phenomenon. The master equation and generating function were used to evaluate the mean length, the growth speed, the cap length probability and catastrophe probability of the microtubule. The results show that the growth of microtubules depends on the concentration of GTPs, of which we have two essential points of the growth speed. From a maximum concentration value, the concentration of GTPs no longer has a significant effect on length growth. The microtubule loses its cap when the length tends towards a maximum length and then, microtubule undergoes catastrophe phenomenon. The catastrophe probability increases with the microtubule length and decreases with the GTPs concentration. By applying Machine Learning techniques, we identified the MT characteristics that distinguish simultaneously all four kinetic states: growth, catastrophe, shortening, and rescue. At the cellular 25 M tubulin concentration, the most important quantities are the MT length L, average longitudinal curvature, MT tip width w, total energy of longitudinal interactions in MT lattice, and the energies of longitudinal and lateral interactions required to complete MT to full cylinder long and lat. These results allow greater insights into what brings about kinetic state stability and the transitions between states involved in MT dynamic instability behavior.

 

    

Machine Learning Model Identification on a Single HPGe Crystal Growth Dynamics-Based LSTM Neural Networks

The growth of high-quality single high-purity germanium (HPGe) crystal is of great significance for rare event searches such as dark matter and neutrinoless double beta decay. A Machine learning model identification based on long short-term memory (LSTM) neural network is implemented to the real-time data sets obtained from the crystal growth process using the Czochralski technique. Some Machine learning algorithms such as support vector machine (SVM), decision tree, and random forest are adopted to predict the growth rate and sampling results such as net impurity concentration and mobility. These models are adopted based on their stability and accuracy This study will lead to a crucial breakthrough in understanding the effective segregation coefficient of impurities in HPGe crystal.