Title: Injecting Domain Knowledge from Empirical Interatomic Potentials to Neural Networks for Predicting Material Properties
For decades, atomistic modeling has played a crucial role in predicting the behavior of materials in numerous fields ranging from nanotechnology to drug discovery. The most accurate methods in this domain are rooted in first-principles quantum mechanical calculations such as density functional theory (DFT). Because these methods have remained computationally prohibitive, practitioners have traditionally focused on defining physically motivated closed-form expressions known as empirical interatomic potentials (EIPs) that approximately model the interactions between atoms in materials. In recent years, neural network (NN)-based potentials trained on quantum mechanical (DFT-labeled) data have emerged as a more accurate alternative to conventional EIPs. However, the generalizability of these models relies heavily on the amount of labeled training data, which is often still insufficient to generate models suitable for general-purpose applications. In this paper, we propose two generic strategies that take advantage of unlabeled training instances to inject domain knowledge from conventional EIPs to NNs in order to increase their generalizability. The first strategy, based on weakly supervised learning, trains an auxiliary classifier on EIPs and selects the best-performing EIP to generate energies to supplement the ground-truth DFT energies in training the NN. The second strategy, based on transfer learning, first pretrains the NN on a large set of easily obtainable EIP energies, and then fine-tunes it on ground-truth DFT energies. Experimental results on three benchmark datasets demonstrate that the first strategy improves baseline NN performance by 5% to 51% while the second improves baseline performance by up to 55%. Combining them further boosts performance.  more » « less
Award ID(s):
1834251
PAR ID:
10382006
Author(s) / Creator(s):
; ; ; ; ;
Date Published:
Journal Name:
NeurIPS 2022
Subject(s) / Keyword(s):
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract Background:Hospital sinks are reservoirs for epidemiologically important pathogens (EIPs), yet practical, effective strategies for sustained decontamination are lacking. Methods:We conducted a randomized controlled trial of 30 in-room sinks (15 intervention, 15 control) in a newly renovated hospital unit to evaluate the efficacy of a hydrogen peroxide/peracetic acid foamed disinfectant in reducing sink contamination. Intervention sinks received foamed disinfectant to sink drains three times weekly; control sinks underwent standard daily surface cleaning. Weekly sampling was performed from three sink locations (top surface, tail pipe, P-trap) over 35 weeks. The primary outcome was sink conversion events (SCEs), defined as first detection of ≥1 EIP, defined asPseudomonas aeruginosa,Stenotrophomonas spp., orAcinetobacterspp., and ESBL-producing or carbapenem-resistant Enterobacterales, in previously negative sinks. Results:A total of 2880 samples were collected. All sinks were negative at baseline for study pathogens. Nearly all sinks (29/30) experienced an SCE during the study period. However, only 44 (9%) intervention sink samples were positive for EIPs, compared to 270 (47%) in control sinks (p< 0.00001). EIPs were recovered from 4% versus 24% of P-traps and 4% versus 39% of tail pipes; sink top/handle contamination was rare and similar (3% vs 4%). The most frequent EIPs wereAcinetobacterspp. andStenotrophomonasspp. Intervention sinks experienced a delayed time to SCE (p= 0.0001). Items were stored on/in sinks in 93% of observations. Conclusion:Regular application of a foamed disinfectant reduced and delayed EIP contamination in renovated hospital sinks. Foam-based protocols may help mitigate environmental reservoirs of multidrug-resistant organisms. 
    more » « less
  2. Conditional-probability density functional theory (CP-DFT) is a formally exact method for finding correlation energies from Kohn-Sham DFT without evaluating an explicit energy functional. We present details on how to generate accurate exchange-correlation energies for the ground-state uniform gas. We also use the exchange hole in a CP antiparallel spin calculation to extract the high-density limit. We give a highly accurate analytic solution to the Thomas-Fermi model for this problem, showing its performance relative to Kohn-Sham and it may be useful at high temperatures. We explore several approximations to the CP potential. Results are compared to accurate parametrizations for both exchange-correlation energies and holes. 
    more » « less
  3. Abstract Upward Terrestrial Gamma‐Ray Flashes (TGFs) are mainly produced during the upward propagating negative leaders inside thunderclouds. The exact source position of TGFs, which is crucial to understanding TGF source properties, is still unclear. The link between positive energetic in‐cloud pulses (+EIPs) and TGFs provides us with a potential target to aim at. In this study, the low‐frequency radio emissions of 75 +EIPs are analyzed to retrieve the source altitudes with an improved ray theory model. Furthermore, the meteorology contexts of +EIPs derived from the ground‐based weather radars and satellite‐based infrared cloud top temperature measurements are investigated. +EIPs are produced at 8.8–13.7 km, with an average of 11.3 km inside thunderclouds, and at an average of ∼2.5 km below cloud tops. These altitudes indicate that a total number of 1.7 × 1016to 2.6 × 1018gamma ray photons with energy greater than 1 MeV are required for an EIP‐TGF to be measured by spaceborne detectors. 
    more » « less
  4. Key Points A framework merging unsupervised clustering and supervised convolutional neural network (CNN) for lightning classification is developed Clustering of positive polarity energetic lightning radio pulses (>150 kA) identifies three processes: +EIPs (6%–7%), +NBEs, and +CGs CNNs detect 95.2% of manually identified +EIPs with up to 98.7% accuracy, enabling studying EIP‐TGF link with lower peak current (>50 kA) 
    more » « less
  5. Ground truth depth information is necessary for many computer vision tasks. Collecting this information is chal-lenging, especially for outdoor scenes. In this work, we propose utilizing single-view depth prediction neural networks pre-trained on synthetic scenes to generate relative depth, which we call pseudo-depth. This approach is a less expen-sive option as the pre-trained neural network obtains ac-curate depth information from synthetic scenes, which does not require any expensive sensor equipment and takes less time. We measure the usefulness of pseudo-depth from pre-trained neural networks by training indoor/outdoor binary classifiers with and without it. We also compare the difference in accuracy between using pseudo-depth and ground truth depth. We experimentally show that adding pseudo-depth to training achieves a 4.4% performance boost over the non-depth baseline model on DIODE, a large stan-dard test dataset, retaining 63.8% of the performance boost achieved from training a classifier on RGB and ground truth depth. It also boosts performance by 1.3% on another dataset, SUN397, for which ground truth depth is not avail-able. Our result shows that it is possible to take information obtained from a model pre-trained on synthetic scenes and successfully apply it beyond the synthetic domain to real-world data. 
    more » « less