Neural Network Training Research Articles

Comparison of Manual vs Artificial Intelligence-Based Muscle MRI Segmentation for Evaluating Disease Progression in Patients With CMT1A.

Intramuscular fat fraction (FF), assessed with quantitative MRI (qMRI), has emerged as one of the few responsive outcome measures in CMT1A patients. The main limitation for its use in future therapeutic trials is the time required for the manual segmentation of individual muscles. This study aimed to evaluate the accuracy and responsiveness of a fully automatic artificial intelligence (AI)-based segmentation pipeline to assess disease progression in a cohort of CMT1A patients over 1 year. Twenty CMT1A patients were included in this observational, prospective, longitudinal study. FF was measured twice a year apart using qMRI in the lower limbs. Individual muscle segmentation was performed fully automatically using a trained convolutional neural network with or without human quality check (QC). The corresponding results were compared with those obtained by fully manual (FM) segmentation using the Dice similarity coefficient (DSC). FF progression and its standardized response mean (SRM) were also computed in individual muscles over the single central slice and a 3D volume to define the most sensitive region of interest. AI-based segmentation showed excellent DSC values (>0.90). Significant global FF progression was observed at thigh (+0.71% ± 1.28%; p = 0.016) and leg (+1.73% ± 2.88%, p = 0.007) levels, similarly to that calculated using the FM technique (p = 0.363 and p = 0.634). FF progression of each individual muscle was comparable when computed from either the central slice or the 3D volume. The best SRM value (0.70) was obtained for the FF progression computed using the AI-based technique with human QC in the 3D volume at the leg level. The time required for fully automatic segmentation using AI with a QC was 10 hours for the entire data set compared with 90 hours for the FM. qMRI combined with AI-based segmentation can be considered as a process ready for assessing longitudinal FF changes in CMT1A patients. Given the slow FF progression at a thigh level and the large heterogeneity between muscles and individuals, FF should be quantified from a 3D volume at the leg level for longitudinal analyses. A QC performed after the AI-based segmentation is still advised given the increased SRM value.

Filtered partial differential equations: a robust surrogate constraint in physics-informed deep learning framework

Embedding physical knowledge into neural network (NN) training has been a hot topic. However, when facing the complex real world, most of the existing methods still strongly rely on the quantity and quality of observation data. Furthermore, the NNs often struggle to converge when the solution to the real equation is very complex. Inspired by large eddy simulation in computational fluid dynamics, we propose an improved method based on filtering. We analysed the causes of the difficulties in physics-informed machine learning, and proposed a surrogate constraint (filtered partial differential equation, FPDE) of the original physical equations to reduce the influence of noisy and sparse observation data. In the noise and sparsity experiment, the proposed FPDE models (which are optimized by FPDE constraints) have better robustness than the conventional PDE models. Experiments demonstrate that the FPDE model can obtain the same quality solution with 100 % higher noise and 12 % quantity of observation data of the baseline. Besides, two groups of real measurement data are used to show the FPDE improvements in real cases. The final results show that the FPDE still gives more physically reasonable solutions when facing the incomplete equation problem and the extremely sparse and high-noise conditions. The proposed FPDE constraint is helpful for merging real-world experimental data into physics-informed training, and it works effectively in two real-world experiments: simulating cell movement in scratches and blood velocity in vessels.

Reconfigurable Acceleration of Neural Networks: A Comprehensive Study of FPGA-based Systems

This paper explores the potential of Field-Programmable Gate Arrays (FPGAs) for accelerating both neural network inference and training. We present a comprehensive analysis of FPGA-based systems, encompassing architecture design, hardware implementation strategies, and performance evaluation. Our study highlights the advantages of FPGAs over traditional CPUs and GPUs for neural network workloads, including their inherent parallelism, reconfigurability, and ability to tailor hardware to specific network needs. We delve into various hardware implementation strategies, from direct mapping to dataflow architectures and specialized hardware blocks, examining their impact on performance. Furthermore, we benchmark FPGA-based systems against traditional platforms, evaluating inference speed, energy efficiency, and memory bandwidth. Finally, we explore emerging trends in FPGA-based neural network acceleration, such as specialized architectures, efficient memory management techniques, and hybrid CPU-FPGA systems. Our analysis underscores the significant potential of FPGAs for accelerating deep learning applications, particularly those requiring high performance, low latency, and energy efficiency.

NitroNet – a machine learning model for the prediction of tropospheric NO2 profiles from TROPOMI observations

Abstract. We introduce NitroNet, a deep learning model for the prediction of tropospheric NO2 profiles from satellite column measurements. NitroNet is a neural network trained on synthetic NO2 profiles from the regional chemistry and transport model WRF-Chem, which was operated on a European domain for the month of May 2019. This WRF-Chem simulation was constrained by in situ and satellite measurements, which were used to optimize important simulation parameters (e.g. the boundary layer scheme). The NitroNet model receives NO2 vertical column densities (VCDs) from the TROPOspheric Monitoring Instrument (TROPOMI) and ancillary variables (meteorology, emissions, etc.) as input, from which it reproduces NO2 concentration profiles. Training of the neural network is conducted on a filtered dataset, meaning that NO2 profiles showing strong disagreement (>20 %) with colocated TROPOMI column measurements are discarded. We present a first evaluation of NitroNet over a variety of geographical and temporal domains (Europe, the US West Coast, India, and China) and different seasons. For this purpose, we validate the NO2 profiles predicted by NitroNet against satellite, in situ, and MAX-DOAS (Multi-Axis Differential Optical Absorption Spectroscopy) measurements. The training data were previously validated against the same datasets. During summertime, NitroNet shows small biases and strong correlations with all three datasets: a bias of +6.7 % and R=0.95 for TROPOMI NO2 VCDs, a bias of −10.5 % and R=0.75 for AirBase surface concentrations, and a bias of −34.3 % to +99.6 % with R=0.83–0.99 for MAX-DOAS measurements. In comparison to TROPOMI satellite data, NitroNet even shows significantly lower errors and stronger correlation than a direct comparison with WRF-Chem numerical results. During wintertime considerable low biases arise because the summertime/late-spring training data are not fully representative of all atmospheric wintertime characteristics (e.g. longer NO2 lifetimes). Nonetheless, the wintertime performance of NitroNet is surprisingly good and comparable to that of classic regional chemistry and transport models. NitroNet can demonstrably be used outside the geographic and temporal domain of the training data with only slight performance reductions. What makes NitroNet unique when compared to similar existing deep learning models is the inclusion of synthetic model data, which offers important benefits: due to the lack of NO2 profile measurements, models trained on empirical datasets are limited to the prediction of surface concentrations learned from in situ measurements. NitroNet, however, can predict full tropospheric NO2 profiles. Furthermore, in situ measurements of NO2 are known to suffer from biases, often larger than +20 %, due to cross-sensitivities to photooxidants, which other models trained on empirical data inevitably reproduce.

INTELLIGENT MONITORING OF THE PHYSIOLOGICAL STATE OF AGRICULTURAL PRODUCTS USING UAV

The article discusses the technology for automated neural network monitoring of the vineyard’s physiological condition. Images of leaves, obtained using an unmanned aerial vehicle (UAV), are the main indicator of the physiological vineyard’s condition. The proposed solution is based on the integrated use of convolutional neural network method (CNN) and machine vision technologies. To determine the optimal neural network (NN) model, a variant analysis was carried out. In accordance with its results, the YOLOv7 model was chosen, which satisfies the introduced time limit and provides the required detection quality. The training of the YOLOv7 neural network was implemented in the Python environment using the PyTorch framework and the OpenCV computer vision library. The dataset consisting of 6320 images of grape leaves (including healthy and diseased ones) has been used for neural network training. The obtained results showed that the detection accuracy is at least 91%. Visualization of monitoring results has been carried out using heatmap, allowing to obtain information about vineyard physiological condition in dynamics. The proposed mathematical model allows to calculate the monitored vineyard’s area made by one complex per day. The obtained results showed that effective monitoring area using one DJI Phantom 4 UAV per day is 2.5 hectares.

Abstract Su803: Sudden Cardiac Arrest Prediction via Deep Learning Electrocardiogram Analysis

Background: Sudden cardiac arrest (SCA) is a potentially fatal event that often occurs without prior indications. Hypothesis: It was hypothesized that electrocardiogram (ECG) data could be used in conjunction with Deep Learning (DL) algorithms to predict the occurrence of SCA’s. Aims: To improve SCA outcomes and enable preventative strategies, the ECG data from the Nightingale Open Science - Subtyping Cardiac Arrest dataset were utilized to create a DL model to serve as a screening tool for sudden cardiac arrest. Methods: A publicly-available dataset containing ten seconds of 12-lead ECGs from individuals who had a SCA and persons who did not have a SCA, and information about time from ECG to arrest, and age and sex were utilized for individualized prediction of SCA. Subsets of the initial data were used to train, validate, and test deep convolution neural network models that processed cardiac cycles from single-lead ECG voltage waveforms. Results: The cohort comprised 221 (mean age: 68.5 and #females: 75) patients who experienced SCA and 1046 controls (mean age: 69.7 and #females: 497) who did not have a cardiac arrest. The SCA ECG-DL base model, with ECGs within one day and with age and sex data, had an area under the receiver operating curve (AUROC) of 0.774 during hold-out testing for discrimination of subsequent SCA. With sensitivity set at 95%, the base model specificity of the SCA ECG-DL model was 31%. Gradient-weighted class activation mapping showed that the SCA ECG-DL model relied most heavily on the QRS complex to make predictions. The performance of the SCA ECG-DL model was similar removing age and sex data (AUROC 0.775) and less reliable when predicting SCA within one month (AUROC 0.734) or between one month and one year (AUROC 0.574). Conclusion: DL interpretation of ECG data is a promising means of screening for SCA, and this method explains differences in SCA’s due to age and sex. Model performance improved when ECGs were obtained at shorter intervals to the incident SCA event, and SCA prediction was more dependent upon QRS complex data than other temporal segments of the ECG.

Water Supply Pipeline Failure Evaluation Model Based on Particle Swarm Optimization Neural Network

The degradation and failure of the urban water supply network may lead to serious safety hazards, including pipe breaks, water supply interruptions, water resource losses, and contaminant intrusions. The risk evaluation of water supply pipeline failure in a distribution network is a challenging task, because most of the available data cannot fully reflect pipeline failure events and many of the mechanisms still cannot be fully understood. Therefore, a predictive model is urgently needed to assess pipeline failure risk based on available data. In this paper, based on the traditional risk assessment theory, seven main factors affecting pipeline failure are selected and scored, and then a pipeline failure model is established by using the particle swarm optimization (PSO) neural network. The model uses the neural network training of historical data to evaluate the failure of the water supply pipeline, and the PSO is used to optimize the neural network to effectively improve the training time and accuracy. The model error and correlation coefficient are 0.003 and 0.987, respectively. The proposed model can be used as a powerful support tool to assist infrastructure managers and pipeline maintainers in their plans and decision-making.

Automatic assessment of nailfold capillaroscopy software: a pilot study

IntroductionCapillaroscopy is a simple method of nailfold capillary imaging, used to diagnose diseases from the systemic sclerosis spectrum. However, the assessment of the capillary image is time-consuming and subjective. This makes it difficult to use for a detailed comparison of studies assessed by various physicians. This pilot study aimed to validate software used for automatic capil­lary counting and image classification as normal or pathological.Material and methodsThe study was based on the assessment of 200 capillaroscopic images obtained from patients suffering from systemic sclerosis or scleroderma spectrum diseases and healthy people. Dinolite MEDL4N Pro was used to perform capillaroscopy. Each image was analysed manually and described using working software. The neural network was trained using the fast.ai library (based on PyTorch). The ResNet-34 deep residual neural network was chosen; 10-fold cross-validation with the validation and test set was performed, using the Darknet-YoloV3 state of the art neural network in a GPU-optimized (P5000 GPU) environment. For the calculation of 1 mm capillaries, an additional detection mechanism was designed.ResultsThe results obtained under neural network training were compared to the results obtained in manual analysis. The sensitivity of the automatic tool relative to manual assessment in classification of correct vs. pathological images was 89.0%, specificity 89.4% for the training group, in validation 89.0% and 86.9% respectively. For the average number of capillaries in 1 mm the precision of real images detected within the region of interest was 96.48%.ConclusionsThe pilot software for fully automatic capillaroscopic image assessment can be a useful tool for the rapid classification of a normal and altered capillaroscopy pattern. In addition, it allows one to quickly calculate the number of capillaries. In the future, the tool will be developed and will make it possible to obtain full imaging characteristics independent of the experience of the examiner.

A Machine Learning Approach for Breast Cancer Risk Prediction in Digital Mammography

Breast cancer is among the most prevalent cancers in the female population globally. Therefore, screening campaigns as well as approaches to identify patients at risk are particularly important for the early detection of suspect lesions. This study aims to propose a workflow for the automatic classification of patients based on one of the most relevant risk factors in breast cancer, which is represented by breast density. The proposed classification methodology takes advantage of the features automatically extracted from mammographic images, as digital mammography represents the major screening tool in women. Textural features were extracted from the breast parenchyma through a radiomics approach, and they were used to train different machine learning algorithms and neural network models to classify the breast density according to the standard Breast Imaging Reporting and Data System (BI-RADS) guidelines. Both binary and multiclass tasks have been carried out and compared in terms of performance metrics. Preliminary results show interesting classification accuracy (93.55% for the binary task and 82.14% for the multiclass task), which are promising compared to the current literature. As the proposed workflow relies on straightforward and computationally efficient algorithms, it could serve as a basis for a fast-track protocol for the screening of mammograms to reduce the radiologists’ workload.

Autonomous UAV Chasing with Monocular Vision: A Learning-Based Approach

In recent years, unmanned aerial vehicles (UAVs) have shown significant potential across diverse applications, drawing attention from both academia and industry. In specific scenarios, UAVs are expected to achieve formation flying without relying on communication or external assistance. In this context, our work focuses on the classic leader-follower formation and presents a learning-based UAV chasing control method that enables a quadrotor UAV to autonomously chase a highly maneuverable fixed-wing UAV. The proposed method utilizes a neural network called Vision Follow Net (VFNet), which integrates monocular visual data with the UAV’s flight state information. Utilizing a multi-head self-attention mechanism, VFNet aggregates data over a time window to predict the waypoints for the chasing flight. The quadrotor’s yaw angle is controlled by calculating the line-of-sight (LOS) angle to the target, ensuring that the target remains within the onboard camera’s field of view during the flight. A simulation flight system is developed and used for neural network training and validation. Experimental results indicate that the quadrotor maintains stable chasing performance through various maneuvers of the fixed-wing UAV and can sustain formation over long durations. Our research explores the use of end-to-end neural networks for UAV formation flying, spanning from perception to control.

Simplifying Distributed Neural Network Training on Massive Graphs: Randomized Partitions Improve Model Aggregation

Distributed Graph Neural Network (GNN) training facilitates learning on massive graphs that surpass the storage and computational capabilities of a single machine. Traditional distributed frameworks strive for performance parity with centralized training by maximally recovering cross-instance node dependencies, relying either on inter-instance communication or periodic fallback to centralized training. However, these processes create overhead and constrain the scalability of the framework. In this work, we propose a streamlined framework for distributed GNN training that eliminates these costly operations, yielding improved scalability, convergence speed, and performance over state-of-the-art approaches. Our framework (1) comprises independent trainers that asynchronously learn local models from locally-available parts of the training graph, and (2) synchronizes these local models only through periodic (time-based) model aggregation. Contrary to prevailing belief, our theoretical analysis shows that it is not essential to maximize the recovery of cross-instance node dependencies to achieve performance parity with centralized training. Instead, our framework leverages randomized assignment of nodes or super-nodes (i.e., collections of original nodes) to partition the training graph in order to enhance data uniformity and minimize discrepancies in gradient and loss function across instances. Experiments on social and e-commerce networks with up to 1.3 billion edges show that our proposed framework achieves state-of-the-art performance and 2.31x speedup compared to the fastest baseline, despite using less training data.

Dataset preparation for CROHME 2019 for training a neural network

Subject of research. This article examines the process of data preparation for training neural network models that address the task of recognizing complex handwritten mathematical expressions and converting them into typesetting systems like LaTeX or MathML. The dataset under study is CROHME 2019, which contains images of handwritten mathematical expressions and their corresponding LaTeX annotations. The objective of this research is to develop and describe a data preparation process that ensures correct model training and minimizes errors in numerical labels and transformations. The study focuses on the following key stages: loading and preprocessing images, parsing LaTeX annotations, linking images with their corresponding annotations, normalizing pixel values, creating and verifying the tokenizer vocabulary, and preparing the tokenizer itself. The methods include data analysis and processing using specialized tools for working with images and text annotations. The main outcome of this work is the creation of a dataset ready for use in training the neural network model. The prepared dataset ensures accurate alignment between images and annotations, as well as correct conversion of mathematical expressions into LaTeX code.

PIPET: A Pipeline to Generate PET Phantom Datasets for Reconstruction Based on Convolutional Neural Network Training

There has been a strong interest in using neural networks to solve several tasks in PET medical imaging. One of the main problems faced when using neural networks is the quality, quantity, and availability of data to train the algorithms. In order to address this issue, we have developed a pipeline that enables the generation of voxelized synthetic PET phantoms, simulates the acquisition of a PET scan, and reconstructs the image from the simulated data. In order to achieve these results, several pieces of software are used in the different steps of the pipeline. This pipeline solves the problem of generating diverse PET datasets and images of high quality for different types of phantoms and configurations. The data obtained from this pipeline can be used to train convolutional neural networks for PET reconstruction.

Neural Thermodynamic Integration: Free Energies from Energy-Based Diffusion Models.

Thermodynamic integration (TI) offers a rigorous method for estimating free-energy differences by integrating over a sequence of interpolating conformational ensembles. However, TI calculations are computationally expensive and typically limited to coupling a small number of degrees of freedom due to the need to sample numerous intermediate ensembles with sufficient conformational-space overlap. In this work, we propose to perform TI along an alchemical pathway represented by a trainable neural network, which we term Neural TI. Critically, we parametrize a time-dependent Hamiltonian interpolating between the interacting and noninteracting systems and optimize its gradient using a score matching objective. The ability of the resulting energy-based diffusion model to sample all intermediate ensembles allows us to perform TI from a single reference calculation. We apply our method to Lennard-Jones fluids, where we report accurate calculations of the excess chemical potential, demonstrating that Neural TI reproduces the underlying changes in free energy without the need for simulations at interpolating Hamiltonians.

Corrigendum: Gradient-free training of recurrent neural networks using random perturbations

Corrigendum: Gradient-free training of recurrent neural networks using random perturbations

Pole structure of Pψ N(4312)+ via machine learning and uniformized S-matrix

Abstract We probed the pole structure of the Pψ N(4312)+ using a trained deep neural network. The training dataset was generated using uniformized independent S-matrix poles to ensure that the resulting interpretation is as model-independent as possible. To prevent possible ambiguity in the interpretation of the pole structure, we included the contribution from the off-diagonal elements of the S-matrix. Our trained model favors a possible three-pole structure of Pψ N(4312)+ signal, with one pole on each of the unphysical sheets. The two poles on two different Riemann sheets can be associated with a pole-shadow pair which is a characteristic of a true resonance. On the other hand, the last isolated pole is most likely associated with the coupled-channel effect in the meson-baryon system. The combined effect of these poles produced a peak below the ΣcD threshold which mimics the line shape of a hadronic molecule.

Development of an automated artificial intelligence-based system for urogenital schistosomiasis diagnosis using digital image analysis techniques and a robotized microscope.

Urogenital schistosomiasis is considered a Neglected Tropical Disease (NTD) by the World Health Organization (WHO). It is estimated to affect 150 million people worldwide, with a high relevance in resource-poor settings of the African continent. The gold-standard diagnosis is still direct observation of Schistosoma haematobium eggs in urine samples by optical microscopy. Novel diagnostic techniques based on digital image analysis by Artificial Intelligence (AI) tools are a suitable alternative for schistosomiasis diagnosis. Digital images of 24 urine sediment samples were acquired in non-endemic settings. S. haematobium eggs were manually labeled in digital images by laboratory professionals and used for training YOLOv5 and YOLOv8 models, which would achieve automatic detection and localization of the eggs. Urine sediment images were also employed to perform binary classification of images to detect erythrocytes/leukocytes with the MobileNetv3Large, EfficientNetv2, and NasNetLarge models. A robotized microscope system was employed to automatically move the slide through the X-Y axis and to auto-focus the sample. A total number of 1189 labels were annotated in 1017 digital images from urine sediment samples. YOLOv5x training demonstrated a 99.3% precision, 99.4% recall, 99.3% F-score, and 99.4% mAP0.5 for S. haematobium detection. NasNetLarge has an 85.6% accuracy for erythrocyte/leukocyte detection with the test dataset. Convolutional neural network training and comparison demonstrated that YOLOv5x for the detection of eggs and NasNetLarge for the binary image classification to detect erythrocytes/leukocytes were the best options for our digital image database. The development of low-cost novel diagnostic techniques based on the detection and identification of S. haematobium eggs in urine by AI tools would be a suitable alternative to conventional microscopy in non-endemic settings. This technical proof-of-principle study allows laying the basis for improving the system, and optimizing its implementation in the laboratories.

AUTOMATED INDOOR ACTIVITY MONITORING FOR ELDERLY AND VISUALLY IMPAIRED PEOPLE USING QUANTUM SALP SWARM ALGORITHM WITH MACHINE LEARNING

Elderly and vision-impaired indoor activity monitoring uses sensor technology to monitor interaction and movement in the living area. This system can identify deviations from regular patterns, provide alerts, and ensure safety in case of emergencies or potential risks. These solutions enhance quality of life by promoting independent living while offering peace of mind to loved ones and caregivers. Visual impairments have become the chief part of these problems over the past few years. Usually, visually impaired people seek help from others to keep doing their daily tasks. An automated human activity recognition (HAR) technique could enhance a vision-impaired person’s transaction activity and safe movement. HAR using deep learning (DL) includes training neural networks to automatically detect and classify various activities carried out by humans based on sensor information. By leveraging DL approaches, the HAR system could accurately identify and classify activities, including sitting, standing, walking, or running, from the information captured by sensors or wearable devices. This system is integral to smart home automation, healthcare, and sports analytics. This study develops an Automated Fractal Quantum Salp Swarm Algorithm with Machine Learning based HAR (QSSA-MLHAR) for assisting elderly and visually impaired people. The QSSA-MLHAR technique uses sensory input data to identify human activities in diverse classes. In the QSSA-MLHAR technique, the min–max scalar is primarily used to scale the input data. Besides, the QSSA-MLHAR technique applies a mixed extreme learning machine (M-ELM) model to classify various human actions. The QSSA-based hyperparameter tuning process is used to improve the detection results of the M-ELM system. A wide range of simulations was performed to ensure the better performance of the QSSA-MLHAR method. The comprehensive outcomes assured that the QSSA-MLHAR technique performs well compared to other models.

Systematic Analysis of Low-Precision Training in Deep Neural Networks: Factors Influencing Matrix Computations

As Deep Neural Networks (DNNs) continue to increase in complexity, the computational demands of their training have become a significant bottleneck. Low-precision training has emerged as a crucial strategy, wherein full-precision values are quantized to lower precisions, reducing computational overhead while aiming to maintain model accuracy. While prior research has primarily focused on minimizing quantization noise and optimizing performance for specific models and tasks, a comprehensive understanding of the general principles governing low-precision computations across diverse DNN architectures has been lacking. In this paper, we address this gap by systematically analyzing the factors that influence low-precision matrix computations, which are fundamental to DNN training. We investigate three critical factors—accumulation in matrix calculations, the frequency of element usage, and the depth of matrices within the model—and their impact on low-precision training. Through controlled experiments on standard models, as well as customized experiments designed to isolate individual factors, we derive several key insights: layers with higher accumulation and matrices with lower usage frequencies demonstrate greater tolerance to low-precision noise, without significantly compromising the stability of model training. Additionally, while the depth of matrices influences the stability of matrix operations to some extent, it does not have a noticeable effect on the overall training outcomes. Our findings contribute to the development of generalizable principles for low-precision training, offering a systematic framework applicable across various DNN architectures. We provide empirical evidence supporting the strategic allocation of training bit-widths based on the analyzed factors, thereby enhancing the efficiency and effectiveness of DNN training in resource-constrained environments.

Anomalously High Apparent Young's Modulus of Ultrathin Freestanding PZT Films Revealed by Machine Learning Empowered Nanoindentation.

The mechanical properties at small length scales are not only significant for understanding the intriguing size-dependent behaviors but also critical for device applications. Nanoindentation via atomic force microscopy is widely used for small-scale mechanical testing, yet determining the Young's modulus of quasi-2D films from freestanding force-displacement curve of nanoindentation remains challenging, complicated by both bending and stretching that are highly nonlinear. To overcome these difficulties, a machine learning model is developed based on the back propagation (BP) neural network and finite element training to accurately determine the Young's modulus, pretension, and thickness of freestanding films from nanoindentation force-displacement curves simultaneously, improving the computational efficiency by two orders of magnitude over conventional brute force curve fitting. Using this technique, anomalously high apparent Young's modulus is discovered of 2.8nm thick PbZr0.2Ti0.8O3 (PZT) film as large as 229.4±12.1GPa, much higher than the bulk value. The enhancement can be attributed to the strain gradient-induced flexoelectric effect, and the corresponding flexoelectric coefficient is estimated to be ≈200nCm-1. The method is developed to enable an artificial intelligence (AI) system to study the mechanical properties of a wide range of low-dimensional materials.