Deep Artificial Neural Network Model Research Articles

Thy-DAMP: deep artificial neural network model for prediction of thyroid cancer mortality.

Despite the rising incidence of differentiated thyroid cancer (DTC), mortality rates have remained relatively low yet crucial for effective patient management. This study aims to develop a deep neural network capable of predicting mortality in patients with differentiated thyroid cancer. Leveraging data from the Surveillance, Epidemiology, and End Results (SEER) database, we developed Thy-DAMP (Deep Artificial Neural Network Model for Prediction of Thyroid Cancer) to forecast mortality in DTC patients. The dataset comprised demographic, histologic, and staging information. Following data normalization and feature encoding, the dataset was partitioned into training, testing, and validation subsets, with model hyperparameters fine-tuned via cross-validation. Among the 63,513 patients, the mean age was 48.22 years (SD = 14.8), with 77.32% being female. Papillary carcinoma emerged as the predominant subtype, representing 62.94% of cases. The majority presented with stage I disease (73.96%). Thy-DAMP demonstrated acceptable performance metrics on both the test and validation sets. Sensitivity was 83.24% (95% CI 76.95-88.40%), specificity was 93.53% (95% CI 93.01-94.02%), and accuracy stood at 93.33% (95% CI 92.82-93.83%). The model exhibited a positive predictive value of 19.76% (95% CI 18.20-21.42%) and a negative predictive value of 99.66% (95% CI 99.53-99.75%). Additionally, Thy-DAMP demonstrated a positive likelihood ratio of 12.86 (95% CI 11.62-14.23), a negative likelihood ratio of 0.18 (95% CI 0.13-0.25), and an area under the receiver operating characteristic curve (AUROC) of 0.95. The model was externally validated on a separate dataset with nearly identical performance. Thy-DAMP showcases considerable promise in accurately predicting mortality in DTC patients, leveraging limited set of patient data.

Using Machine Learning Algorithms to Develop a Predictive Model for Computing the Maximum Deflection of Horizontally Curved Steel I-Beams

Horizontally curved steel I-beams exhibit a complicated mechanical response as they experience a combination of bending, shear, and torsion, which varies based on the geometry of the beam at hand. The behaviour of these beams is therefore quite difficult to predict, as they can fail due to either flexure, shear, torsion, lateral torsional buckling, or a combination of these types of failure. This therefore necessitates the usage of complicated nonlinear analyses in order to accurately model their behaviour. Currently, little guidance is provided by international design standards in consideration of the serviceability limit states of horizontally curved steel I-beams. In this research, an experimentally validated dataset was created and was used to train numerous machine learning (ML) algorithms for predicting the midspan deflection at failure as well as the failure load of numerous horizontally curved steel I-beams. According to the experimental and numerical investigation, the deep artificial neural network model was found to be the most accurate when used to predict the validation dataset, where a mean absolute error of 6.4 mm (16.20%) was observed. This accuracy far surpassed that of Castigliano’s second theorem, where the mean absolute error was found to be equal to 49.84 mm (126%). The deep artificial neural network was also capable of estimating the failure load with a mean absolute error of 30.43 kN (22.42%). This predictive model, which is the first of its kind in the international literature, can be used by professional engineers for the design of curved steel I-beams since it is currently the most accurate model ever developed.

A Deep Artificial Neural Network Model for Predicting the Mechanical Behavior of Triply Periodic Minimal Surfaces under Damage Loading

A Deep Artificial Neural Network Model for Predicting the Mechanical Behavior of Triply Periodic Minimal Surfaces under Damage Loading

A Hierarchy of Variables That Influence the Force–Velocity Profile of Acrobatic Gymnasts: A Tool Based on Artificial Intelligence

Jumping performance is considered an overall indicator of gymnastics ability. Acrobatic Gymnastics involves base and top gymnasts, considering the type of training that is performed and the distinct anthropometric traits of each gymnast. This work aims to investigate a hierarchy of variables that influence the force–velocity (F-V) profile of top and base acrobatic gymnasts through a deep artificial neural network model. Twenty-eight first division and elite acrobatic gymnasts (eleven tops and seventeen bases) performed two evaluations to assess the F-V profile during the Countermovement Jump and its mechanical variables, using My Jump 2 (a total of 56 evaluations). A training background survey and anthropometric assessments were conducted. The final model (R = 0.97) showed that the F-V imbalance (F-Vimb) increases with higher force and decreases with higher maximal power, fat percentage, velocity, and height. Coaches should prioritize the development of force, followed by maximal power, and velocity for the optimization of gymnasts’ F-Vimb. For training planning, the influences of body mass and push-off height are higher for the bases, and the influences of years of practice and competition level are higher for the tops.

Improved deep artificial neural network-powered prediction of extreme mechanical performances of fractal architectures with high hierarchical rank

This work explores the mechanical performances, including anisotropy, wave propagation, and buckling resistance capability of the Menger sponge and Jerusalem fractal structures, with the high rank, using an improved deep artificial neural network (ANN) model inserted by a loop with multiple conditions. Where, the effective mechanical attributes, including Young’s modulus, shear modulus, and Poisson’s ratio of the Menger sponge structure with rank 4 are attained using high-performance computer cluster (HPCC) that are then used to train and also validate the ANN model. The accuracy of the deep ANN model trained by data of fractal structures with and without rank 4 is verified by the numerical model and experiment with the good agreement found or the maximum percent difference being 15.6% where the numerical and experimental data don’t belong to its training samples. Remarkably, the deep ANN model demonstrates an extremely low cost, but very fast compared with the numerical and experimental works. Namely, the improved deep ANN model implemented by a regular computer can accurately predict the mechanical attributes of Menger sponge fractal structures with full ranks in 3 min. While it takes a week for numerical work using HPCC or experiment method to produce the same mechanical properties of the Menger sponge with only rank 4, displaying at least thousands of times slower than the ANN model. The ANN results indicate that the onset of anisotropy of considered fractal structures is from rank 1, and the reversal of anisotropy will take place upon an increase in the rank while such reversal is not seen for the case of lattice structure. The anisotropy reversal could be one of main reasons why the fractal structure tends to have a high rank in nature. In addition, the buckling stress and phase wave velocity are observed to decrease with rising rank, with a nonlinear pattern.

核検知及び核セキュリティ事案初動対応を支援する深層ニューラルネットワークを用いた核種判定アルゴリズム

Rapid and precise radioisotope identification in the scene of nuclear detection and nuclear security incidents is one of the challenging issues for the prompt response to the detection alarm or the incidents. A radioisotope identification algorithm using a deep artificial neural network model applicable to handheld γ-ray detectors has been proposed in the present paper. The proposed algorithm automatically identifies gamma-emitting radioisotopes based on the count contribution ratio (CCR) from each of them estimated by the deep artificial neural network model trained by simulated γ-ray spectra. The automated radioisotope identification algorithm can support first responders of nuclear detection and nuclear security incidents without sufficient experience and knowledge in radiation measurement. The authors tested the performance of the proposed algorithm using two different types of deep artificial neural network models in application to handheld detectors having high or low energy resolution. The proposed algorithm showed high performance in identifying artificial radioisotopes for actually measured γ-ray spectra. It was also confirmed that the algorithm is applicable to identifying 235U and automated uranium categorization by analyzing estimated CCRs by the deep artificial neural network models. The authors also compared the performance of the proposed algorithm with a conventional radioisotope identification method and discussed promising ways to improve the performance of the algorithm using the deep artificial neural network.

Color Image Encryption using Chaotic Algorithm and 2D Sin-Cos Henon Map for High Security

In every form of electronic communication, data security must be an absolute top priority. As the prevalence of Internet and other forms of electronic communication continues to expand, so too does the need for visual content. There are numerous options for protecting transmitted data. It's important that the transmission of hidden messages in images remain unnoticed to avoid raising any red flags. In this paper, we propose a new deep learning-based image encryption algorithm for safe image retrieval. The proposed algorithm employs a deep artificial neural network model to extract features via sample training, allowing for more secure image network transmission. The algorithm is incorporated into a deep learning-based image retrieval process with Convolution Neural Networks(CNN), improving the efficiency of retrieval while also guaranteeing the security of ciphertext images. Experiments conducted on five different datasets demonstrate that the proposed algorithm vastly improves retrieval efficiency and strengthens data security. Also hypothesised a 2D Sin-Cos-Henon (2D-SCH)-based encryption algorithm for highly secure colour images. We demonstrate that this algorithm is secure against a variety of attacks and that it can encrypt all three colour channels of an image simultaneously.

Forecasting road traffic accident using deep artificial neural network approach in case of Oromia Special Zone

Millions of people are dying, and billions of properties are damaged by road traffic accidents each year worldwide. In the case of our country Ethiopia, the effect of traffic accidents is even more by causing injuries, death, and property damage. Forecasting road traffic accident and predicting the severity of road traffic accident contributes a role indirectly in reducing road traffic accidents. This study deals with forecasting the number of accident and prediction of the severity of an accident in the Oromia Special Zone using deep artificial neural network models. Around 6170 road traffic accidents data are collected from Oromia Police Commission Excel data and Oromia Special Zone Traffic Police Department; the dataset consists of accidents in the Special Zone of Oromia Districts (Woredas) from 2005 to 2012 with 15 accidents attributes. 5928 or (80%) of the dataset was used for the training model, and 1482 or (20%) of the dataset was used for the testing model. This study proposed six different neural network architectures such as backpropagation neural network (BPNN), feedforward neural network (FFNN), multilayer perceptron neural network (MLPNN), recurrent neural networks (RNN), radial basis function neural network (RBFNN) and long short-term memory (LSTM) models for accident severity prediction and the LSTM model for a time serious forecasting of number accidents within specified years. The models will take input data, classify accidents, and predict the severity of an accident. Accident predictor GUI has been created using Python Tkinter library for easy accident severity prediction. According to the model performance results, RNN model showed the best prediction accuracy of 97.18%, whereas MLP, LTSM, RBFNN, FFNN, and BPNN models showed the accuracy of 97.13%, 91.00%, 87.00%, 80.56%, and 77.26%, respectively. LTSM model forecasted accident for three years which is 3555 where the actual accident number is 3561. The prediction and forecast result obtained from the model will be helpful in planning and management of road traffic accidents.

Investigating the role of psychological needs in predicting the educational sustainability of Metaverse using a deep learning-based hybrid SEM-ANN technique

ABSTRACT Metaverse is an immersive three-dimensional (3D) virtual world inhabited by avatars beyond the physical realm. The COVID-19 pandemic has disrupted the education system and the need to accelerate the digitalization of education has received a lot of attention. Metaverse can be an alternative solution for sociocultural interaction and to continue education. Accordingly, this study aimed to identify key factors in predicting the educational sustainability of Metaverse. The study empirically tested the role of psychological needs (i.e. hedonic motivation, affiliation, dominance, achievement, and autonomy) in predicting educational sustainability. The study employed a hybrid method integrating covariance-based structural-equation-modeling (CB-SEM) and deep artificial neural network (ANN) model. The CB-SEM results indicated that the need for autonomy and hedonic motivation significantly predicted the educational sustainability of Metaverse. Further, deep ANN models indicated that hedonic motivation was the most important input factor, followed by autonomy, affiliation, dominance, and achievement. Findings have practical implications for developers of the Metaverse environments and theoretical contributions for educators who manage and implement such environments.

Reducing smearing effect in contribution plots and improving fault detection via polynomial approximated isomap embeddings

AbstractDimension reduction is an essential method used in multivariate statistical process monitoring for fault detection and diagnosis. Principal component analysis (PCA) and independent component analysis (ICA) are the most frequently used linear dimensional reduction tools, and the contribution plot is the most popular fault isolation method in the absence of any prior information on the faults. These methods, however, come with their shortcomings. The fault detection capability of linear methods may not be sufficient for non‐linear processes, and smearing effect is known to deteriorate the diagnostics obtained from contribution plots. While the fault detection rate may be increased by kernelized methods or deep artificial neural network models, tuning data‐dependent hyperparameter(s) and network structure with limited historical data is not an easy task. Furthermore, the resulting non‐linear models often do not directly possess fault isolation capability. In the current study, we aim to devise a novel method named ICApIso‐PCA, which offers non‐linear fault detection and isolation in a rather straightforward manner. The rationale of ICApIso‐PCA mainly involves building a non‐linear scores matrix, composed of principal component scores and high‐order polynomial approximated isomap embeddings, followed by implementation of the ICA‐PCA algorithm on this matrix. Applications on a toy dataset and the Tennessee Eastman plant show that the I2 index from ICApIso‐PCA yields a high fault detection rate and offers accurate contribution plots with diminished smearing effects compared to those from traditional monitoring methods. Easy implementation and the potential for future research are further advantages of the proposed method.

Modeling Bitcoin Prices using Signal Processing Methods, Bayesian Optimization, and Deep Neural Networks.

Bitcoin is a volatile financial asset that runs on a decentralized peer-to-peer Blockchain network. Investors need accurate price forecasts to minimize losses and maximize profits. Extreme volatility, speculative nature, and dependence on intrinsic and external factors make Bitcoin price forecast challenging. This research proposes a reliable forecasting framework by reducing the inherent noise in Bitcoin time series and by examining the predictive power of three distinct types of predictors, namely fundamental indicators, technical indicators, and univariate lagged prices. We begin with a three-step hybrid feature selection procedure to identify the variables with the highest predictive ability, then use Hampel and Savitzky-Golay filters to impute outliers and remove signal noise from the Bitcoin time series. Next, we use several deep neural networks tuned by Bayesian Optimization to forecast short-term prices for the next day, three days, five days, and seven days ahead intervals. We found that the Deep Artificial Neural Network model created using technical indicators as input data outperformed other benchmark models like Long Short Term Memory, Bi-directional LSTM (BiLSTM), and Convolutional Neural Network (CNN)-BiLSTM. The presented results record a high accuracy and outperform all existing models available in the past literature with an absolute percentage error as low as 0.28% for the next day forecast and 2.25% for the seventh day for the latest out of sample period ranging from Jan 1, 2021, to Nov 1, 2021. With contributions in feature selection, data-preprocessing, and hybridizing deep learning models, this work contributes to researchers and traders in fundamental and technical domains.

Image Security Retrieval Based on Chaotic Algorithm and Deep Learning

At present, most image retrieval methods are based on plain-text images which poses a threat to some professional fields, such as medicine, military, and finance. In order to achieve greater security for the network transmission security of the image, we establish a deep artificial neural network model to extract features by sample training. Then an image-encryption algorithm that matches and secures image retrieval is designed and integrated into an image-retrieval process based on deep learning. Experiments on multiple authoritative datasets show that the proposed algorithm can not only achieve secure retrieval of ciphertext images, but also improve retrieval efficiency obviously. Specifically, the experimental results on five data sets show that compared with the average performance of 16 comparison algorithms, each evaluation indicator has been significantly improved in our research, with <i>Pre</i> increased by 12.54% - 88.20%, <i>Rec</i> increased by 1.46% - 10.95%, <i>F1</i> increased by 2.86% - 13.55% and <i>mAP</i> increased by 16.64% - 82.47%. Futhermore, the successful realization of the ciphertext retrieval provides some reference for the information security retrieval research.

An integrated model for the usage and acceptance of stickers in WhatsApp through SEM-ANN approach

This analysis integrates the “technology acceptance model (TAM)” with the “use of gratifications theory (U&amp;G)” to develop an embedded model that predicts the use and satisfaction of emotional icons called stickers through WhatsApp. The explanation for combining these two theories is that U&amp;G offers accurate information and a thorough knowledge of use, while TAM theory has been firmly established in several technical implementations. A newly developed hybrid analysis procedure has been applied within this research. Using an artificial neural network (ANN), and the structural equation model (SEM) have been combined. The research also uses the importance-performance map analysis (IPMA) to present each factor’s performance as well as importance. The ANN and IPMA research have both indicated that for sticker use intention, a highly essential predictor is Socialization. An online questionnaire survey was developed to assess the recommended model. The intention to use stickers was significantly affected by “Socialization, Self Presentation, Enjoyment, Novelty, Unique Function, Perceived Ease of Use, and Perceived Usefulness”. The research's main achievement is the convergence of two separate theories into a single conceptualization to accurately calculate the TAM components when it comes to the usage of stickers in WhatsApp. Theoretically, the recommended model provides enough insight for aspects which affect the intention to use stickers with relevance to the socialization’s factors considering interpersonal aspects. Practically, the higher education decision-makers along with professionals would extract variables that are important as compared to others and policies would be developed accordingly. The deep ANN model competence has been analyzed within the research to decide upon the non-linear associations between variables of the theoretical model, methodologically.

On the Neural Network Solution of One-Dimensional Wave Problem

Artificial neural network has become an emerging popular method to handle various problems, especially in case where it has deep multiple neural layers. In this study, we use a deep artificial neural network model to solve one-dimensional wave equation, without any external datasets. Different type of boundary conditions, i.e., Dirichlet, Neumann, and Robin, are used. We analyze the model learning capabilities in a set of settings, such as data setup and the model width and depth. We also present some discussions of advantages and disadvantages of the model in comparison with other matured existing techniques to solve wave equation.

A Deep Learning Algorithm for High-Dimensional Exploratory Item Factor Analysis.

Marginal maximum likelihood (MML) estimation is the preferred approach to fitting item response theory models in psychometrics due to the MML estimator's consistency, normality, and efficiency as the sample size tends to infinity. However, state-of-the-art MML estimation procedures such as the Metropolis-Hastings Robbins-Monro (MH-RM) algorithm as well as approximate MML estimation procedures such as variational inference (VI) are computationally time-consuming when the sample size and the number of latent factors are very large. In this work, we investigate a deep learning-based VI algorithm for exploratory item factor analysis (IFA) that is computationally fast even in large data sets with many latent factors. The proposed approach applies a deep artificial neural network model called an importance-weighted autoencoder (IWAE) for exploratory IFA. The IWAE approximates the MML estimator using an importance sampling technique wherein increasing the number of importance-weighted (IW) samples drawn during fitting improves the approximation, typically at the cost of decreased computational efficiency. We provide a real data application that recovers results aligning with psychological theory across random starts. Via simulation studies, we show that the IWAE yields more accurate estimates as either the sample size or the number of IW samples increases (although factor correlation and intercepts estimates exhibit some bias) and obtains similar results to MH-RM in less time. Our simulations also suggest that the proposed approach performs similarly to and is potentially faster than constrained joint maximum likelihood estimation, a fast procedure that is consistent when the sample size and the number of items simultaneously tend to infinity.

A data driven deep neural network model for predicting boiling heat transfer in helical coils under high gravity

In this article, a deep artificial neural network (ANN) model has been proposed to predict the boiling heat transfer in helical coils under high gravity conditions, which is compared with experimental data. A test rig is set up to provide high gravity up to 11 g with a heat flux up to 15100 W/m2 and the mass velocity range from 40 to 2000 kg m−2 s−1. In the current work, a total 531 data samples have been used in the ANN model. The proposed model was developed in a Python Keras environment with Feed-forward Back-propagation (FFBP) Multi-layer Perceptron (MLP) using eight features (mass flow rate, thermal power, inlet temperature, inlet pressure, direction, acceleration, tube inner surface area, helical coil diameter) as the inputs and two features (wall temperature, heat transfer coefficient) as the outputs. The deep ANN model composed of three hidden layers with a total number of 1098 neurons and 300,266 trainable parameters has been found as optimal according to statistical error analysis. Performance evaluation is conducted based on six verification statistic metrics (R2, MSE, MAE, MAPE, RMSE and cosine proximity) between the experimental data and predicted values. The results demonstrate that a 8-512-512-64-2 neural network has the best performance in predicting the helical coil characteristics with (R2=0.853, MSE=0.018, MAE=0.074, MAPE=1.110, RMSE=0.136, cosine proximity=1.000) in the testing stage. It is indicated that with the utilisation of deep learning, the proposed model is able to successfully predict the heat transfer performance in helical coils, and especially achieved excellent performance in predicting outputs that have a very large range of value differences.

Unknown Security Attack Detection Using Shallow and Deep ANN Classifiers

Advancements in machine learning and artificial intelligence have been widely utilised in the security domain, including but not limited to intrusion detection techniques. With the large training datasets of modern traffic, intelligent algorithms and powerful machine learning tools, security researchers have been able to greatly improve on the intrusion detection models and enhance their ability to detect malicious traffic more accurately. Nonetheless, the problem of detecting completely unknown security attacks is still an open area of research. The enormous number of newly developed attacks constitutes an eccentric challenge for all types of intrusion detection systems. Additionally, the lack of a standard definition of what constitutes an unknown security attack in the literature and the industry alike adds to the problem. In this paper, the researchers reviewed the studies on detecting unknown attacks over the past 10 years and found that they tended to use inconsistent definitions. This formulates the need for a standard consistent definition to have comparable results. The researchers proposed a new categorisation of two types of unknown attacks, namely Type-A, which represents a completely new category of unknown attacks, and Type-B, which represents unknown attacks within already known categories of attacks. The researchers conducted several experiments and evaluated modern intrusion detection systems based on shallow and deep artificial neural network models and their ability to detect Type-A and Type-B attacks using two well-known benchmark datasets for network intrusion detection. The research problem was studied as both a binary and multi-class classification problem. The results showed that the evaluated models had poor overall generalisation error measures, where the classification error rate in detecting several types of unknown attacks from 92 experiments was 50.09%, which highlights the need for new approaches and techniques to address this problem.

CLASSIFICATION PERFORMANCE COMPARISONS OF DEEP LEARNING MODELS IN PNEUMONIA DIAGNOSIS USING CHEST X-RAY IMAGES

In recent years, the analysis of medical images using deep learning techniques has become an area of increasing popularity. Advances in this area have been particularly evident after the discovery of deep artificial neural network models and achieving more successful performance results than other traditional models. In this study, the performance comparison of different deep learning models used to efficiently diagnose pneumonia on chest x-ray images was performed. The data set used in the study consists of a total of 5840 chest x-ray images of individuals. In order to classify these data, three different deep learning models are used: Convolutional Neural Network, Convolutional Neural Network with Data Augmentation and Transfer Learning. The images in the data set were classified into two categories as pneumonia and healthy people using these three deep learning models. The performances of these three deep learning models used in classification were compared in terms of loss and accuracy. In the comparison of three different deep learning models with two different performance values, 5216 chest x-ray images in the data set were used to train the deep learning model and the remaining 624 were used to test the model. At the end of the study, the most successful performance result was obtained by convolutional neural network model applied with data augmentation technique. According to the best results of this study, this model was able to accurately predict the class of 93.4% of the test data.

Solution and validation of inverse kinematics using Deep Artificial neural network

Researchers and roboticists face one of the most complex problems of solving inverse kinematics while designing the robotic manipulator. For the sake of this research, a Deep Artificial neural network (DANN) model is used to solve the inverse kinematics of a 5-axis serial link robotic manipulator. The joints in this robotic manipulator are rotary. MATLAB is used to train the proposed DANN model. Transformation matrix and forward kinematics of the robotic manipulator is used for generation of the training dataset. An experimental robotic model is constructed to authenticate inverse calculations made by this model. A Basler ACA smart camera system is connected with MATLAB. This arrangement calculates deviation at the end needle in real-time. Image processing is used for this purpose. Results of training of DANN showed that there was a deviation of ±0.1 rad in each joint. This result feed-forward satisfactory as it was also validated experimentally. Even with fewer input data, one could achieve the desired target coordinate points with a robotic manipulator using the proposed method.

Experimental Validation for Inverse Kinematics of a Five Axis Robotic Manipulator using Deep Artificial Neural Network

The use of robotics is to improvise and simplify human life. Robotic manipulators have been around for a while now and are being used in many different sectors such as industries, households, warehouses, medicine etc. Solving of inverse kinematics is one of the most complex issues faced while designing the robotic manipulator. In this research a Deep Artificial neural network (D-ANN) model is proposed to solve inverse kinematics of a 5-axis robotic manipulator with rotary joints. The D-ANN model is trained in MATLAB. Training dataset was generated using forward kinematics equations obtained easily from transformation matrix of the robotic manipulator. To validate predictions made by this model an experimental robotic arm manipulator Is fabricated. A smart camera setup has been linked to MATLAB for real time image processing and calculating the deviation of the end needle in reaching the desired target coordinate. The trained model yielded satisfactory results with ±0.03 radians error and this was also validated experimentally. This research will help the robotic manipulator reach the desired target coordinates even when one does not have enough input data.Paper Setup must be in A4 size with Margin: Top 0.7”, Bottom 0.7”, Left 0.65”, 0.65”, Gutter 0”, and Gutter Position Top. Paper must be in two Columns after Authors Name with Width 8.27”, height 11.69” Spacing 0.2”. Whole paper must be with: Font Name Times New Roman, Font Size 10, Line Spacing 1.05 EXCEPT Abstract, Keywords (Index Term), Paper Tile, References, Author Profile (in the last page of the paper, maximum 400 words), All Headings, and Manuscript Details (First Page, Bottom, left side).