Artificial Neural Network Learning Research Articles

Is Learning in Biological Neural Networks Based on Stochastic Gradient Descent? An Analysis Using Stochastic Processes.

In recent years, there has been an intense debate about how learning in biological neural networks (BNNs) differs from learning in artificial neural networks. It is often argued that the updating of connections in the brain relies only on local information, and therefore a stochastic gradient-descent type optimization method cannot be used. In this article, we study a stochastic model for supervised learning in BNNs. We show that a (continuous) gradient step occurs approximately when each learning opportunity is processed by many local updates. This result suggests that stochastic gradient descent may indeed play a role in optimizing BNNs.

Identification of Elastoplastic Constitutive Model of GaN Thin Films Using Instrumented Nanoindentation and Machine Learning Technique

This study presents a novel inverse identification approach to determine the elastoplastic parameters of a 2 µm thick GaN semiconductor thin film deposited on a sapphire substrate. This approach combines instrumented nanoindentation with finite element (FE) simulations and an artificial neural network (ANN) model. Experimental load–depth curves were obtained using a Berkovich indenter. To generate a comprehensive database for the inverse analysis, FE models were constructed to simulate load–depth responses across a wide range of GaN thin film properties. The accuracy of both 2D and 3D simulations was compared to select the optimal model for database generation. The Box–Behnken design-based data sampling method was used to define the number of simulations and input variables for the FE models. The ANN technique was then employed to establish the complex mapping between the simulated load–depth curves (input) and the corresponding stress–strain curve (output). The generated database was used to train and test the ANN model. Then, the learned ANN model was used to achieve high accuracy in identifying the stress–strain curve of the GaN thin film from the experimental load–depth data. This work demonstrates the successful application of an inverse analysis framework, combining experimental nanoindentation tests, FE modeling, and an ANN model, for the characterization of the elastoplastic behavior of GaN thin films.

Brain-inspired learning in artificial neural networks: A review

Artificial neural networks (ANNs) have emerged as an essential tool in machine learning, achieving remarkable success across diverse domains, including image and speech generation, game playing, and robotics. However, there exist fundamental differences between ANNs’ operating mechanisms and those of the biological brain, particularly concerning learning processes. This paper presents a comprehensive review of current brain-inspired learning representations in artificial neural networks. We investigate the integration of more biologically plausible mechanisms, such as synaptic plasticity, to improve these networks’ capabilities. Moreover, we delve into the potential advantages and challenges accompanying this approach. In this review, we pinpoint promising avenues for future research in this rapidly advancing field, which could bring us closer to understanding the essence of intelligence.

Successive bootstrapping deep learning approach and airborne EM-borehole data fusion to understand salt water in the Mississippi River Valley Alluvial Aquifer

Increasing demands from agriculture and urbanization have decreased groundwater level and increased salinity worldwide. Better aquifer characterization and soil salinity mapping are important for proactive groundwater management. Airborne electromagnetic (AEM) is a powerful tool for aquifer characterization and salinity delineation. However, AEM needs to be interpreted with caution before being used for groundwater quality analysis. This study introduces a framework that utilizes the AEM data for both lithologic modeling and salinity delineation. A resistivity-to-lithology (R2L) model is developed to interpret AEM resistivity to lithology based a depth-dependent multi-resistivity thresholds. Then, a cokriging method is used to integrate AEM data from two different EM systems to predict resistivity at the aquifer. Finally, a resistivity-to-chloride concentration (R2C) model utilizes the resistivity model to estimate chloride concentrations at sand facies. A deep learning artificial neural network (DL-ANN) model is introduced with a successive bootstrapping approach to estimate total dissolved solids first and then use it together with resistivity data to estimate chloride concentration. The methodology was applied to delineating salinity plumes in the Mississippi River Valley alluvial aquifer (MRVA). This study found that the salinity distribution in MRVA is highly correlated with the Jurassic salt basin, salt domes, faulting, seismicity, and river water quality. The result indicates salinity upconing due to excessive pumping.

An efficient parameterized neural network enhanced multiscale finite element modeling for triply periodic minimal surface meta-structures and its applications for femur

Triply periodic minimal surface (TPMS) multiscale structures are meta-structures known for their low-weight-high-strength property. Because of their delicate geometry and sheet-networks feature, a fine mesh is required for finite element analysis, which is computationally expensive for conventional direct numerical simulation. To resolve this issue, this study develops a neural network enhanced finite element method (NN-FEM) for parameterized TPMS multiscale meta-structures. The multiscale homogenization approach replaces the complicated representative volume elements (RVEs) with a surrogate homogenized material model. The surrogate process is accomplished by deep learning of artificial neural networks (NN) with an effective boundary displacement gradient and the Helmholtz free energy of RVEs. The NN-trained surrogate material model is deployed on the macroscale Galerkin finite element model for various field tests. The parameterized structural information of TPMS can also be embedded into the NN model to resolve the hidden constitutive relation of RVEs with unique geometries. Numerical examples are provided to benchmark the efficacy and efficiency of NN-FEM on the parameterized TPMS structure. The proposed model can be applied to femurs for rapid stress analysis and osteoporosis detection.

A novel metaheuristic optimization and soft computing techniques for improved hydrological drought forecasting

Drought is one of the costliest natural disasters worldwide and weakens countries economically by causing negative impacts on hydropower and agricultural production. Therefore, it is necessary to create drought risk management plans by monitoring and predicting droughts. Various drought indicators have been developed to monitor droughts. This study aims to forecast Streamflow Drought Index (SDI) values with various novel metaheuristic optimization-based Artificial Neural Network (ANN) and deep learning models to predict 1-month lead-time hydrological droughts on 1, 3, and 12-month time scales in the Konya closed basin, one of the driest basins in Turkey. To achieve this goal, the ANN model was integrated with the Firefly Algorithm (FFA), Genetic Algorithm (GA), and Particle Swarm Optimization (PSO) techniques and compared against long short-term memory (LSTM) networks. While establishing the SDI prediction model, lag values exceeding the 95% confidence intervals in the partial autocorrelation function graphs were used. Model performance was evaluated according to scatter matrix, radar, time series, bee swarm graphs, and statistical performance metrics. As a result of the analysis, the PSO-ANN hybrid model with (R2:0.468–0.931) at station 1611 and the FFA-ANN hybrid model with (R2:0.443–0.916) at station 1612 generally have the highest accuracy.

Improved Artificial Neural Network Classification Model based Metaheuristic Optimization for Handwritten Character Recognition

This study addresses the concerns regarding the performance of Handwritten Character Recognition (HCR) systems, focusing on the classification stage. It is widely acknowledged that the development of the classification model significantly impacts the overall performance of HCR. The problems identified specifically pertain to the classification model, particularly in the context of the Artificial Neural Network (ANN) learning problem, leading to low accuracy in recognizing handwritten characters. The objective of this study is to improve and refine the ANN classification model to achieve better HCR. To achieve this goal, this study proposed a hybrid Flower Pollination Algorithm with Artificial Neural Network (FPA-ANN) classification model for HCR. The FPA is one of the metaheuristic approaches is utilized as an optimization technique to enhance the performance of ANN, particularly by optimizing the network training process of ANN. The experimentation phase involves using the National Institute of Standards and Technology (NIST) handwritten character database. Finally, the proposed FPA-ANN classification model is analysed based on generated confusion matrix and evaluated performance of the classification model in terms of precision, sensitivity, specificity, F-score and accuracy.

Crescent Moon Visibility: A New Criterion using Deep learned Artificial Neural-Network

Many authors investigated the problem of the early visibility of the new crescent moon after the conjunction and proposed many criteria addressing this issue in the literature. This article presented a proposed criterion for early crescent moon sighting based on a deep-learned pattern recognizer artificial neural network (ANN) performance. Moon sight datasets were collected from various sources and used to learn the ANN. The new criterion relied on the crescent width and the arc of vision from the edge of the crescent bright limb. The result of that criterion was a control value indicating the moon's visibility condition, which separated the datasets into four regions: invisible, telescope only, probably visible, and certainly visible. This criterion was used on the dataset for ANN learning to compare its efficiency with the actual moon visibility events.

Improving learning in Artificial Neural Networks using better weight initializations

Improving learning in Artificial Neural Networks using better weight initializations

Design of urban road fault detection system based on artificial neural network and deep learning.

In urban traffic management, the timely detection of road faults plays a crucial role in improving traffic efficiency and safety. However, conventional methods often fail to fully leverage the information from road topology and traffic data. To address this issue, we propose an innovative detection system that combines Artificial Neural Networks (ANNs), specifically Graph Convolutional Networks (GCN), Bidirectional Gated Recurrent Units (BiGRU), and self-attention mechanisms. Our approach begins by representing the road topology as a graph and utilizing GCN to model it. This allows us to learn the relationships between roads and capture their structural dependencies. By doing so, we can effectively incorporate the spatial information provided by the road network. Next, we employ BiGRU to model the historical traffic data, enabling us to capture the temporal dynamics and patterns in the traffic flow. The BiGRU architecture allows for bidirectional processing, which aids in understanding the traffic conditions based on both past and future information. This temporal modeling enhances our system's ability to handle time-varying traffic patterns. To further enhance the feature representations, we leverage self-attention mechanisms. By combining the hidden states of the BiGRU with self-attention, we can assign importance weights to different temporal features, focusing on the most relevant information. This attention mechanism helps to extract salient features from the traffic data. Subsequently, we merge the features learned by GCN from the road topology and BiGRU from the traffic data. This fusion of spatial and temporal information provides a comprehensive representation of the road status. By employing a Multilayer Perceptron (MLP) as a classifier, we can effectively determine whether a road is experiencing a fault. The MLP model is trained using labeled road fault data through supervised learning, optimizing its performance for fault detection. Experimental evaluations of our system demonstrate excellent performance in road fault detection. Compared to traditional methods, our system achieves more accurate fault detection, thereby improving the efficiency of urban traffic management. This is of significant importance for city administrators, as they can promptly identify road faults and take appropriate measures for repair and traffic diversion.

Code-Free Machine Learning Solutions for Microscopy Image Processing: Deep Learning.

In recent years, there has been a significant expansion in the realm of processing microscopy images, thanks to the advent of machine learning techniques. These techniques offer diverse applications for image processing. Currently, numerous methods are used for processing microscopy images in the field of biology, ranging from conventional machine learning algorithms to sophisticated deep learning artificial neural networks with millions of parameters. However, a comprehensive grasp of the intricacies of these methods usually necessitates proficiency in programming and advanced mathematics. In our comprehensive review, we explore various widely used deep learning approaches tailored for the processing of microscopy images. Our emphasis is on algorithms that have gained popularity in the field of biology and have been adapted to cater to users lacking programming expertise. In essence, our target audience comprises biologists interested in exploring the potential of deep learning algorithms, even without programming skills. Throughout the review, we elucidate each algorithm's fundamental concepts and capabilities without delving into mathematical and programming complexities. Crucially, all the highlighted algorithms are accessible on open platforms without requiring code, and we provide detailed descriptions and links within our review. It's essential to recognize that addressing each specific problem demands an individualized approach. Consequently, our focus is not on comparing algorithms but on delineating the problems they are adept at solving. In practical scenarios, researchers typically select multiple algorithms suited to their tasks and experimentally determine the most effective one. It is worth noting that microscopy extends beyond the realm of biology; its applications span diverse fields such as geology and material science. Although our review predominantly centers on biomedical applications, the algorithms and principles outlined here are equally applicable to other scientific domains. Furthermore, a number of the proposed solutions can be modified for use in entirely distinct computer vision cases.

Digital twin model with machine learning and optimization for resilient production–distribution systems under disruptions

Inspired by a real-life problem in the semiconductor industry, we introduce a novel digital twin model for a company subject to the adverse effects of unpredictable disruptions. Specifically, this company manufactures a product using a raw material provided by an external supplier, whose lead times may abruptly change due to disruptive events. The Smoothing Order-Up-To rule is adopted by the company as a replenishment policy. It is characterized by three control parameters, which must be optimized to enhance the resilience of the system. To this end, the digital twin learns from the real production–distribution data and periodically self-adjusts the replenishment parameters based on the evolution of the external environment. The digital twin architecture combines data analytics, simulation modeling, machine learning, and a metaheuristic. More specifically, an Artificial Neural Network learns from the manufacturer’s operations and generates predictive models. These are embedded in a Particle Swarm Optimization, which provides the optimal combination of the replenishment parameters. An experimental campaign was performed to demonstrate that the digital twin outperforms the traditional strategy in which the replenishment parameters are kept unchanged. The numerical results show that the digital twin strongly improves the manufacturer’s performance, in particular in terms of time-to-recover and time-to-survive, used to measure the resilience of the system subject to disruption.

An Effective Heart Disease Detection and Classification using Machine Learning

About 20.5 million people die yearly due to cardiovascular disease (CVD). Early prediction of heart disease and accurate heart severity identification can save the induvial life with timely medications. In many cases, most deaths occur due to inaccurate diagnosis of the heart. In computer science, various existing researchers have dealt with the classification of heart disease on synthetic as well as real-time datasets. But still, those systems have a challenge such as low classification accuracy, high error rate and inaccurate heart severity detection. After identifying all these challenges, we proposed an effective heart disease detection and classification using hybrid machine learning techniques. In this article, we describe how various feature extraction and hybrid machine learning classifiers produce accurate severity of heart disease on real-time datasets. In the first phase, we collected a Cleveland dataset from Kaggle and then applied preprocessing and normalization for data balancing. Various feature extraction and selection methods, such as TF-IDF, Co-relation coefficient, N-Gram, and bi-Gram features, are used for practical module training. The different machine learning we used on normalized datasets for classification. The five supervised machine learning algorithms are used, such as Support Vector Machine (SVM), Naïve Bayes (NB), Artificial Neural Network (ANN) and Hybrid Machine Learning (HML) etc. The HML achieves 99.30% accuracy on the Cleveland dataset using Weka 3.8 machine learning framework. As a result, the proposed system compares with various heart disease predictions using machine learning techniques.

Predictive analysis of terrorist activities in Thailand's Southern provinces: a deep learning approach

Terrorist activities have been on the rise globally, with Thailand experiencing significant challenges, particularly in its three southern border provinces. This study offers a comprehensive analysis aiming to predict forthcoming terrorist events in these provinces. We employed historical data, categorized into nine groups based on military expert recommendations, to train our prediction model. This research tested the prediction capabilities of various methodologies, including decision trees, naïve Bayesian learning techniques, and deep learning artificial neural networks. Notably, the deep neural network emerged as the superior predictive tool, achieving an impressive accuracy of 98.21% and a root mean square error (RMSE) of 0.59%. The primary anticipated events include bombings, shootings, assaults, and acts of vandalism. Our findings also revealed that Pattani Province was the most affected, accounting for 45% of incidents. Specific districts, such as Panare and Yarang, exhibited high crime rates of 40% and 36.84%, respectively. Yala Province, particularly Bannang Sata District, was identified as the hotspot for shooting incidents, with a rate of 34%.

Comparative Analysis of Ensemble Learning Techniques and ANN for Prediction of Compressive Strength of Concrete Mix

Abstract: In this study, non-linear machine learning regressors and an artificial neural network (ANN) were modeled to predict the compressive strength of concrete mixes. A dataset of 1030 mix design data points with a wide range of compressive strength varying from M10 to M80 is taken into account. The data set comprised of 8 input parameters namely- cement, water, fine & coarse aggregate, flyash, slag, superplasticizer, and age. The paper also analyzes effects on the prediction of compressive strength due to waiving of the parameters with lesser importance factors. To improve the performances of models by reducing the variance, biases, and errors, K-fold cross-validation was examined. To examine the performances of the models, metrics such as coefficient of determination (R2) and mean square error (MSE) were employed. Most of the models can predict the compressive strength with good accuracy. Out of the 7-machine learning regressors and ANN models employed, Gradient Boosting Regressor yielded the best prediction accuracy of 0.9561 with the lowest MSE value of 0.0513.

Prediction of Hydrodynamic Parameters of the State of the Bottomhole Zone of Wells Using Machine Learning Methods

The relevance of the development of a methodology for the operational assessment of the bottom-hole formation zone (the permeability of the bottom-hole formation zone and the skin factor) is primarily due to economic considerations, since existing approaches to its definition based on hydrodynamic studies lead to shortages and increased risks of failure to ensure the output of the well. In this regard, the use of modern methods of working with big data, such as deep learning of artificial neural networks, will ensure monitoring of the condition of the bottom-hole zone of the well formation without stopping them for hydrodynamic tests, which will reduce losses for oil production enterprises. It will allow for operational analysis for effective and timely application of intensification technologies, enhanced oil recovery. The authors analyzed the existing methods for determining the bottom-hole characteristics of the formation and machine learning approaches in the direction of solving this problem. The article presents a methodology for the operational assessment of the state of the bottom-hole formation zone: the permeability of the near bottomhole zone (NBHZ) and the skin factor using artificial neural network training approaches based on geological, operational data and the results of interpretation of hydrodynamic studies on the example of sandstones of oil fields in the Perm Region. A fully connected neural network was used to predict the NBHZ permeability. The article presents the results of testing various neural network architectures: the number of layers and neurons in layers with the choice of the best one. Some techniques were used to prevent over-training of models. The author’s methodology for assessing the skin factor of wells is proposed using a comprehensive analysis of the constructed statistical models and training models of artificial neural networks to solve the regression problem. In future studies, it is planned to use recurrent and convolutional neural networks to study the dynamic components of the formation of the bottom-hole formation zone and create an integrated approach to solve the problem.

Motion sensitive network for action recognition in control and decision-making of autonomous systems.

Spatial-temporal modeling is crucial for action recognition in videos within the field of artificial intelligence. However, robustly extracting motion information remains a primary challenge due to temporal deformations of appearances and variations in motion frequencies between different actions. In order to address these issues, we propose an innovative and effective method called the Motion Sensitive Network (MSN), incorporating the theories of artificial neural networks and key concepts of autonomous system control and decision-making. Specifically, we employ an approach known as Spatial-Temporal Pyramid Motion Extraction (STP-ME) module, adjusting convolution kernel sizes and time intervals synchronously to gather motion information at different temporal scales, aligning with the learning and prediction characteristics of artificial neural networks. Additionally, we introduce a new module called Variable Scale Motion Excitation (DS-ME), utilizing a differential model to capture motion information in resonance with the flexibility of autonomous system control. Particularly, we employ a multi-scale deformable convolutional network to alter the motion scale of the target object before computing temporal differences across consecutive frames, providing theoretical support for the flexibility of autonomous systems. Temporal modeling is a crucial step in understanding environmental changes and actions within autonomous systems, and MSN, by integrating the advantages of Artificial Neural Networks (ANN) in this task, provides an effective framework for the future utilization of artificial neural networks in autonomous systems. We evaluate our proposed method on three challenging action recognition datasets (Kinetics-400, Something-Something V1, and Something-Something V2). The results indicate an improvement in accuracy ranging from 1.1% to 2.2% on the test set. When compared with state-of-the-art (SOTA) methods, the proposed approach achieves a maximum performance of 89.90%. In ablation experiments, the performance gain of this module also shows an increase ranging from 2% to 5.3%. The introduced Motion Sensitive Network (MSN) demonstrates significant potential in various challenging scenarios, providing an initial exploration into integrating artificial neural networks into the domain of autonomous systems.

Machine learning technology for predicting intraocular lens power. Diagnostics data generalization

Backgraund. Implantation of recent IOLs allows ophthalmologists to effectively solve the problems of surgical rehabilitation of patients with cataracts. The degree of improving the patient's visual functions is directly dependent on the accuracy of the preoperative calculation of the optical power of IOLs. The most famous formulas to calculate this indicator are SRK II, SRK/T, Hoffer-Q, Holladay II, Haigis, Barrett. All of them work well for an "average patient", however, they are not adequate enough at the boundaries of input variables ranges.
 Aims. To study the possibility of using mathematical models obtained as a result of deep learning of artificial neural networks (ANN models) to generalize data and predict the optical power of modern intraocular lenses.
 Materials and methods. ANN models were trained on large-scale samples, including depersonalized data for patients of ophthalmology clinic. Data provided in 2021 by ophthalmologist K.K. Syrykh, and reflect the results of both preoperative and postoperative observation of patients. The source file used to build the ANN model includes 455 records - 26 columns of input factors and one column for the output factor - calculating IOL (diopters). To conveniently build ANN models, we used a simulator program, previously developed by the authors.
 Results. The resulting models, in contrast to the traditionally used formulas, reflect the regional specificity of patients to a much greater extent. They also make it possible to retrain and optimize the structure based on newly received data, which allows taking into account the non-stationarity of the object. A distinctive feature of such ANN models in comparison with the well-known formulas SRK II, SRK/T, Hoffer-Q, Holladay II, Haigis, Barrett, widely used in the surgical cataract treatment in ophthalmology, is their ability to take into account a significant number of recorded input quantities, which makes it possible to reduce the mean relative error in calculating the optical power of IOL from 10–12% to 3.5%.
 Conclusions. This work shows the fundamental possibility of generalizing a significant amount of empirical data on calculating the optical power of the IOL using training ANN models that have a significantly larger number of input variables than when using traditional formulas and methods. The results obtained allow, in principle, to build an intelligent expert system with a continuous flow of new data from a source and a step-by-step retraining of the ANN model.

A comparative study of 11 non-linear regression models highlighting autoencoder, DBN, and SVR, enhanced by SHAP importance analysis in soybean branching prediction.

To explore a robust tool for advancing digital breeding practices through an artificial intelligence-driven phenotype prediction expert system, we undertook a thorough analysis of 11 non-linear regression models. Our investigation specifically emphasized the significance of Support Vector Regression (SVR) and SHapley Additive exPlanations (SHAP) in predicting soybean branching. By using branching data (phenotype) of 1918 soybean accessions and 42k SNP (Single Nucleotide Polymorphism) polymorphic data (genotype), this study systematically compared 11 non-linear regression AI models, including four deep learning models (DBN (deep belief network) regression, ANN (artificial neural network) regression, Autoencoders regression, and MLP (multilayer perceptron) regression) and seven machine learning models (e.g., SVR (support vector regression), XGBoost (eXtreme Gradient Boosting) regression, Random Forest regression, LightGBM regression, GPs (Gaussian processes) regression, Decision Tree regression, and Polynomial regression). After being evaluated by four valuation metrics: R2 (R-squared), MAE (Mean Absolute Error), MSE (Mean Squared Error), and MAPE (Mean Absolute Percentage Error), it was found that the SVR, Polynomial Regression, DBN, and Autoencoder outperformed other models and could obtain a better prediction accuracy when they were used for phenotype prediction. In the assessment of deep learning approaches, we exemplified the SVR model, conducting analyses on feature importance and gene ontology (GO) enrichment to provide comprehensive support. After comprehensively comparing four feature importance algorithms, no notable distinction was observed in the feature importance ranking scores across the four algorithms, namely Variable Ranking, Permutation, SHAP, and Correlation Matrix, but the SHAP value could provide rich information on genes with negative contributions, and SHAP importance was chosen for feature selection. The results of this study offer valuable insights into AI-mediated plant breeding, addressing challenges faced by traditional breeding programs. The method developed has broad applicability in phenotype prediction, minor QTL (quantitative trait loci) mining, and plant smart-breeding systems, contributing significantly to the advancement of AI-based breeding practices and transitioning from experience-based to data-based breeding.

Biomarkers and computational models for predicting efficacy to tumor ICI immunotherapy.

Numerous studies have shown that immune checkpoint inhibitor (ICI) immunotherapy has great potential as a cancer treatment, leading to significant clinical improvements in numerous cases. However, it benefits a minority of patients, underscoring the importance of discovering reliable biomarkers that can be used to screen for potential beneficiaries and ultimately reduce the risk of overtreatment. Our comprehensive review focuses on the latest advancements in predictive biomarkers for ICI therapy, particularly emphasizing those that enhance the efficacy of programmed cell death protein 1 (PD-1)/programmed cell death-ligand 1 (PD-L1) inhibitors and cytotoxic T-lymphocyte antigen-4 (CTLA-4) inhibitors immunotherapies. We explore biomarkers derived from various sources, including tumor cells, the tumor immune microenvironment (TIME), body fluids, gut microbes, and metabolites. Among them, tumor cells-derived biomarkers include tumor mutational burden (TMB) biomarker, tumor neoantigen burden (TNB) biomarker, microsatellite instability (MSI) biomarker, PD-L1 expression biomarker, mutated gene biomarkers in pathways, and epigenetic biomarkers. TIME-derived biomarkers include immune landscape of TIME biomarkers, inhibitory checkpoints biomarkers, and immune repertoire biomarkers. We also discuss various techniques used to detect and assess these biomarkers, detailing their respective datasets, strengths, weaknesses, and evaluative metrics. Furthermore, we present a comprehensive review of computer models for predicting the response to ICI therapy. The computer models include knowledge-based mechanistic models and data-based machine learning (ML)models. Among the knowledge-based mechanistic models are pharmacokinetic/pharmacodynamic (PK/PD) models, partial differential equation (PDE) models, signal networks-based models, quantitative systems pharmacology (QSP) models, and agent-based models (ABMs). ML models include linear regression models, logistic regression models, support vector machine (SVM)/random forest/extra trees/k-nearest neighbors (KNN) models, artificial neural network (ANN) and deep learning models. Additionally, there are hybrid models of systems biology and ML. We summarized the details of these models, outlining the datasets they utilize, their evaluation methods/metrics, and their respective strengths and limitations. By summarizing the major advances in the research on predictive biomarkers and computer models for the therapeutic effect and clinical utility of tumor ICI, we aim to assist researchers in choosing appropriate biomarkers or computer models for research exploration and help clinicians conduct precision medicine by selecting the best biomarkers.