Interpretable Neural Network Research Articles

Interpretable neural network classification model using first-order logic rules

Over the past decade, the field of neural networks has made significant strides, particularly in deep learning. However, their limited interpretability has constrained their application in certain critical domains, drawing widespread criticism. Researchers have proposed various methods for explaining neural networks to address this challenge. This paper focuses on rule-based explanations for neural network classification problems. We propose IRCnet, a scalable classification model based on first-order logic rules. IRCnet consists of layers for learning conjunction and disjunction rules, utilizing binary logic activation functions to enhance interpretability. The model is initially trained using a continuous-weight version, which is later binarized to produce a discrete-weight version. During training, we innovatively employed gradient approximation method to handle the non-differentiable weight binarization function, thereby enabling the training of split matrices used for binarization. Finally, Conjunctive Normal Form (CNF) or Disjunctive Normal Form (DNF) rules are extracted from the model’s discrete-weight version. Experimental results indicate that our model achieves the highest or near-highest performance across various classification metrics in multiple structured datasets while demonstrating significant scalability. It effectively balances classification accuracy with the complexity of the generated rules.

PepNet: an interpretable neural network for anti-inflammatory and antimicrobial peptides prediction using a pre-trained protein language model

Identifying anti-inflammatory peptides (AIPs) and antimicrobial peptides (AMPs) is crucial for the discovery of innovative and effective peptide-based therapies targeting inflammation and microbial infections. However, accurate identification of AIPs and AMPs remains a computational challenge mainly due to limited utilization of peptide sequence information. Here, we propose PepNet, an interpretable neural network for predicting both AIPs and AMPs by applying a pre-trained protein language model to fully utilize the peptide sequence information. It first captures the information of residue arrangements and physicochemical properties using a residual dilated convolution block, and then seizes the function-related diverse information by introducing a residual Transformer block to characterize the residue representations generated by a pre-trained protein language model. After training and testing, PepNet demonstrates great superiority over other leading AIP and AMP predictors and shows strong interpretability of its learned peptide representations. A user-friendly web server for PepNet is freely available at http://liulab.top/PepNet/server.

Classifying Neuronal Cell Types Based on Shared Electrophysiological Information from Humans and Mice.

The brain is an intricate system that controls a variety of functions. It consists of a vast number of cells that exhibit diverse characteristics. To understand brain function in health and disease, it is crucial to classify neurons accurately. Recent advancements in machine learning have provided a way to classify neurons based on their electrophysiological activity. This paper presents a deep-learning framework that classifies neurons solely on this basis. The framework uses data from the Allen Cell Types database, which contains a survey of biological features derived from single-cell recordings from mice and humans. The shared information from both sources is used to classify neurons into their broad types with the help of a joint model. An accurate domain-adaptive model, integrating electrophysiological data from both mice and humans, is implemented. Furthermore, data from mouse neurons, which also includes labels of transgenic mouse lines, is further classified into subtypes using an interpretable neural network model. The framework provides state-of-the-art results in terms of accuracy and precision while also providing explanations for the predictions.

Improving the accuracy of genomic prediction in dairy cattle using the biologically annotated neural networks framework

BackgroundBiologically annotated neural networks (BANNs) are feedforward Bayesian neural network models that utilize partially connected architectures based on SNP-set annotations. As an interpretable neural network, BANNs model SNP and SNP-set effects in their input and hidden layers, respectively. Furthermore, the weights and connections of the network are regarded as random variables with prior distributions reflecting the manifestation of genetic effects at various genomic scales. However, its application in genomic prediction has yet to be explored.ResultsThis study extended the BANNs framework to the area of genomic selection and explored the optimal SNP-set partitioning strategies by using dairy cattle datasets. The SNP-sets were partitioned based on two strategies–gene annotations and 100 kb windows, denoted as BANN_gene and BANN_100kb, respectively. The BANNs model was compared with GBLUP, random forest (RF), BayesB and BayesCπ through five replicates of five-fold cross-validation using genotypic and phenotypic data on milk production traits, type traits, and one health trait of 6,558, 6,210 and 5,962 Chinese Holsteins, respectively. Results showed that the BANNs framework achieves higher genomic prediction accuracy compared to GBLUP, RF and Bayesian methods. Specifically, the BANN_100kb demonstrated superior accuracy and the BANN_gene exhibited generally suboptimal accuracy compared to GBLUP, RF, BayesB and BayesCπ across all traits. The average accuracy improvements of BANN_100kb over GBLUP, RF, BayesB and BayesCπ were 4.86%, 3.95%, 3.84% and 1.92%, and the accuracy of BANN_gene was improved by 3.75%, 2.86%, 2.73% and 0.85% compared to GBLUP, RF, BayesB and BayesCπ, respectively across all seven traits. Meanwhile, both BANN_100kb and BANN_gene yielded lower overall mean square error values than GBLUP, RF and Bayesian methods.ConclusionOur findings demonstrated that the BANNs framework performed better than traditional genomic prediction methods in our tested scenarios, and might serve as a promising alternative approach for genomic prediction in dairy cattle.

Enhancing information transmission in FSO-OAM systems through multiscale interpretable neural networks under turbulent conditions

The paper proposes a solution to improve the information transmission efficiency of FSO-OAM systems under turbulent conditions by combining a multiscale interpretable neural network model, 4RK-MSNN. We use a multiscale structure to design the overall architecture of the neural network, which enables the comprehensive analysis of information in different dimensions. Based on the fourth-order Runge-Kutta correlation theory, a core network module, 4RK, is constructed, which can be explained in terms of dynamical systems. The 4RK-MSNN model, which couples the multiscale structure and the 4RK module, has a lower number of parameters, allowing for layered feature extraction in an interpretable framework. This facilitates low-cost, rapid sharing and transmission of feature information at different scales. The proposed solution is validated by transmitting image data under different turbulence intensities and transmission distances. The results indicate the feasibility of the proposed information transfer system. After adding redundant training data, the 4RK-MSNN model significantly improves the quality of the transmitted data and maintains satisfactory results even under strong turbulence and long distances.

A propagation path-based interpretable neural network model for fault detection and diagnosis in chemical process systems

Process monitoring through automated fault detection and diagnosis (FDD) plays a crucial role in maintaining a productive and reliable chemical process system. Developments in AI and machine learning have boosted FDD model performances especially with deep learning methods. However, these neural network models are considered black-boxes where the reasoning behind a diagnosis is unclear, preventing industrial adoption. Therefore, in this study, an interpretable neural network model is proposed for FDD in chemical processes. This framework detects and diagnoses faults based on the propagation paths of different faults which are embedded into the architecture through graph convolutional networks. A mechanism for interpreting the node activations which represent process variables is developed for decision verification. The proposed method is evaluated on the benchmark Tennessee Eastman Process where it achieves a 93.56% accuracy on selected faults.

Exploring the Relative Importance of the MJO and ENSO to North Pacific Subseasonal Predictability

AbstractHere we explore the relative contribution of the Madden‐Julian Oscillation (MJO) and El Niño Southern Oscillation (ENSO) to midlatitude subseasonal predictive skill of upper atmospheric circulation over the North Pacific, using an inherently interpretable neural network applied to pre‐industrial control runs of the Community Earth System Model version 2. We find that this interpretable network generally favors the state of ENSO, rather than the MJO, to make correct predictions on a range of subseasonal lead times and predictand averaging windows. Moreover, the predictability of positive circulation anomalies over the North Pacific is comparatively lower than that of their negative counterparts, especially evident when the ENSO state is important. However, when ENSO is in a neutral state, our findings indicate that the MJO provides some predictive information, particularly for positive anomalies. We identify three distinct evolutions of these MJO states, offering fresh insights into opportune forecasting windows for MJO teleconnections.

SPIN: sex-specific and pathway-based interpretable neural network for sexual dimorphism analysis.

Sexual dimorphism in prevalence, severity and genetic susceptibility exists for most common diseases. However, most genetic and clinical outcome studies are designed in sex-combined framework considering sex as a covariate. Few sex-specific studies have analyzed males and females separately, which failed to identify gene-by-sex interaction. Here, we propose a novel unified biologically interpretable deep learning-based framework (named SPIN) for sexual dimorphism analysis. We demonstrate that SPIN significantly improved the C-index up to 23.6% in TCGA cancer datasets, and it was further validated using asthma datasets. In addition, SPIN identifies sex-specific and -shared risk loci that are often missed in previous sex-combined/-separate analysis. We also show that SPIN is interpretable for explaining how biological pathways contribute to sexual dimorphism and improve risk prediction in an individual level, which can result in the development of precision medicine tailored to a specific individual's characteristics.

NCART: Neural Classification and Regression Tree for tabular data

Deep learning models have become popular in the analysis of tabular data, as they address the limitations of decision trees and enable valuable applications like semi-supervised learning, online learning, and transfer learning. However, these deep-learning approaches often encounter a trade-off. On one hand, they can be computationally demanding when dealing with large-scale or high-dimensional datasets. On the other hand, they may lack interpretability and may not be suitable for small-scale datasets. In this study, we propose a novel interpretable neural network called Neural Classification and Regression Tree (NCART) to overcome these challenges. NCART is a modified version of Residual Networks that replaces fully-connected layers with multiple differentiable oblivious decision trees. By integrating decision trees into the architecture, NCART maintains its interpretability while benefiting from the end-to-end capabilities of neural networks. The simplicity of the NCART architecture makes it well-suited for datasets of varying sizes and reduces computational costs compared to state-of-the-art deep learning models. Extensive numerical experiments demonstrate the superior performance of NCART compared to existing deep learning models, establishing it as a strong competitor to tree-based models. The code is available at https://github.com/Luojiaqimath/NCART.

Learning Traveling Solitary Waves Using Separable Gaussian Neural Networks.

In this paper, we apply a machine-learning approach to learn traveling solitary waves across various physical systems that are described by families of partial differential equations (PDEs). Our approach integrates a novel interpretable neural network (NN) architecture, called Separable Gaussian Neural Networks (SGNN) into the framework of Physics-Informed Neural Networks (PINNs). Unlike the traditional PINNs that treat spatial and temporal data as independent inputs, the present method leverages wave characteristics to transform data into the so-called co-traveling wave frame. This reformulation effectively addresses the issue of propagation failure in PINNs when applied to large computational domains. Here, the SGNN architecture demonstrates robust approximation capabilities for single-peakon, multi-peakon, and stationary solutions (known as "leftons") within the (1+1)-dimensional, b-family of PDEs. In addition, we expand our investigations, and explore not only peakon solutions in the ab-family but also compacton solutions in (2+1)-dimensional, Rosenau-Hyman family of PDEs. A comparative analysis with multi-layer perceptron (MLP) reveals that SGNN achieves comparable accuracy with fewer than a tenth of the neurons, underscoring its efficiency and potential for broader application in solving complex nonlinear PDEs.

Evolutionary Perspectives on Neural Network Generations: A Critical Examination of Models and Design Strategies

Abstract: In the last few years, Neural Networks have become more common in different areas due to their ability to learn intricate patterns and provide precise predictions. Nonetheless, creating an efficient neural network model is a difficult task that demands careful thought of multiple factors, such as architecture, optimization method, and regularization technique. This paper aims to comprehensively overview the state-of-the-art artificial neural network (ANN) generation and highlight key challenges and opportunities in machine learning applications. It provides a critical analysis of current neural network model design methodologies, focusing on the strengths and weaknesses of different approaches. Also, it explores the use of different deep neural networks (DNN) in image recognition, natural language processing, and time series analysis. In addition, the text explores the advantages of selecting optimal values for various components of an Artificial Neural Network (ANN). These components include the number of input/output layers, the number of hidden layers, the type of activation function used, the number of epochs, and the model type selection. Setting these components to their ideal values can help enhance the model's overall performance and generalization. Furthermore, it identifies some common pitfalls and limitations of existing design methodologies, such as overfitting, lack of interpretability, and computational complexity. Finally, it proposes some directions for future research, such as developing more efficient and interpretable neural network architectures, improving the scalability of training algorithms, and exploring the potential of new paradigms, such as Spiking Neural Networks, quantum neural networks, and neuromorphic computing.

Interpretable prediction of mRNA abundance from promoter sequence using contextual regression models.

While machine learning models have been successfully applied to predicting gene expression from promoter sequences, it remains a great challenge to derive intuitive interpretation of the model and reveal DNA motif grammar such as motif cooperation and distance constraint between motif sites. Previous interpretation approaches are often time-consuming or have difficulty to learn the combinatory rules. In this work, we designed interpretable neural network models to predict the mRNA expression levels from DNA sequences. By applying the Contextual Regression framework we developed, we extracted weighted features to cluster samples into different groups, which have different gene expression levels. We performed motif analysis in each cluster and found motifs with active or repressive regulation on gene expression. By comparing the co-occurrence locations of discovered motifs, we also uncovered multiple grammars of motif combination including communities of cooperative motifs and distance constraints between motif pairs. These results revealed new insights of the regulatory architecture of promoter sequences.

CTV-Net: Complex-Valued TV-Driven Network With Nested Topology for 3-D SAR Imaging.

The regularization-based approaches offer promise in improving synthetic aperture radar (SAR) imaging quality while reducing system complexity. However, the widely applied l1 regularization model is hindered by their hypothesis of inherent sparsity, causing unreal estimations of surface-like targets. Inspired by the edge-preserving property of total variation (TV), we propose a new complex-valued TV (CTV)-driven interpretable neural network with nested topology, i.e., CTV-Net, for 3-D SAR imaging. In our scheme, based on the 2-D holography imaging operator, the CTV-driven optimization model is constructed to pursue precise estimations in weakly sparse scenarios. Subsequently, a nested algorithmic framework, i.e., complex-valued TV-driven fast iterative shrinkage thresholding (CTV-FIST), is derived from the theory of proximal gradient descent (PGD) and FIST algorithm, theoretically supporting the design of CTV-Net. In CTV-Net, the trainable weights are layer-varied and functionally relevant to the hyperparameters of CTV-FIST, which aims to constrain the algorithmic parameters to update in a well-conditioned tendency. All weights are learned by end-to-end training based on a two-term cost function, which bounds the measurement fidelity and TV norm simultaneously. Under the guidance of the SAR signal model, a reasonably sized training set is generated, by randomly selecting reference images from the MNIST set and consequently synthesizing complex-valued label signals. Finally, the methodology is validated, numerically and visually, by extensive SAR simulations and real-measured experiments, and the results demonstrate the viability and efficiency of the proposed CTV-Net in the cases of recovering 3-D SAR images from incomplete echoes.

GINN-LP: A Growing Interpretable Neural Network for Discovering Multivariate Laurent Polynomial Equations

Traditional machine learning is generally treated as a black-box optimization problem and does not typically produce interpretable functions that connect inputs and outputs. However, the ability to discover such interpretable functions is desirable. In this work, we propose GINN-LP, an interpretable neural network to discover the form and coefficients of the underlying equation of a dataset, when the equation is assumed to take the form of a multivariate Laurent Polynomial. This is facilitated by a new type of interpretable neural network block, named the “power-term approximator block”, consisting of logarithmic and exponential activation functions. GINN-LP is end-to-end differentiable, making it possible to use backpropagation for training. We propose a neural network growth strategy that will enable finding the suitable number of terms in the Laurent polynomial that represents the data, along with sparsity regularization to promote the discovery of concise equations. To the best of our knowledge, this is the first model that can discover arbitrary multivariate Laurent polynomial terms without any prior information on the order. Our approach is first evaluated on a subset of data used in SRBench, a benchmark for symbolic regression. We first show that GINN-LP outperforms the state-of-the-art symbolic regression methods on datasets generated using 48 real-world equations in the form of multivariate Laurent polynomials. Next, we propose an ensemble method that combines our method with a high-performing symbolic regression method, enabling us to discover non-Laurent polynomial equations. We achieve state-of-the-art results in equation discovery, showing an absolute improvement of 7.1% over the best contender, by applying this ensemble method to 113 datasets within SRBench with known ground-truth equations.

Abstract 912: CheckpointPx, an interpretable radiology AI tool, predicts checkpoint blockade benefit independent of PDL1 status in non-small cell lung cancers (NSCLC): A multi-institutional validation study

Abstract Background: Immune Checkpoint Inhibitors (ICIs) are a cornerstone in the treatment of cancers such as Non-Small Cell Lung Cancer (NSCLC). Despite the standard use of biomarkers such as PD-L1 expression for selecting ICI regimens, recent studies suggest missed opportunities for benefit in PD-L1 low/negative patients. Therefore, there is an urgent need for new biomarker strategies capable of guiding the use of ICIs in NSCLC. We introduce and validate an AI-powered radiology tool for ICI patient selection that provides interpretable predictions by quantifying aspects of tumor biology on baseline CT scans directly associated with ICI efficacy, such as tumor heterogeneity and twistedness of the tumor-associated vasculature. Methods: We analyzed CT scans of 439 patients treated with ICI at 4 institutions (D1-D4) using the Picture Health Px platform. Experienced radiologists and physicians across institutions delineated target lesions on baseline CT scans. A deep learning model was used to segment pulmonary vessels. CheckpointPx v1.11, an interpretable neural network incorporating quantitative features extracted from within the tumor and its vasculature, was trained to predict ICI response (defined as disease control per RECIST best overall response) on n=247 patients (D1-D3). It was assessed with respect to response and progression free-survival (PFS) on 192 patients (D2-D4), with D4 (n=105) being an external validation set. CheckpointPx was also evaluated within PDL1 subsets among testing set patients where available (n=138). Results: The cohort was predominantly late stage (Stage 3/4) (>85%) and was mixed with respect to ICI line of treatment (1st-4th). 122 of 247 (49%) and 119 of 192 (62%) were responders in training and test cohorts respectively. CheckpointPx included 16 features, such as entropy-based heterogeneity and quantitative vessel tortuosity (twistedness). The model predicted response to ICI with an AUC=0.65 on the test set. The CheckpointPx High Risk group significantly stratified patients by PFS (HR=1.67 [1.22-2.28], p=0.001). This separation remained significant within the subset of PDL1-negative (HR=2.71 [1.35-5.44], p=0.005) and PDL1-positive patients (HR=2.05 [1.22-3.46], p=0.007). Conclusions: From baseline radiology, CheckpointPx was shown to strongly predict ICI outcomes across multiple institutions, NSCLC stages, and lines of ICI. Furthermore, the tool stratified patients by ICI benefit within both PDL1-positive and negative subsets - suggesting the potential of CheckpointPx to address critical gaps in the NSCLC biomarker landscape. CheckpointPx is driven by interpretable imaging biomarkers, and thus its predictions are tied to specific and quantifiable phenotypic attributes of the tumor and its microenvironment. Citation Format: Amogh Hiremath, Seyoung Lee, Jeeyeon Lee, Peter Haseok Kim, Kai Zhang, Salie Lee, Monica Yadav, Maria Jose Aguilera Chuchuca, Taegyu Um, Myungwoo Nam, Liam Il-Young Chung, Hye Sung Kim, Jisang Yu, Trie Arni Djunadi, Leeseul Kim, Youjin Oh, Sungmi Yoon, Zunairah Shah, Yuchan Kim, Ilene Hong, Grace Kang, Jessica Jang, Amy Cho, Soowon Lee, Cecilia Nam, Timothy Hong, Yury S. Velichko, Amit Gupta, Vamsidhar Velcheti, Anant Madabhushi, Nathaniel Braman, Young Kwang Chae. CheckpointPx, an interpretable radiology AI tool, predicts checkpoint blockade benefit independent of PDL1 status in non-small cell lung cancers (NSCLC): A multi-institutional validation study [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 912.

Joint intra and inter-channel nonlinear compensation scheme based on improved learned digital back propagation for WDM systems.

In this paper, we improve the learned digital back propagation (LDBP) and propose a novel joint intra and inter-channel nonlinearity compensation scheme for polarization division multiplexing wavelength-division multiplexed (PDM-WDM) systems. From the perspective of interpretable neural network, the scheme realizes the alternating compensation of chromatic dispersion (CD) and nonlinearity based on physical models. The chromatic dispersion compensation (CDC) adopts one-dimensional convolution operation in the time domain. Moreover, the pulse-broadening effect is introduced into the overlap-and-save method. For nonlinear compensation, the improved joint model is applied, and the impact of the intra-channel pulse broadening and the walk-off effect between different channels caused by CD on the nonlinear effect is considered. To validate the effectiveness of the proposed scheme, we construct an 11-channel simulation system of 36 GBaud PDM uniform 16 quadrature amplitude modulation (PDM-16QAM) 1600 km and 64 GBaud PDM-64QAM 400 km, as well as a 5-channel experimental system of 28 GBaud PDM-16QAM 806.4 km. The simulation results show that the performance of PDM-16QAM with 0.5 steps per span and PDM-64QAM with 2 steps per span improve the Q-factor by approximately 0.75 dB and 0.54 dB at the optimal launch power, compared with the linear compensation scheme. The transmission performance of PDM-16QAM is higher than that of digital back propagation with 5 steps per span (DBP-5StPS), and the complexity is only 31.36% of that of DBP-5StPS. The performance of PDM-64QAM is higher than that of DBP-10StPS, with a complexity of 62.72%. The experimental results show that the performance of PDM-16QAM with 0.5 steps per span is improved with 0.86 dB Q-factor improvement compared with the linear compensation scheme at the optimal launch power, and the performance of the proposed scheme is higher than that of DBP-5StPS with a complexity of only 23.68%.

Incorporating metabolic activity, taxonomy and community structure to improve microbiome-based predictive models for host phenotype prediction

ABSTRACT We developed MicroKPNN, a prior-knowledge guided interpretable neural network for microbiome-based human host phenotype prediction. The prior knowledge used in MicroKPNN includes the metabolic activities of different bacterial species, phylogenetic relationships, and bacterial community structure, all in a shallow neural network. Application of MicroKPNN to seven gut microbiome datasets (involving five different human diseases including inflammatory bowel disease, type 2 diabetes, liver cirrhosis, colorectal cancer, and obesity) shows that incorporation of the prior knowledge helped improve the microbiome-based host phenotype prediction. MicroKPNN outperformed fully connected neural network-based approaches in all seven cases, with the most improvement of accuracy in the prediction of type 2 diabetes. MicroKPNN outperformed a recently developed deep-learning based approach DeepMicro, which selects the best combination of autoencoder and machine learning approach to make predictions, in all of the seven cases. Importantly, we showed that MicroKPNN provides a way for interpretation of the predictive models. Using importance scores estimated for the hidden nodes, MicroKPNN could provide explanations for prior research findings by highlighting the roles of specific microbiome components in phenotype predictions. In addition, it may suggest potential future research directions for studying the impacts of microbiome on host health and diseases. MicroKPNN is publicly available at https://github.com/mgtools/MicroKPNN.

A hybrid neural network approach for classifying diabetic retinopathy subtypes.

Diabetic retinopathy is a prevalent complication among diabetic patients that, if not predicted and treated promptly, can lead to blindness. This paper proposes a method for accurately and swiftly predicting the degree of diabetic retinopathy using a hybrid neural network model. Timely prediction of diabetic retinopathy is crucial in preventing blindness associated with this condition. This study aims to enhance the prediction accuracy of diabetic retinopathy by utilizing the hybrid neural network model EfficientNet and Swin Transformer. The specific methodology includes: (1) combining local and global features to accurately capture lesion characteristics by leveraging the strengths of both Swin Transformer and EfficientNet models; (2) improving prediction accuracy through a comprehensive analysis of the model's training details and applying data augmentation techniques such as Gaussian blur to enhance the hybrid model's performance; (3) validating the effectiveness and utility of the proposed hybrid model for diabetic retinopathy detection through extensive experimental evaluations and comparisons with other deep learning models. The hybrid model was trained and tested on the large-scale real-world diabetic retinopathy detection dataset APTOS 2019 Blindness Detection. The experimental results show that the hybrid model in this paper achieves the best results in all metrics, including sensitivity of 0.95, specificity of 0.98, accuracy of 0.97, and AUC of 0.97. The performance of the model is significantly improved compared to the mainstream methods currently employed. In addition, the model provides interpretable neural network details through class activation maps, which enables the visualization of diabetic retinopathy. This feature helps physicians to make more accurate diagnosis and treatment decisions. The model proposed in this paper shows higher accuracy in detecting and diagnosing diabetic retinopathy, which is crucial for the treatment and rehabilitation of diabetic patients. The hybrid neural network model based on EfficientNet and Swin Transformer significantly contributes to the prediction of diabetic retinopathy. By combining local and global features, the model achieves improved prediction accuracy. The validity and utility of the model are verified through experimental evaluations. This research provides robust support for the early diagnosis and treatment of diabetic patients.

Synchronization-Inspired Interpretable Neural Networks.

Synchronization is a ubiquitous phenomenon in nature that enables the orderly presentation of information. In the human brain, for instance, functional modules such as the visual, motor, and language cortices form through neuronal synchronization. Inspired by biological brains and previous neuroscience studies, we propose an interpretable neural network incorporating a synchronization mechanism. The basic idea is to constrain each neuron, such as a convolution filter, to capture a single semantic pattern while synchronizing similar neurons to facilitate the formation of interpretable functional modules. Specifically, we regularize the activation map of a neuron to surround its focus position of the activated pattern in a sample. Moreover, neurons locally interact with each other, and similar ones are synchronized together during the training phase adaptively. Such local aggregation preserves the globally distributed representation nature of the neural network model, enabling a reasonably interpretable representation. To analyze the neuron interpretability comprehensively, we introduce a series of novel evaluation metrics from multiple aspects. Qualitative and quantitative experiments demonstrate that the proposed method outperforms many state-of-the-art algorithms in terms of interpretability. The resulting synchronized functional modules show module consistency across data and semantic specificity within modules.

A Hybrid Model for Spatiotemporal Air Quality Prediction Based on Interpretable Neural Networks and a Graph Neural Network

To effectively address air pollution and enhance air quality, governments must be able to predict the air quality index with high accuracy and reliability. However, air quality prediction is subject to ambiguity and instability because of the atmosphere’s fluidity, making it challenging to identify the temporal and spatial correlations using a single model. Therefore, a new hybrid model is proposed based on an interpretable neural network and a graph neural network (INNGNN), which simulates the temporal and spatial dependence of air quality and achieves accurate multi-step air quality prediction. A time series is first interpreted using interpretable neural networks (INN) to extract the potentially important aspects that are easily overlooked in the data; second, a self-attention mechanism catches the local and global dependencies and associations in the time series. Lastly, a city map is created using a graph neural network (GNN) to determine the relationships between cities in order to extract the spatially dependent features. In the experimental evaluation, the results show that the INNGNN model performs better than comparable algorithms. Therefore, it is confirmed that the INNGNN model can effectively capture the temporal and spatial relationships and better predict air quality.