This article proposes a brain-inspired generative (BIG) model that merges an impulsive-attention neural network and a variational autoencoder (VAE) for identifying cognitive states based on electroencephalography (EEG) data. A hybrid learning method is presented for training the model by integrating gradient-based learning and heteroassociative memory. The BIG model is capable of achieving multi-task objectives: EEG classification, generating new EEG, and brain network interpretation, alleviating the limitations of excessive data training and high computational cost in conventional approaches. Experimental results on two public EEG datasets with different sampling rates demonstrate that the BIG model achieves a classification accuracy above 89\%, comparable with state-of-the-art methods, while reducing computational cost by nearly 11\% over the baseline EEGNet. Incorporating the generated EEG data for training, the BIG model exhibits comparative performance in a few-shot pattern. Ablation studies justify the poised brain-inspired characteristic regarding the impulsive-attention module and the hybrid learning method. Thanks to the performance advantages with interpretable outputs, this BIG model has application potential for building digital twins of the brain.
The inherent "black box" nature of deep neural networks (DNNs) compromises their transparency and reliability. Recently, explainable AI (XAI) has garnered increasing attention from researchers. Several perturbation-based interpretations have emerged. However, these methods often fail to adequately consider feature dependencies. To solve this problem, we introduce a perturbation-based interpretation guided by feature coalitions, which leverages deep information of network to extract correlated features. Then, we proposed a carefully-designed consistency loss to guide network interpretation. Both quantitative and qualitative experiments are conducted to validate the effectiveness of our proposed method. Code is available at github.com/Teriri1999/Perturebation-on-Feature-Coalition.
Several natural phenomena and complex systems are often represented as networks. Discovering their community structure is a fundamental task for understanding these networks. Many algorithms have been proposed, but recently, Graph Neural Networks (GNN) have emerged as a compelling approach for enhancing this task.In this paper, we introduce a simple, efficient, and clustering-oriented model based on unsupervised \textbf{G}raph Attention \textbf{A}uto\textbf{E}ncoder for community detection in attributed networks (GAECO). The proposed model adeptly learns representations from both the network's topology and attribute information, simultaneously addressing dual objectives: reconstruction and community discovery. It places a particular emphasis on discovering compact communities by robustly minimizing clustering errors. The model employs k-means as an objective function and utilizes a multi-head Graph Attention Auto-Encoder for decoding the representations. Experiments conducted on three datasets of attributed networks show that our method surpasses state-of-the-art algorithms in terms of NMI and ARI. Additionally, our approach scales effectively with the size of the network, making it suitable for large-scale applications. The implications of our findings extend beyond biological network interpretation and social network analysis, where knowledge of the fundamental community structure is essential.
Magnetic resonance imaging (MRI) is an important non-invasive imaging method in clinical diagnosis. Beyond the common image structures, parametric imaging can provide the intrinsic tissue property thus could be used in quantitative evaluation. The emerging deep learning approach provides fast and accurate parameter estimation but still encounters the lack of network interpretation and enough training data. Even with a large amount of training data, the mismatch between the training and target data may introduce errors. Here, we propose one way that solely relies on the target scanned data and does not need a pre-defined training database. We provide a proof-of-concept that embeds the physical rule of MRI, the Bloch equation, into the loss of physics-informed neural network (PINN). PINN enables learning the Bloch equation, estimating the T2 parameter, and generating a series of physically synthetic data. Experimental results are conducted on phantom and cardiac imaging to demonstrate its potential in quantitative MRI.
Neural network interpretation methods, particularly feature attribution methods, are known to be fragile with respect to adversarial input perturbations. To address this, several methods for enhancing the local smoothness of the gradient while training have been proposed for attaining \textit{robust} feature attributions. However, the lack of considering the normalization of the attributions, which is essential in their visualizations, has been an obstacle to understanding and improving the robustness of feature attribution methods. In this paper, we provide new insights by taking such normalization into account. First, we show that for every non-negative homogeneous neural network, a naive $\ell_2$-robust criterion for gradients is \textit{not} normalization invariant, which means that two functions with the same normalized gradient can have different values. Second, we formulate a normalization invariant cosine distance-based criterion and derive its upper bound, which gives insight for why simply minimizing the Hessian norm at the input, as has been done in previous work, is not sufficient for attaining robust feature attribution. Finally, we propose to combine both $\ell_2$ and cosine distance-based criteria as regularization terms to leverage the advantages of both in aligning the local gradient. As a result, we experimentally show that models trained with our method produce much more robust interpretations on CIFAR-10 and ImageNet-100 without significantly hurting the accuracy, compared to the recent baselines. To the best of our knowledge, this is the first work to verify the robustness of interpretation on a larger-scale dataset beyond CIFAR-10, thanks to the computational efficiency of our method.
Visual question answering (VQA) is a hallmark of vision and language reasoning and a challenging task under the zero-shot setting. We propose Plug-and-Play VQA (PNP-VQA), a modular framework for zero-shot VQA. In contrast to most existing works, which require substantial adaptation of pretrained language models (PLMs) for the vision modality, PNP-VQA requires no additional training of the PLMs. Instead, we propose to use natural language and network interpretation as an intermediate representation that glues pretrained models together. We first generate question-guided informative image captions, and pass the captions to a PLM as context for question answering. Surpassing end-to-end trained baselines, PNP-VQA achieves state-of-the-art results on zero-shot VQAv2 and GQA. With 11B parameters, it outperforms the 80B-parameter Flamingo model by 8.5% on VQAv2. With 738M PLM parameters, PNP-VQA achieves an improvement of 9.1% on GQA over FewVLM with 740M PLM parameters. Code is released at https://github.com/salesforce/LAVIS/tree/main/projects/pnp-vqa
Uncovering rationales behind predictions of graph neural networks (GNNs) has received increasing attention over recent years. Instance-level GNN explanation aims to discover critical input elements, like nodes or edges, that the target GNN relies upon for making predictions. Though various algorithms are proposed, most of them formalize this task by searching the minimal subgraph which can preserve original predictions. However, an inductive bias is deep-rooted in this framework: several subgraphs can result in the same or similar outputs as the original graphs. Consequently, they have the danger of providing spurious explanations and fail to provide consistent explanations. Applying them to explain weakly-performed GNNs would further amplify these issues. To address this problem, we theoretically examine the predictions of GNNs from the causality perspective. Two typical reasons of spurious explanations are identified: confounding effect of latent variables like distribution shift, and causal factors distinct from the original input. Observing that both confounding effects and diverse causal rationales are encoded in internal representations, we propose a simple yet effective countermeasure by aligning embeddings. Concretely, concerning potential shifts in the high-dimensional space, we design a distribution-aware alignment algorithm based on anchors. This new objective is easy to compute and can be incorporated into existing techniques with no or little effort. Theoretical analysis shows that it is in effect optimizing a more faithful explanation objective in design, which further justifies the proposed approach.
Dropout is designed to relieve the overfitting problem in high-level vision tasks but is rarely applied in low-level vision tasks, like image super-resolution (SR). As a classic regression problem, SR exhibits a different behaviour as high-level tasks and is sensitive to the dropout operation. However, in this paper, we show that appropriate usage of dropout benefits SR networks and improves the generalization ability. Specifically, dropout is better embedded at the end of the network and is significantly helpful for the multi-degradation settings. This discovery breaks our common sense and inspires us to explore its working mechanism. We further use two analysis tools -- one is from recent network interpretation works, and the other is specially designed for this task. The analysis results provide side proofs to our experimental findings and show us a new perspective to understand SR networks.
Here we propose and investigate the use of visibility graphs to model the feature map of a neural network. The model, initially devised for studies on complex networks, is employed here for the classification of texture images. The work is motivated by an alternative viewpoint provided by these graphs over the original data. The performance of the proposed method is verified in the classification of four benchmark databases, namely, KTHTIPS-2b, FMD, UIUC, and UMD and in a practical problem, which is the identification of plant species using scanned images of their leaves. Our method was competitive with other state-of-the-art approaches, confirming the potential of techniques used for data analysis in different contexts to give more meaningful interpretation to the use of neural networks in texture classification.
We design, implement, and evaluate DeepEverest, a system for the efficient execution of interpretation by example queries over the activation values of a deep neural network. DeepEverest consists of an efficient indexing technique and a query execution algorithm with various optimizations. We prove that the proposed query execution algorithm is instance optimal. Experiments with our prototype show that DeepEverest, using less than 20% of the storage of full materialization, significantly accelerates individual queries by up to 63x and consistently outperforms other methods on multi-query workloads that simulate DNN interpretation processes.
Gaussian processes (GPs) are ubiquitously used in sciences and engineering as metamodels. Standard GPs, however, can only handle numerical or quantitative variables. In this paper, we introduce latent map Gaussian processes (LMGPs) that inherit the attractive properties of GPs and are also applicable to mixed data which have both quantitative and qualitative inputs. The core idea behind LMGPs is to learn a continuous, low-dimensional latent space or manifold which encodes all qualitative inputs. To learn this manifold, we first assign a unique prior vector representation to each combination of qualitative inputs. We then use a low-rank linear map to project these priors on a manifold that characterizes the posterior representations. As the posteriors are quantitative, they can be directly used in any standard correlation function such as the Gaussian or Matern. Hence, the optimal map and the corresponding manifold, along with other hyperparameters of the correlation function, can be systematically learned via maximum likelihood estimation. Through a wide range of analytic and real-world examples, we demonstrate the advantages of LMGPs over state-of-the-art methods in terms of accuracy and versatility. In particular, we show that LMGPs can handle variable-length inputs, have an explainable neural network interpretation, and provide insights into how qualitative inputs affect the response or interact with each other. We also employ LMGPs in Bayesian optimization and illustrate that they can discover optimal compound compositions more efficiently than conventional methods that convert compositions to qualitative variables via manual featurization.
The lack of interpretability and trust is a much-criticised feature of deep neural networks. In fully connected nets, the signalling between inner layers is scrambled because backpropagation training does not require perceptrons to be arranged in any particular order. The result is a black box; this problem is particularly severe in scientific computing and digital signal processing (DSP), where neutral nets perform abstract mathematical transformations that do not reduce to features or concepts. We present here a group-theoretical procedure that attempts to bring inner layer signalling into a human-readable form, the assumption being that this form exists and has identifiable and quantifiable features - for example, smoothness or locality. We applied the proposed method to DEERNet (a DSP network used in electron spin resonance) and managed to descramble it. We found considerable internal sophistication: the network spontaneously invents a bandpass filter, a notch filter, a frequency axis rectifier, frequency division multiplexing, group embedding, spectral filtering regularisation, and a Fourier-like transform that maps harmonic functions into Chebyshev polynomials - in ten minutes of unattended training from a random initial guess.
Neural networks embedded in safety-sensitive applications such as self-driving cars and wearable health monitors rely on two important techniques: input attribution for hindsight analysis and network compression to reduce its size for edge-computing. In this paper, we show that these seemingly unrelated techniques conflict with each other as network compression deforms the produced attributions, which could lead to dire consequences for mission-critical applications. This phenomenon arises due to the fact that conventional network compression methods only preserve the predictions of the network while ignoring the quality of the attributions. To combat the attribution inconsistency problem, we present a framework that can preserve the attributions while compressing a network. By employing the Weighted Collapsed Attribution Matching regularizer, we match the attribution maps of the network being compressed to its pre-compression former self. We demonstrate the effectiveness of our algorithm both quantitatively and qualitatively on diverse compression methods.
Recent works have empirically shown that there exist adversarial examples that can be hidden from neural network interpretability (namely, making network interpretation maps visually similar), or interpretability is itself susceptible to adversarial attacks. In this paper, we theoretically show that with a proper measurement of interpretation, it is actually difficult to prevent prediction-evasion adversarial attacks from causing interpretation discrepancy, as confirmed by experiments on MNIST, CIFAR-10 and Restricted ImageNet. Spurred by that, we develop an interpretability-aware defensive scheme built only on promoting robust interpretation (without the need for resorting to adversarial loss minimization). We show that our defense achieves both robust classification and robust interpretation, outperforming state-of-the-art adversarial training methods against attacks of large perturbation in particular.
Neural networks have opened up many new opportunities to utilize remotely sensed images in meteorology. Common applications include image classification, e.g., to determine whether an image contains a tropical cyclone, and image translation, e.g., to emulate radar imagery for satellites that only have passive channels. However, there are yet many open questions regarding the use of neural networks in meteorology, such as best practices for evaluation, tuning and interpretation. This article highlights several strategies and practical considerations for neural network development that have not yet received much attention in the meteorological community, such as the concept of effective receptive fields, underutilized meteorological performance measures, and methods for NN interpretation, such as synthetic experiments and layer-wise relevance propagation. We also consider the process of neural network interpretation as a whole, recognizing it as an iterative scientist-driven discovery process, and breaking it down into individual steps that researchers can take. Finally, while most work on neural network interpretation in meteorology has so far focused on networks for image classification tasks, we expand the focus to also include networks for image translation.
Deep learning based medical image classifiers have shown remarkable prowess in various application areas like ophthalmology, dermatology, pathology, and radiology. However, the acceptance of these Computer-Aided Diagnosis (CAD) systems in real clinical setups is severely limited primarily because their decision-making process remains largely obscure. This work aims at elucidating a deep learning based medical image classifier by verifying that the model learns and utilizes similar disease-related concepts as described and employed by dermatologists. We used a well-trained and high performing neural network developed by REasoning for COmplex Data (RECOD) Lab for classification of three skin tumours, i.e. Melanocytic Naevi, Melanoma and Seborrheic Keratosis and performed a detailed analysis on its latent space. Two well established and publicly available skin disease datasets, PH2 and derm7pt, are used for experimentation. Human understandable concepts are mapped to RECOD image classification model with the help of Concept Activation Vectors (CAVs), introducing a novel training and significance testing paradigm for CAVs. Our results on an independent evaluation set clearly shows that the classifier learns and encodes human understandable concepts in its latent representation. Additionally, TCAV scores (Testing with CAVs) suggest that the neural network indeed makes use of disease-related concepts in the correct way when making predictions. We anticipate that this work can not only increase confidence of medical practitioners on CAD but also serve as a stepping stone for further development of CAV-based neural network interpretation methods.
Deep neural networks (DNNs) have had many successes, but they suffer from two major issues: (1) a vulnerability to adversarial examples and (2) a tendency to elude human interpretation. Interestingly, recent empirical and theoretical evidence suggests these two seemingly disparate issues are actually connected. In particular, robust models tend to provide more interpretable gradients than non-robust models. However, whether this relationship works in the opposite direction remains obscure. With this paper, we seek empirical answers to the following question: can models acquire adversarial robustness when they are trained to have interpretable gradients? We introduce a theoretically inspired technique called Interpretation Regularization (IR), which encourages a model's gradients to (1) match the direction of interpretable target salience maps and (2) have small magnitude. To assess model performance and tease apart factors that contribute to adversarial robustness, we conduct extensive experiments on MNIST and CIFAR-10 with both $\ell_2$ and $\ell_\infty$ attacks. We demonstrate that training the networks to have interpretable gradients improves their robustness to adversarial perturbations. Applying the network interpretation technique SmoothGrad yields additional performance gains, especially in cross-norm attacks and under heavy perturbations. The results indicate that the interpretability of the model gradients is a crucial factor for adversarial robustness. Code for the experiments can be found at https://github.com/a1noack/interp_regularization.
Neural networks have become increasingly prevalent within the geosciences, although a common limitation of their usage has been a lack of methods to interpret what the networks learn and how they make decisions. As such, neural networks have often been used within the geosciences to most accurately identify a desired output given a set of inputs, with the interpretation of what the network learns used as a secondary metric to ensure the network is making the right decision for the right reason. Neural network interpretation techniques have become more advanced in recent years, however, and we therefore propose that the ultimate objective of using a neural network can also be the interpretation of what the network has learned rather than the output itself. We show that the interpretation of neural networks can enable the discovery of scientifically meaningful connections within geoscientific data. In particular, we use two methods for neural network interpretation called backwards optimization and layerwise relevance propagation, both of which project the decision pathways of a network back onto the original input dimensions. To the best of our knowledge, LRP has not yet been applied to geoscientific research, and we believe it has great potential in this area. We show how these interpretation techniques can be used to reliably infer scientifically meaningful information from neural networks by applying them to common climate patterns. These results suggest that combining interpretable neural networks with novel scientific hypotheses will open the door to many new avenues in neural network-related geoscience research.
The remarkable performance of convolutional neural networks (CNNs) is entangled with their huge number of uninterpretable parameters, which has become the bottleneck limiting the exploitation of their full potential. Towards network interpretation, previous endeavors mainly resort to the single filter analysis, which however ignores the relationship between filters. In this paper, we propose a novel architecture decoupling method to interpret the network from a perspective of investigating its calculation paths. More specifically, we introduce a novel architecture controlling module in each layer to encode the network architecture by a vector. By maximizing the mutual information between the vectors and input images, the module is trained to select specific filters to distill a unique calculation path for each input. Furthermore, to improve the interpretability and compactness of the decoupled network, the output of each layer is encoded to align the architecture encoding vector with the constraint of sparsity regularization. Unlike conventional pixel-level or filter-level network interpretation methods, we propose a path-level analysis to explore the relationship between the combination of filter and semantic concepts, which is more suitable to interpret the working rationale of the decoupled network. Extensive experiments show that the decoupled network achieves several applications, i.e., network interpretation, network acceleration, and adversarial samples detection.
In this paper, we propose a trainable multiplication layer (TML) for a neural network that can be used to calculate the multiplication between the input features. Taking an image as an input, the TML raises each pixel value to the power of a weight and then multiplies them, thereby extracting the higher-order local auto-correlation from the input image. The TML can also be used to extract co-occurrence from the feature map of a convolutional network. The training of the TML is formulated based on backpropagation with constraints to the weights, enabling us to learn discriminative multiplication patterns in an end-to-end manner. In the experiments, the characteristics of the TML are investigated by visualizing learned kernels and the corresponding output features. The applicability of the TML for classification and neural network interpretation is also evaluated using public datasets.