Zijun Zhang

Michail Chatzianastasis, Michalis Vazirgiannis, Zijun Zhang

The identification of cancer genes is a critical yet challenging problem in cancer genomics research. Existing computational methods, including deep graph neural networks, fail to exploit the multilayered gene-gene interactions or provide limited explanation for their predictions. These methods are restricted to a single biological network, which cannot capture the full complexity of tumorigenesis. Models trained on different biological networks often yield different and even opposite cancer gene predictions, hindering their trustworthy adaptation. Here, we introduce an Explainable Multilayer Graph Neural Network (EMGNN) approach to identify cancer genes by leveraging multiple genegene interaction networks and pan-cancer multi-omics data. Unlike conventional graph learning on a single biological network, EMGNN uses a multilayered graph neural network to learn from multiple biological networks for accurate cancer gene prediction. Our method consistently outperforms all existing methods, with an average 7.15% improvement in area under the precision-recall curve (AUPR) over the current state-of-the-art method. Importantly, EMGNN integrated multiple graphs to prioritize newly predicted cancer genes with conflicting predictions from single biological networks. For each prediction, EMGNN provided valuable biological insights via both model-level feature importance explanations and molecular-level gene set enrichment analysis. Overall, EMGNN offers a powerful new paradigm of graph learning through modeling the multilayered topological gene relationships and provides a valuable tool for cancer genomics research.

Christian Pedersen, Tiberiu Tesileanu, Tinghui Wu et al. · cambridge

In Elmarakeby et al., "Biologically informed deep neural network for prostate cancer discovery", a feedforward neural network with biologically informed, sparse connections (P-NET) was presented to model the state of prostate cancer. We verified the reproducibility of the study conducted by Elmarakeby et al., using both their original codebase, and our own re-implementation using more up-to-date libraries. We quantified the contribution of network sparsification by Reactome biological pathways, and confirmed its importance to P-NET's superior performance. Furthermore, we explored alternative neural architectures and approaches to incorporating biological information into the networks. We experimented with three types of graph neural networks on the same training data, and investigated the clinical prediction agreement between different models. Our analyses demonstrated that deep neural networks with distinct architectures make incorrect predictions for individual patient that are persistent across different initializations of a specific neural architecture. This suggests that different neural architectures are sensitive to different aspects of the data, an important yet under-explored challenge for clinical prediction tasks.

Qiao Li, Cong Wu, Jing Chen et al.

Deep neural networks (DNNs) are increasingly used in critical applications such as identity authentication and autonomous driving, where robustness against adversarial attacks is crucial. These attacks can exploit minor perturbations to cause significant prediction errors, making it essential to enhance the resilience of DNNs. Traditional defense methods often rely on access to detailed model information, which raises privacy concerns, as model owners may be reluctant to share such data. In contrast, existing black-box defense methods fail to offer a universal defense against various types of adversarial attacks. To address these challenges, we introduce DUCD, a universal black-box defense method that does not require access to the target model's parameters or architecture. Our approach involves distilling the target model by querying it with data, creating a white-box surrogate while preserving data privacy. We further enhance this surrogate model using a certified defense based on randomized smoothing and optimized noise selection, enabling robust defense against a broad range of adversarial attacks. Comparative evaluations between the certified defenses of the surrogate and target models demonstrate the effectiveness of our approach. Experiments on multiple image classification datasets show that DUCD not only outperforms existing black-box defenses but also matches the accuracy of white-box defenses, all while enhancing data privacy and reducing the success rate of membership inference attacks.

Luoxiao Yang, Yun Wang, Xinqi Fan et al.

Time series forecasting (TSF) possesses great practical values in various fields, including power and energy, transportation, etc. TSF methods have been studied based on knowledge from classical statistics to modern deep learning. Yet, all of them were developed based on one fundamental concept, the numerical data fitting. Thus, the models developed have long been known to be problem-specific and lacking application generalizability. Practitioners expect a TSF foundation model that serves TSF tasks in different applications. The central question is then how to develop such a TSF foundation model. This paper offers one pioneering study in the TSF foundation model development method and proposes a vision intelligence-powered framework, ViTime, for the first time. ViTime fundamentally shifts TSF from numerical fitting to operations based on a binary image-based time series metric space and naturally supports both point and probabilistic forecasting. We also provide rigorous theoretical analyses of ViTime, including quantization-induced system error bounds and principled strategies for optimal parameter selection. Furthermore, we propose RealTS, an innovative synthesis algorithm generating diverse and realistic training samples, effectively enriching the training data and significantly enhancing model generalizability. Extensive experiments demonstrate ViTime's state-of-the-art performance. In zero-shot scenarios, ViTime outperforms TimesFM by 9-15\%. With just 10\% fine-tuning data, ViTime surpasses both leading foundation models and fully-supervised benchmarks, a gap that widens with 100\% fine-tuning. ViTime also exhibits exceptional robustness, effectively handling missing data and outperforming TimesFM by 20-30\% under various data perturbations, validating the power of its visual space data operation paradigm.

Luoxiao Yang, Xinqi Fan, Zijun Zhang

Time series forecasting (TSF) has been a challenging research area, and various models have been developed to address this task. However, almost all these models are trained with numerical time series data, which is not as effectively processed by the neural system as visual information. To address this challenge, this paper proposes a novel machine vision assisted deep time series analysis (MV-DTSA) framework. The MV-DTSA framework operates by analyzing time series data in a novel binary machine vision time series metric space, which includes a mapping and an inverse mapping function from the numerical time series space to the binary machine vision space, and a deep machine vision model designed to address the TSF task in the binary space. A comprehensive computational analysis demonstrates that the proposed MV-DTSA framework outperforms state-of-the-art deep TSF models, without requiring sophisticated data decomposition or model customization. The code for our framework is accessible at https://github.com/IkeYang/ machine-vision-assisted-deep-time-series-analysis-MV-DTSA-.

Luoxiao Yang, Zhong Zheng, Zijun Zhang

The convolutional neural network (CNN) has been widely applied to process the industrial data based tensor input, which integrates data records of distributed industrial systems from the spatial, temporal, and system dynamics aspects. However, unlike images, information in the industrial data based tensor is not necessarily spatially ordered. Thus, directly applying CNN is ineffective. To tackle such issue, we propose a plug and play module, the Rubik's Cube Operator (RCO), to adaptively permutate the data organization of the industrial data based tensor to an optimal or suboptimal order of attributes before being processed by CNNs, which can be updated with subsequent CNNs together via the gradient-based optimizer. The proposed RCO maintains K binary and right stochastic permutation matrices to permutate attributes of K axes of the input industrial data based tensor. A novel learning process is proposed to enable learning permutation matrices from data, where the Gumbel-Softmax is employed to reparameterize elements of permutation matrices, and the soft regularization loss is proposed and added to the task-specific loss to ensure the feature diversity of the permuted data. We verify the effectiveness of the proposed RCO via considering two representative learning tasks processing industrial data via CNNs, the wind power prediction (WPP) and the wind speed prediction (WSP) from the renewable energy domain. Computational experiments are conducted based on four datasets collected from different wind farms and the results demonstrate that the proposed RCO can improve the performance of CNN based networks significantly.

Luoxiao Yang, Long Wang, Zijun Zhang

This paper develops a novel self-training U-net (STU-net) based method for the automated WPC model generation without requiring data pre-processing. The self-training (ST) process of STU-net has two steps. First, different from traditional studies regarding the WPC modeling as a curve fitting problem, in this paper, we renovate the WPC modeling formulation from a machine vision aspect. To develop sufficiently diversified training samples, we synthesize supervisory control and data acquisition (SCADA) data based on a set of S-shape functions depicting WPCs. These synthesized SCADA data and WPC functions are visualized as images and paired as training samples(I_x, I_wpc). A U-net is then developed to approximate the model recovering I_wpc from I_x. The developed U-net is applied into observed SCADA data and can successfully generate the I_wpc. Moreover, we develop a pixel mapping and correction process to derive a mathematical form f_wpc representing I_wpcgenerated previously. The proposed STU-net only needs to train once and does not require any data preprocessing in applications. Numerical experiments based on 76 WTs are conducted to validate the superiority of the proposed method by benchmarking against classical WPC modeling methods. To demonstrate the repeatability of the presented research, we release our code at https://github.com/IkeYang/STU-net.

Zijun Zhang, Linqi Zhou, Liangke Gou et al.

We report a neural architecture search framework, BioNAS, that is tailored for biomedical researchers to easily build, evaluate, and uncover novel knowledge from interpretable deep learning models. The introduction of knowledge dissimilarity functions in BioNAS enables the joint optimization of predictive power and biological knowledge through searching architectures in a model space. By optimizing the consistency with existing knowledge, we demonstrate that BioNAS optimal models reveal novel knowledge in both simulated data and in real data of functional genomics. BioNAS provides a useful tool for domain experts to inject their prior belief into automated machine learning and therefore making deep learning easily accessible to practitioners. BioNAS is available at https://github.com/zj-zhang/BioNAS-pub.

Junwei Su, Difan Zou, Zijun Zhang et al.

Incremental learning is a machine learning approach that involves training a model on a sequence of tasks, rather than all tasks at once. This ability to learn incrementally from a stream of tasks is crucial for many real-world applications. However, incremental learning is a challenging problem on graph-structured data, as many graph-related problems involve prediction tasks for each individual node, known as Node-wise Graph Incremental Learning (NGIL). This introduces non-independent and non-identically distributed characteristics in the sample data generation process, making it difficult to maintain the performance of the model as new tasks are added. In this paper, we focus on the inductive NGIL problem, which accounts for the evolution of graph structure (structural shift) induced by emerging tasks. We provide a formal formulation and analysis of the problem, and propose a novel regularization-based technique called Structural-Shift-Risk-Mitigation (SSRM) to mitigate the impact of the structural shift on catastrophic forgetting of the inductive NGIL problem. We show that the structural shift can lead to a shift in the input distribution for the existing tasks, and further lead to an increased risk of catastrophic forgetting. Through comprehensive empirical studies with several benchmark datasets, we demonstrate that our proposed method, Structural-Shift-Risk-Mitigation (SSRM), is flexible and easy to adapt to improve the performance of state-of-the-art GNN incremental learning frameworks in the inductive setting.

Huixin Zhan, Zijun Zhang

Clinical variant classification of pathogenic versus benign genetic variants remains a pivotal challenge in clinical genetics. Recently, the proposition of protein language models has improved the generic variant effect prediction (VEP) accuracy via weakly-supervised or unsupervised training. However, these VEPs are not disease-specific, limiting their adaptation at point-of-care. To address this problem, we propose a disease-specific \textsc{pro}tein language model for variant \textsc{path}ogenicity, termed ProPath, to capture the pseudo-log-likelihood ratio in rare missense variants through a siamese network. We evaluate the performance of ProPath against pre-trained language models, using clinical variant sets in inherited cardiomyopathies and arrhythmias that were not seen during training. Our results demonstrate that ProPath surpasses the pre-trained ESM1b with an over $5\%$ improvement in AUC across both datasets. Furthermore, our model achieved the highest performances across all baselines for both datasets. Thus, our ProPath offers a potent disease-specific variant effect prediction, particularly valuable for disease associations and clinical applicability.

Huixin Zhan, Ying Nian Wu, Zijun Zhang

Although DNA foundation models have advanced the understanding of genomes, they still face significant challenges in the limited scale and diversity of genomic data. This limitation starkly contrasts with the success of natural language foundation models, which thrive on substantially larger scales. Furthermore, genome understanding involves numerous downstream genome annotation tasks with inherent data heterogeneity, thereby necessitating more efficient and robust fine-tuning methods tailored for genomics. Here, we present \textsc{Lingo}: \textsc{L}anguage prefix f\textsc{In}e-tuning for \textsc{G}en\textsc{O}mes. Unlike DNA foundation models, \textsc{Lingo} strategically leverages natural language foundation models' contextual cues, recalibrating their linguistic knowledge to genomic sequences. \textsc{Lingo} further accommodates numerous, heterogeneous downstream fine-tune tasks by an adaptive rank sampling method that prunes and stochastically reintroduces pruned singular vectors within small computational budgets. Adaptive rank sampling outperformed existing fine-tuning methods on all benchmarked 14 genome understanding tasks, while requiring fewer than 2\% of trainable parameters as genomic-specific adapters. Impressively, applying these adapters on natural language foundation models matched or even exceeded the performance of DNA foundation models. \textsc{Lingo} presents a new paradigm of efficient and scalable genome understanding via genomic-specific adapters on language models.

Eric Xue, Ruiyi Zhang, Zijun Zhang et al.

Transformer models are foundational to natural language processing (NLP) applications, yet remain vulnerable to backdoor attacks introduced through poisoned data, which implant hidden behaviors during training. To strengthen the ability to prevent such compromises, recent research has focused on designing increasingly stealthy attacks to stress-test existing defenses, pairing backdoor behaviors with stylized artifact or token-level perturbation triggers. However, this trend diverts attention from the harder and more realistic case: making the model respond to semantic triggers such as specific names or entities, where a successful backdoor could manipulate outputs tied to real people or events in deployed systems. Motivated by this growing disconnect, we introduce SteganoBackdoor, bringing stealth techniques back into line with practical threat models. Leveraging innocuous properties from natural-language steganography, SteganoBackdoor applies a gradient-guided data optimization process to transform semantic trigger seeds into steganographic carriers that embed a high backdoor payload, remain fluent, and exhibit no representational resemblance to the trigger. Across diverse experimental settings, SteganoBackdoor achieves over 99% attack success at an order-of-magnitude lower data-poisoning rate than prior approaches while maintaining unparalleled evasion against a comprehensive suite of data-level defenses. By revealing this practical and covert attack, SteganoBackdoor highlights an urgent blind spot in current defenses and demands immediate attention to adversarial data defenses and real-world threat modeling.

Selena Song, Ziming Xu, Zijun Zhang et al.

Diffusion Transformer(DiT) based video generation models have recently achieved impressive visual quality and temporal coherence, but they still frequently violate basic physical laws and commonsense dynamics, revealing a lack of explicit world knowledge. In this work, we explore how to equip them with a plug-and-play memory that injects useful world knowledge. Motivated by in-context memory in Transformer-based LLMs, we conduct empirical studies to show that DiT can be steered via interventions on its hidden states, and simple low-pass and high-pass filters in the embedding space naturally disentangle low-level appearance and high-level physical/semantic cues, enabling targeted guidance. Building on these observations, we propose a learnable memory encoder DiT-Mem, composed of stacked 3D CNNs, low-/high-pass filters, and self-attention layers. The encoder maps reference videos into a compact set of memory tokens, which are concatenated as the memory within the DiT self-attention layers. During training, we keep the diffusion backbone frozen, and only optimize the memory encoder. It yields a rather efficient training process on few training parameters (150M) and 10K data samples, and enables plug-and-play usage at inference time. Extensive experiments on state-of-the-art models demonstrate the effectiveness of our method in improving physical rule following and video fidelity. Our code and data are publicly released here: https://thrcle421.github.io/DiT-Mem-Web/.

Zijun Zhang, Adam R. Lamson, Michael Shelley et al.

Finely-tuned enzymatic pathways control cellular processes, and their dysregulation can lead to disease. Creating predictive and interpretable models for these pathways is challenging because of the complexity of the pathways and of the cellular and genomic contexts. Here we introduce Elektrum, a deep learning framework which addresses these challenges with data-driven and biophysically interpretable models for determining the kinetics of biochemical systems. First, it uses in vitro kinetic assays to rapidly hypothesize an ensemble of high-quality Kinetically Interpretable Neural Networks (KINNs) that predict reaction rates. It then employs a novel transfer learning step, where the KINNs are inserted as intermediary layers into deeper convolutional neural networks, fine-tuning the predictions for reaction-dependent in vivo outcomes. Elektrum makes effective use of the limited, but clean in vitro data and the complex, yet plentiful in vivo data that captures cellular context. We apply Elektrum to predict CRISPR-Cas9 off-target editing probabilities and demonstrate that Elektrum achieves state-of-the-art performance, regularizes neural network architectures, and maintains physical interpretability.

Tiange Wang, Zijun Zhang, Kwok-Leung Tsui

Accurate forecasting of traffic conditions is critical for improving safety, stability, and efficiency of a city transportation system. In reality, it is challenging to produce accurate traffic forecasts due to the complex and dynamic spatiotemporal correlations. Most existing works only consider partial characteristics and features of traffic data, and result in unsatisfactory performances on modeling and forecasting. In this paper, we propose a periodic spatial-temporal deep neural network (PSTN) with three pivotal modules to improve the forecasting performance of traffic conditions through a novel integration of three types of information. First, the historical traffic information is folded and fed into a module consisting of a graph convolutional network and a temporal convolutional network. Second, the recent traffic information together with the historical output passes through the second module consisting of a graph convolutional network and a gated recurrent unit framework. Finally, a multi-layer perceptron is applied to process the auxiliary road attributes and output the final predictions. Experimental results on two publicly accessible real-world urban traffic data sets show that the proposed PSTN outperforms the state-of-the-art benchmarks by significant margins for short-term traffic conditions forecasting

Tiange Wang, Zijun Zhang, Fangfang Yang et al.

The automatic detection of major rail components using railway images is beneficial to ensure the rail transport safety. In this paper, we propose an attention-powered deep convolutional network (AttnConv-net) to detect multiple rail components including the rail, clips, and bolts. The proposed method consists of a deep convolutional neural network (DCNN) as the backbone, cascading attention blocks (CAB), and two feed forward networks (FFN). Two types of positional embedding are applied to enrich information in latent features extracted from the backbone. Based on processed latent features, the CAB aims to learn the local context of rail components including their categories and component boundaries. Final categories and bounding boxes are generated via two FFN implemented in parallel. To enhance the detection of small components, various data augmentation methods are employed in the training process. The effectiveness of the proposed AttnConv-net is validated with one real dataset and another synthesized dataset. Compared with classic convolutional neural network based methods, our proposed method simplifies the detection pipeline by eliminating the need of prior- and post-processing, which offers a new speed-quality solution to enable faster and more accurate image-based rail component detections

Tiange Wang, Zijun Zhang, Fangfang Yang et al.

Automated inspection and detection of foreign objects on railways is important for rail transportation safety as it helps prevent potential accidents and trains derailment. Most existing vision-based approaches focus on the detection of frontal intrusion objects with prior labels, such as categories and locations of the objects. In reality, foreign objects with unknown categories can appear anytime on railway tracks. In this paper, we develop a semi-supervised convolutional autoencoder based framework that only requires railway track images without prior knowledge on the foreign objects in the training process. It consists of three different modules, a bottleneck feature generator as encoder, a photographic image generator as decoder, and a reconstruction discriminator developed via adversarial learning. In the proposed framework, the problem of detecting the presence, location, and shape of foreign objects is addressed by comparing the input and reconstructed images as well as setting thresholds based on reconstruction errors. The proposed method is evaluated through comprehensive studies under different performance criteria. The results show that the proposed method outperforms some well-known benchmarking methods. The proposed framework is useful for data analytics via the train Internet-of-Things (IoT) systems

Zijun Zhang, Ruixiang Zhang, Zongpeng Li et al.

Modern generative models are usually designed to match target distributions directly in the data space, where the intrinsic dimension of data can be much lower than the ambient dimension. We argue that this discrepancy may contribute to the difficulties in training generative models. We therefore propose to map both the generated and target distributions to a latent space using the encoder of a standard autoencoder, and train the generator (or decoder) to match the target distribution in the latent space. Specifically, we enforce the consistency in both the data space and the latent space with theoretically justified data and latent reconstruction losses. The resulting generative model, which we call a perceptual generative autoencoder (PGA), is then trained with a maximum likelihood or variational autoencoder (VAE) objective. With maximum likelihood, PGAs generalize the idea of reversible generative models to unrestricted neural network architectures and arbitrary number of latent dimensions. When combined with VAEs, PGAs substantially improve over the baseline VAEs in terms of sample quality. Compared to other autoencoder-based generative models using simple priors, PGAs achieve state-of-the-art FID scores on CIFAR-10 and CelebA.

Zijun Zhang, Yining Zhang, Zongpeng Li

Multiplicative noise, including dropout, is widely used to regularize deep neural networks (DNNs), and is shown to be effective in a wide range of architectures and tasks. From an information perspective, we consider injecting multiplicative noise into a DNN as training the network to solve the task with noisy information pathways, which leads to the observation that multiplicative noise tends to increase the correlation between features, so as to increase the signal-to-noise ratio of information pathways. However, high feature correlation is undesirable, as it increases redundancy in representations. In this work, we propose non-correlating multiplicative noise (NCMN), which exploits batch normalization to remove the correlation effect in a simple yet effective way. We show that NCMN significantly improves the performance of standard multiplicative noise on image classification tasks, providing a better alternative to dropout for batch-normalized networks. Additionally, we present a unified view of NCMN and shake-shake regularization, which explains the performance gain of the latter.

Zijun Zhang, Lin Ma, Zongpeng Li et al.

Adaptive optimization algorithms, such as Adam and RMSprop, have shown better optimization performance than stochastic gradient descent (SGD) in some scenarios. However, recent studies show that they often lead to worse generalization performance than SGD, especially for training deep neural networks (DNNs). In this work, we identify the reasons that Adam generalizes worse than SGD, and develop a variant of Adam to eliminate the generalization gap. The proposed method, normalized direction-preserving Adam (ND-Adam), enables more precise control of the direction and step size for updating weight vectors, leading to significantly improved generalization performance. Following a similar rationale, we further improve the generalization performance in classification tasks by regularizing the softmax logits. By bridging the gap between SGD and Adam, we also hope to shed light on why certain optimization algorithms generalize better than others.