Temporal Feature Fusion with Sampling Pattern Optimization for Multi-echo Gradient Echo Acquisition and Image Reconstruction

Quantitative imaging in MRI usually involves acquisition and reconstruction of a series of images at multi-echo time points, which possibly requires more scan time and specific reconstruction technique compared to conventional qualitative imaging. In this work, we focus on optimizing the acquisition and reconstruction process of multi-echo gradient echo pulse sequence for quantitative susceptibility mapping as one important quantitative imaging method in MRI. A multi-echo sampling pattern optimization block extended from LOUPE-ST is proposed to optimize the k-space sampling patterns along echoes. Besides, a recurrent temporal feature fusion block is proposed and inserted into a backbone deep ADMM network to capture the signal evolution along echo time during reconstruction. Experiments show that both blocks help improve multi-echo image reconstruction performance.

NeRD: Neural Representation of Distribution for Medical Image Segmentation

We introduce Neural Representation of Distribution (NeRD) technique, a module for convolutional neural networks (CNNs) that can estimate the feature distribution by optimizing an underlying function mapping image coordinates to the feature distribution. Using NeRD, we propose an end-to-end deep learning model for medical image segmentation that can compensate the negative impact of feature distribution shifting issue caused by commonly used network operations such as padding and pooling. An implicit function is used to represent the parameter space of the feature distribution by querying the image coordinate. With NeRD, the impact of issues such as over-segmenting and missing have been reduced, and experimental results on the challenging white matter lesion segmentation and left atrial segmentation verify the effectiveness of the proposed method. The code is available via https://github.com/tinymilky/NeRD.

On the Memory Mechanism of Tensor-Power Recurrent Models

Tensor-power (TP) recurrent model is a family of non-linear dynamical systems, of which the recurrence relation consists of a p-fold (a.k.a., degree-p) tensor product. Despite such the model frequently appears in the advanced recurrent neural networks (RNNs), to this date there is limited study on its memory property, a critical characteristic in sequence tasks. In this work, we conduct a thorough investigation of the memory mechanism of TP recurrent models. Theoretically, we prove that a large degree p is an essential condition to achieve the long memory effect, yet it would lead to unstable dynamical behaviors. Empirically, we tackle this issue by extending the degree p from discrete to a differentiable domain, such that it is efficiently learnable from a variety of datasets. Taken together, the new model is expected to benefit from the long memory effect in a stable manner. We experimentally show that the proposed model achieves competitive performance compared to various advanced RNNs in both the single-cell and seq2seq architectures.

Expectation-Maximization Regularized Deep Learning for Weakly Supervised Tumor Segmentation for Glioblastoma

We present an Expectation-Maximization (EM) Regularized Deep Learning (EMReDL) model for the weakly supervised tumor segmentation. The proposed framework was tailored to glioblastoma, a type of malignant tumor characterized by its diffuse infiltration into the surrounding brain tissue, which poses significant challenge to treatment target and tumor burden estimation based on conventional structural MRI. Although physiological MRI can provide more specific information regarding tumor infiltration, the relatively low resolution hinders a precise full annotation. This has motivated us to develop a weakly supervised deep learning solution that exploits the partial labelled tumor regions. EMReDL contains two components: a physiological prior prediction model and EM-regularized segmentation model. The physiological prior prediction model exploits the physiological MRI by training a classifier to generate a physiological prior map. This map was passed to the segmentation model for regularization using the EM algorithm. We evaluated the model on a glioblastoma dataset with the available pre-operative multiparametric MRI and recurrence MRI. EMReDL was shown to effectively segment the infiltrated tumor from the partially labelled region of potential infiltration. The segmented core and infiltrated tumor showed high consistency with the tumor burden labelled by experts. The performance comparison showed that EMReDL achieved higher accuracy than published state-of-the-art models. On MR spectroscopy, the segmented region showed more aggressive features than other partial labelled region. The proposed model can be generalized to other segmentation tasks with partial labels, with the CNN architecture flexible in the framework.

On the Practicality of Differential Privacy in Federated Learning by Tuning Iteration Times

In spite that Federated Learning (FL) is well known for its privacy protection when training machine learning models among distributed clients collaboratively, recent studies have pointed out that the naive FL is susceptible to gradient leakage attacks. In the meanwhile, Differential Privacy (DP) emerges as a promising countermeasure to defend against gradient leakage attacks. However, the adoption of DP by clients in FL may significantly jeopardize the model accuracy. It is still an open problem to understand the practicality of DP from a theoretic perspective. In this paper, we make the first attempt to understand the practicality of DP in FL through tuning the number of conducted iterations. Based on the FedAvg algorithm, we formally derive the convergence rate with DP noises in FL. Then, we theoretically derive: 1) the conditions for the DP based FedAvg to converge as the number of global iterations (GI) approaches infinity; 2) the method to set the number of local iterations (LI) to minimize the negative influence of DP noises. By further substituting the Laplace and Gaussian mechanisms into the derived convergence rate respectively, we show that: 3) The DP based FedAvg with the Laplace mechanism cannot converge, but the divergence rate can be effectively prohibited by setting the number of LIs with our method; 4) The learning error of the DP based FedAvg with the Gaussian mechanism can converge to a constant number finally if we use a fixed number of LIs per GI. To verify our theoretical findings, we conduct extensive experiments using two real-world datasets. The results not only validate our analysis results, but also provide useful guidelines on how to optimize model accuracy when incorporating DP into FL

Protecting Big Data Privacy Using Randomized Tensor Network Decomposition and Dispersed Tensor Computation

Data privacy is an important issue for organizations and enterprises to securely outsource data storage, sharing, and computation on clouds / fogs. However, data encryption is complicated in terms of the key management and distribution; existing secure computation techniques are expensive in terms of computational / communication cost and therefore do not scale to big data computation. Tensor network decomposition and distributed tensor computation have been widely used in signal processing and machine learning for dimensionality reduction and large-scale optimization. However, the potential of distributed tensor networks for big data privacy preservation have not been considered before, this motivates the current study. Our primary intuition is that tensor network representations are mathematically non-unique, unlinkable, and uninterpretable; tensor network representations naturally support a range of multilinear operations for compressed and distributed / dispersed computation. Therefore, we propose randomized algorithms to decompose big data into randomized tensor network representations and analyze the privacy leakage for 1D to 3D data tensors. The randomness mainly comes from the complex structural information commonly found in big data; randomization is based on controlled perturbation applied to the tensor blocks prior to decomposition. The distributed tensor representations are dispersed on multiple clouds / fogs or servers / devices with metadata privacy, this provides both distributed trust and management to seamlessly secure big data storage, communication, sharing, and computation. Experiments show that the proposed randomization techniques are helpful for big data anonymization and efficient for big data storage and computation.

Improving BERT with Syntax-aware Local Attention

Pre-trained Transformer-based neural language models, such as BERT, have achieved remarkable results on varieties of NLP tasks. Recent works have shown that attention-based models can benefit from more focused attention over local regions. Most of them restrict the attention scope within a linear span, or confine to certain tasks such as machine translation and question answering. In this paper, we propose a syntax-aware local attention, where the attention scopes are restrained based on the distances in the syntactic structure. The proposed syntax-aware local attention can be integrated with pretrained language models, such as BERT, to render the model to focus on syntactically relevant words. We conduct experiments on various single-sentence benchmarks, including sentence classification and sequence labeling tasks. Experimental results show consistent gains over BERT on all benchmark datasets. The extensive studies verify that our model achieves better performance owing to more focused attention over syntactically relevant words.

NN-EMD: Efficiently Training Neural Networks using Encrypted Multi-sourced Datasets

Training a machine learning model over an encrypted dataset is an existing promising approach to address the privacy-preserving machine learning task, however, it is extremely challenging to efficiently train a deep neural network (DNN) model over encrypted data for two reasons: first, it requires large-scale computation over huge datasets; second, the existing solutions for computation over encrypted data, such as homomorphic encryption, is inefficient. Further, for an enhanced performance of a DNN model, we also need to use huge training datasets composed of data from multiple data sources that may not have pre-established trust relationships among each other. We propose a novel framework, NN-EMD, to train DNN over multiple encrypted datasets collected from multiple sources. Toward this, we propose a set of secure computation protocols using hybrid functional encryption schemes. We evaluate our framework for performance with regards to the training time and model accuracy on the MNIST datasets. Compared to other existing frameworks, our proposed NN-EMD framework can significantly reduce the training time, while providing comparable model accuracy and privacy guarantees as well as supporting multiple data sources. Furthermore, the depth and complexity of neural networks do not affect the training time despite introducing a privacy-preserving NN-EMD setting.

Bayesian optimization assisted unsupervised learning for efficient intra-tumor partitioning in MRI and survival prediction for glioblastoma patients

Glioblastoma is profoundly heterogeneous in microstructure and vasculature, which may lead to tumor regional diversity and distinct treatment response. Although successful in tumor sub-region segmentation and survival prediction, radiomics based on machine learning algorithms, is challenged by its robustness, due to the vague intermediate process and track changes. Also, the weak interpretability of the model poses challenges to clinical application. Here we proposed a machine learning framework to semi-automatically fine-tune the clustering algorithms and quantitatively identify stable sub-regions for reliable clinical survival prediction. Hyper-parameters are automatically determined by the global minimum of the trained Gaussian Process (GP) surrogate model through Bayesian optimization(BO) to alleviate the difficulty of tuning parameters for clinical researchers. To enhance the interpretability of the survival prediction model, we incorporated the prior knowledge of intra-tumoral heterogeneity, by segmenting tumor sub-regions and extracting sub-regional features. The results demonstrated that the global minimum of the trained GP surrogate can be used as sub-optimal hyper-parameter solutions for efficient. The sub-regions segmented based on physiological MRI can be applied to predict patient survival, which could enhance the clinical interpretability for the machine learning model.

Channel Pruning via Multi-Criteria based on Weight Dependency

Channel pruning has demonstrated its effectiveness in compressing ConvNets. In many prior arts, the importance of an output feature map is only determined by its associated filter. However, these methods ignore a small part of weights in the next layer which disappear as the feature map is removed. They ignore the dependency of the weights, so that, a part of weights are pruned without being evaluated. In addition, many pruning methods use only one criterion for evaluation, and find a sweet-spot of pruning structure and accuracy in a trial-and-error fashion, which can be time-consuming. To address the above issues, we proposed a channel pruning algorithm via multi-criteria based on weight dependency, CPMC, which can compress a variety of models efficiently. We design the importance of the feature map in three aspects, including its associated weight value, computational cost and parameter quantity. Use the phenomenon of weight dependency, We get the importance by assessing its associated filter and the corresponding partial weights of the next layer. Then we use global normalization to achieve cross-layer comparison. Our method can compress various CNN models, including VGGNet, ResNet and DenseNet, on various image classification datasets. Extensive experiments have shown CPMC outperforms the others significantly.