Machine learning (ML) is increasingly being adopted in a wide variety of application domains. Usually, a well-performing ML model, especially, emerging deep neural network model, relies on a large volume of training data and high-powered computational resources. The need for a vast volume of available data raises serious privacy concerns because of the risk of leakage of highly privacy-sensitive information and the evolving regulatory environments that increasingly restrict access to and use of privacy-sensitive data. Furthermore, a trained ML model may also be vulnerable to adversarial attacks such as membership/property inference attacks and model inversion attacks. Hence, well-designed privacy-preserving ML (PPML) solutions are crucial and have attracted increasing research interest from academia and industry. More and more efforts of PPML are proposed via integrating privacy-preserving techniques into ML algorithms, fusing privacy-preserving approaches into ML pipeline, or designing various privacy-preserving architectures for existing ML systems. In particular, existing PPML arts cross-cut ML, system, security, and privacy; hence, there is a critical need to understand state-of-art studies, related challenges, and a roadmap for future research. This paper systematically reviews and summarizes existing privacy-preserving approaches and proposes a PGU model to guide evaluation for various PPML solutions through elaborately decomposing their privacy-preserving functionalities. The PGU model is designed as the triad of Phase, Guarantee, and technical Utility. Furthermore, we also discuss the unique characteristics and challenges of PPML and outline possible directions of future work that benefit a wide range of research communities among ML, distributed systems, security, and privacy areas.

Human Motion Prediction is a crucial task in computer vision and robotics. It has versatile application potentials such as in the area of human-robot interactions, human action tracking for airport security systems, autonomous car navigation, computer gaming to name a few. However, predicting human motion based on past actions is an extremely challenging task due to the difficulties in detecting spatial and temporal features correctly. To detect temporal features in human poses, we propose an Inception Residual Block(IRB), due to its inherent capability of processing multiple kernels to capture salient features. Here, we propose to use multiple 1-D Convolution Neural Network (CNN) with different kernel sizes and input sequence lengths and concatenate them to get proper embedding. As kernels strides over different receptive fields, they detect smaller and bigger salient features at multiple temporal scales. Our main contribution is to propose a residual connection between input and the output of the inception block to have a continuity between the previously observed pose and the next predicted pose. With this proposed architecture, it learns prior knowledge much better about human poses and we achieve much higher prediction accuracy as detailed in the paper. Subsequently, we further propose to feed the output of the inception residual block as an input to the Graph Convolution Neural Network (GCN) due to its better spatial feature learning capability. We perform a parametric analysis for better designing of our model and subsequently, we evaluate our approach on the Human 3.6M dataset and compare our short-term as well as long-term predictions with the state of the art papers, where our model outperforms most of the pose results, the detailed reasons of which have been elaborated in the paper.

The advances in deep neural networks (DNN) have significantly enhanced real-time detection of anomalous data in IoT applications. However, the complexity-accuracy-delay dilemma persists: complex DNN models offer higher accuracy, but typical IoT devices can barely afford the computation load, and the remedy of offloading the load to the cloud incurs long delay. In this paper, we address this challenge by proposing an adaptive anomaly detection scheme with hierarchical edge computing (HEC). Specifically, we first construct multiple anomaly detection DNN models with increasing complexity, and associate each of them to a corresponding HEC layer. Then, we design an adaptive model selection scheme that is formulated as a contextual-bandit problem and solved by using a reinforcement learning policy network. We also incorporate a parallelism policy training method to accelerate the training process by taking advantage of distributed models. We build an HEC testbed using real IoT devices, implement and evaluate our contextual-bandit approach with both univariate and multivariate IoT datasets. In comparison with both baseline and state-of-the-art schemes, our adaptive approach strikes the best accuracy-delay tradeoff on the univariate dataset, and achieves the best accuracy and F1-score on the multivariate dataset with only negligibly longer delay than the best (but inflexible) scheme.

Several studies have reported the inability of Transformer models to generalize compositionally, a key type of generalization in many NLP tasks such as semantic parsing. In this paper we explore the design space of Transformer models showing that the inductive biases given to the model by several design decisions significantly impact compositional generalization. Through this exploration, we identified Transformer configurations that generalize compositionally significantly better than previously reported in the literature in a diverse set of compositional tasks, and that achieve state-of-the-art results in a semantic parsing compositional generalization benchmark (COGS), and a string edit operation composition benchmark (PCFG).

Deep learning models achieve strong performance for radiology image classification, but their practical application is bottlenecked by the need for large labeled training datasets. Semi-supervised learning (SSL) approaches leverage small labeled datasets alongside larger unlabeled datasets and offer potential for reducing labeling cost. In this work, we introduce NoTeacher, a novel consistency-based SSL framework which incorporates probabilistic graphical models. Unlike Mean Teacher which maintains a teacher network updated via a temporal ensemble, NoTeacher employs two independent networks, thereby eliminating the need for a teacher network. We demonstrate how NoTeacher can be customized to handle a range of challenges in radiology image classification. Specifically, we describe adaptations for scenarios with 2D and 3D inputs, uni and multi-label classification, and class distribution mismatch between labeled and unlabeled portions of the training data. In realistic empirical evaluations on three public benchmark datasets spanning the workhorse modalities of radiology (X-Ray, CT, MRI), we show that NoTeacher achieves over 90-95% of the fully supervised AUROC with less than 5-15% labeling budget. Further, NoTeacher outperforms established SSL methods with minimal hyperparameter tuning, and has implications as a principled and practical option for semisupervised learning in radiology applications.

Deep Neural Networks (DNNs) are vulnerable to adversarial examples which would inveigle neural networks to make prediction errors with small per- turbations on the input images. Researchers have been devoted to promoting the research on the universal adversarial perturbations (UAPs) which are gradient-free and have little prior knowledge on data distributions. Procedural adversarial noise at- tack is a data-free universal perturbation generation method. In this paper, we propose two universal adversarial perturbation (UAP) generation methods based on procedural noise functions: Simplex noise and Worley noise. In our framework, the shading which disturbs visual classification is generated with rendering technology. Without changing the semantic representations, the adversarial examples generated via our methods show superior performance on the attack.

For the past ten years, CNN has reigned supreme in the world of computer vision, but recently, Transformer is on the rise. However, the quadratic computational cost of self-attention has become a severe problem of practice. There has been much research on architectures without CNN and self-attention in this context. In particular, MLP-Mixer is a simple idea designed using MLPs and hit an accuracy comparable to the Vision Transformer. However, the only inductive bias in this architecture is the embedding of tokens. Thus, there is still a possibility to build a non-convolutional inductive bias into the architecture itself, and we built in an inductive bias using two simple ideas. A way is to divide the token-mixing block vertically and horizontally. Another way is to make spatial correlations denser among some channels of token-mixing. With this approach, we were able to improve the accuracy of the MLP-Mixer while reducing its parameters and computational complexity. Compared to other MLP-based models, the proposed model, named RaftMLP has a good balance of computational complexity, the number of parameters, and actual memory usage. In addition, our work indicates that MLP-based models have the potential to replace CNNs by adopting inductive bias. The source code in PyTorch version is available at \url{https://github.com/okojoalg/raft-mlp}.

Natural numerical networks are introduced as a new classification algorithm based on the numerical solution of nonlinear partial differential equations of forward-backward diffusion type on complete graphs. The proposed natural numerical network is applied to open important environmental and nature conservation task, the automated identification of protected habitats by using satellite images. In the natural numerical network, the forward diffusion causes the movement of points in a feature space toward each other. The opposite effect, keeping the points away from each other, is caused by backward diffusion. This yields the desired classification. The natural numerical network contains a few parameters that are optimized in the learning phase of the method. After learning parameters and optimizing the topology of the network graph, classification necessary for habitat identification is performed. A relevancy map for each habitat is introduced as a tool for validating the classification and finding new Natura 2000 habitat appearances.

The performance of deep neural networks typically increases with the number of training images. However, not all images have the same importance towards improved performance and robustness. In fetal brain MRI, abnormalities exacerbate the variability of the developing brain anatomy compared to non-pathological cases. A small number of abnormal cases, as is typically available in clinical datasets used for training, are unlikely to fairly represent the rich variability of abnormal developing brains. This leads machine learning systems trained by maximizing the average performance to be biased toward non-pathological cases. This problem was recently referred to as hidden stratification. To be suited for clinical use, automatic segmentation methods need to reliably achieve high-quality segmentation outcomes also for pathological cases. In this paper, we show that the state-of-the-art deep learning pipeline nnU-Net has difficulties to generalize to unseen abnormal cases. To mitigate this problem, we propose to train a deep neural network to minimize a percentile of the distribution of per-volume loss over the dataset. We show that this can be achieved by using Distributionally Robust Optimization (DRO). DRO automatically reweights the training samples with lower performance, encouraging nnU-Net to perform more consistently on all cases. We validated our approach using a dataset of 368 fetal brain T2w MRIs, including 124 MRIs of open spina bifida cases and 51 MRIs of cases with other severe abnormalities of brain development.

1) The local environment and land usages have changed a lot during the past one hundred years. Historical documents and materials are crucial in understanding and following these changes. Historical documents are, therefore, an important piece in the understanding of the impact and consequences of land usage change. This, in turn, is important in the search of restoration projects that can be conducted to turn and reduce harmful and unsustainable effects originating from changes in the land-usage. 2) This work extracts information on the historical location and geographical distribution of wetlands, from hand-drawn maps. This is achieved by using deep learning (DL), and more specifically a convolutional neural network (CNN). The CNN model is trained on a manually pre-labelled dataset on historical wetlands in the area of J\"onk\"oping county in Sweden. These are all extracted from the historical map called "Generalstabskartan". 3) The presented CNN performs well and achieves a $F_1$-score of 0.886 when evaluated using a 10-fold cross validation over the data. The trained models are additionally used to generate a GIS layer of the presumable historical geographical distribution of wetlands for the area that is depicted in the southern collection in Generalstabskartan, which covers the southern half of Sweden. This GIS layer is released as an open resource and can be freely used. 4) To summarise, the presented results show that CNNs can be a useful tool in the extraction and digitalisation of non-textual information in historical documents, such as historical maps. A modern GIS material that can be used to further understand the past land-usage change is produced within this research.