With the rapid rise of neural architecture search, the ability to understand its complexity from the perspective of a search algorithm is desirable. Recently, Traor\'e et al. have proposed the framework of Fitness Landscape Footprint to help describe and compare neural architecture search problems. It attempts at describing why a search strategy might be successful, struggle or fail on a target task. Our study leverages this methodology in the context of searching across sensors, including sensor data fusion. In particular, we apply the Fitness Landscape Footprint to the real-world image classification problem of So2Sat LCZ42, in order to identify the most beneficial sensor to our neural network hyper-parameter optimization problem. From the perspective of distributions of fitness, our findings indicate a similar behaviour of the search space for all sensors: the longer the training time, the larger the overall fitness, and more flatness in the landscapes (less ruggedness and deviation). Regarding sensors, the better the fitness they enable (Sentinel-2), the better the search trajectories (smoother, higher persistence). Results also indicate very similar search behaviour for sensors that can be decently fitted by the search space (Sentinel-2 and fusion).
Approximating wind flows using computational fluid dynamics (CFD) methods can be time-consuming. Creating a tool for interactively designing prototypes while observing the wind flow change requires simpler models to simulate faster. Instead of running numerical approximations resulting in detailed calculations, data-driven methods in deep learning might be able to give similar results in a fraction of the time. This work rephrases the problem from computing 3D flow fields using CFD to a 2D image-to-image translation-based problem on the building footprints to predict the flow field at pedestrian height level. We investigate the use of generative adversarial networks (GAN), such as Pix2Pix [1] and CycleGAN [2] representing state-of-the-art for image-to-image translation task in various domains as well as U-Net autoencoder [3]. The models can learn the underlying distribution of a dataset in a data-driven manner, which we argue can help the model learn the underlying Reynolds-averaged Navier-Stokes (RANS) equations from CFD. We experiment on novel simulated datasets on various three-dimensional bluff-shaped buildings with and without height information. Moreover, we present an extensive qualitative and quantitative evaluation of the generated images for a selection of models and compare their performance with the simulations delivered by CFD. We then show that adding positional data to the input can produce more accurate results by proposing a general framework for injecting such information on the different architectures. Furthermore, we show that the models performances improve by applying attention mechanisms and spectral normalization to facilitate stable training.
The evolution of wireless systems has led to a continuous increase in the demand for radio frequency spectrum. To address this issue, a technology that has received a lot of attention is In-Band Full-Duplex (IBFD). The interest in IBFD systems stems from its capability to simultaneously transmit and receive data in the same frequency. Cancelling the self interference (SI) from the transmitter to the collocated receiver plays a pivotal role in the performance of the system. There are two types of SI cancellation (SIC) approaches, passive and active. In this research, the focus is on active cancellation and, in particular, SIC in the digital domain. Among the direct and backscattered SI, the former has been studied for a long time; therefore, the backscatter is considered in this research and two SIC approaches are analyzed. The first achieves SIC through beamforming. This requires knowing the angle of the received SI to put the beam null-space in this direction. The second method removes SI by employing an Artificial Neural Networks (ANNs). Using an ANN, there is no need to know the direction of the SI. The neural network is trained with pilots which results in the network being able to separate the desired signal from the SI at the receiver. Bayesian Neural Networks show the importance of the weights and assign a parameter that facilitates ignoring the less significant ones. Through comparative simulations we demonstrate that the ANN-based SIC achieves equivalent bit error rate performance as two beamforming methods.
Metric $k$-center clustering is a fundamental unsupervised learning primitive. Although widely used, this primitive is heavily affected by noise in the data, so that a more sensible variant seeks for the best solution that disregards a given number $z$ of points of the dataset, called outliers. We provide efficient algorithms for this important variant in the streaming model under the sliding window setting, where, at each time step, the dataset to be clustered is the window $W$ of the most recent data items. Our algorithms achieve $O(1)$ approximation and, remarkably, require a working memory linear in $k+z$ and only logarithmic in $|W|$. As a by-product, we show how to estimate the effective diameter of the window $W$, which is a measure of the spread of the window points, disregarding a given fraction of noisy distances. We also provide experimental evidence of the practical viability of our theoretical results.
Anomaly detection is the process of identifying abnormal instances or events in data sets which deviate from the norm significantly. In this study, we propose a signatures based machine learning algorithm to detect rare or unexpected items in a given data set of time series type. We present applications of signature or randomized signature as feature extractors for anomaly detection algorithms; additionally we provide an easy, representation theoretic justification for the construction of randomized signatures. Our first application is based on synthetic data and aims at distinguishing between real and fake trajectories of stock prices, which are indistinguishable by visual inspection. We also show a real life application by using transaction data from the cryptocurrency market. In this case, we are able to identify pump and dump attempts organized on social networks with F1 scores up to 88% by means of our unsupervised learning algorithm, thus achieving results that are close to the state-of-the-art in the field based on supervised learning.
Cancer is one of the leading causes of death worldwide, and head and neck (H&N) cancer is amongst the most prevalent types. Positron emission tomography and computed tomography are used to detect and segment the tumor region. Clinically, tumor segmentation is extensively time-consuming and prone to error. Machine learning, and deep learning in particular, can assist to automate this process, yielding results as accurate as the results of a clinician. In this research study, we develop a vision transformers-based method to automatically delineate H&N tumor, and compare its results to leading convolutional neural network (CNN)-based models. We use multi-modal data of CT and PET scans to do this task. We show that the selected transformer-based model can achieve results on a par with CNN-based ones. With cross validation, the model achieves a mean dice similarity coefficient of 0.736, mean precision of 0.766 and mean recall of 0.766. This is only 0.021 less than the 2020 competition winning model in terms of the DSC score. This indicates that the exploration of transformer-based models is a promising research area.
Brain tissue segmentation has demonstrated great utility in quantifying MRI data through Voxel-Based Morphometry and highlighting subtle structural changes associated with various conditions within the brain. However, manual segmentation is highly labor-intensive, and automated approaches have struggled due to properties inherent to MRI acquisition, leaving a great need for an effective segmentation tool. Despite the recent success of deep convolutional neural networks (CNNs) for brain tissue segmentation, many such solutions do not generalize well to new datasets, which is critical for a reliable solution. Transformers have demonstrated success in natural image segmentation and have recently been applied to 3D medical image segmentation tasks due to their ability to capture long-distance relationships in the input where the local receptive fields of CNNs struggle. This study introduces a novel CNN-Transformer hybrid architecture designed for brain tissue segmentation. We validate our model's performance across four multi-site T1w MRI datasets, covering different vendors, field strengths, scan parameters, time points, and neuropsychiatric conditions. In all situations, our model achieved the greatest generality and reliability. Out method is inherently robust and can serve as a valuable tool for brain-related T1w MRI studies. The code for the TABS network is available at: https://github.com/raovish6/TABS.
Training deep neural networks (DNNs) is time-consuming. While most existing solutions try to overlap/schedule computation and communication for efficient training, this paper goes one step further by skipping computing and communication through DNN layer freezing. Our key insight is that the training progress of internal DNN layers differs significantly, and front layers often become well-trained much earlier than deep layers. To explore this, we first introduce the notion of training plasticity to quantify the training progress of internal DNN layers. Then we design KGT, a knowledge-guided DNN training system that employs semantic knowledge from a reference model to accurately evaluate individual layers' training plasticity and safely freeze the converged ones, saving their corresponding backward computation and communication. Our reference model is generated on the fly using quantization techniques and runs forward operations asynchronously on available CPUs to minimize the overhead. In addition, KGT caches the intermediate outputs of the frozen layers with prefetching to further skip the forward computation. Our implementation and testbed experiments with popular vision and language models show that KGT achieves 19%-43% training speedup w.r.t. the state-of-the-art without sacrificing accuracy.
Existing explanation methods for black-box supervised learning models generally work by building local models that explain the models behaviour for a particular data item. It is possible to make global explanations, but the explanations may have low fidelity for complex models. Most of the prior work on explainable models has been focused on classification problems, with less attention on regression. We propose a new manifold visualization method, SLISEMAP, that at the same time finds local explanations for all of the data items and builds a two-dimensional visualization of model space such that the data items explained by the same model are projected nearby. We provide an open source implementation of our methods, implemented by using GPU-optimized PyTorch library. SLISEMAP works both on classification and regression models. We compare SLISEMAP to most popular dimensionality reduction methods and some local explanation methods. We provide mathematical derivation of our problem and show that SLISEMAP provides fast and stable visualizations that can be used to explain and understand black box regression and classification models.
The latest sheet stamping processes enable efficient manufacturing of complex shape structural components that have high stiffness to weight ratios, but these processes can introduce defects. To assist component design for stamping processes, this paper presents a novel deep-learning-based platform for optimising 3D component geometries. The platform adopts a non-parametric modelling approach that is capable of optimising arbitrary geometries from multiple geometric parameterisation schema. This approach features the interaction of two neural networks: 1) a geometry generator and 2) a manufacturing performance evaluator. The generator predicts continuous 3D signed distance fields (SDFs) for geometries of different classes, and each SDF is conditioned on a latent vector. The zero-level-set of each SDF implicitly represents a generated geometry. Novel training strategies for the generator are introduced and include a new loss function which is tailored for sheet stamping applications. These strategies enable the differentiable generation of high quality, large scale component geometries with tight local features for the first time. The evaluator maps a 2D projection of these generated geometries to their post-stamping physical (e.g., strain) distributions. Manufacturing constraints are imposed based on these distributions and are used to formulate a novel objective function for optimisation. A new gradient-based optimisation technique is employed to iteratively update the latent vectors, and therefore geometries, to minimise this objective function and thus meet the manufacturing constraints. Case studies based on optimising box geometries subject to a sheet thinning constraint for a hot stamping process are presented and discussed. The results show that expressive geometric changes are achievable, and that these changes are driven by stamping performance.