We predict restaurant ratings from Yelp reviews based on Yelp Open Dataset. Data distribution is presented, and one balanced training dataset is built. Two vectorizers are experimented for feature engineering. Four machine learning models including Naive Bayes, Logistic Regression, Random Forest, and Linear Support Vector Machine are implemented. Four transformer-based models containing BERT, DistilBERT, RoBERTa, and XLNet are also applied. Accuracy, weighted F1 score, and confusion matrix are used for model evaluation. XLNet achieves 70% accuracy for 5-star classification compared with Logistic Regression with 64% accuracy.
Recognizing material from color images is still a challenging problem today. While deep neural networks provide very good results on object recognition and has been the topic of a huge amount of papers in the last decade, their adaptation to material images still requires some works to reach equivalent accuracies. Nevertheless, recent studies achieve very good results in material recognition with deep learning and we propose, in this paper, to review most of them by focusing on three aspects: material image datasets, influence of the context and ad hoc descriptors for material appearance. Every aspect is introduced by a systematic manner and results from representative works are cited. We also present our own studies in this area and point out some open challenges for future works.
According to 't Hooft, restoring Poincar\'e invariance in a holographic cellular automaton (CA) requires two distinct evolution laws that commute. We explore how this is realized in the AdS/CFT framework, assuming commutativity as a fundamental principle--much like general covariance once did--for encoding curvature. In our setup, physical processes in a given spacetime are encoded in a CA; to preserve Poincar\'e symmetry, the spacetime curvature must effectively vanish, so we consider a near-extremal black D3-brane solution, in which both the stretched horizon and the conformal boundary are approximated by Minkowski space. AdS/CFT implies a spatial evolution law connecting these hypersurfaces. Commutativity means the final state does not depend on the order of time evolution on each hypersurface and spatial evolution between them, forcing the time evolution law on the horizon and boundary to coincide. To satisfy all these conditions, we aim to demonstrate that the spatial evolution law inevitably encapsulates the curvature of the bulk, including quantum effects. For a computational model, we compactify the hyperplanes to tori, reducing the degrees of freedom to a finite number; taking these tori to infinite size then restores Poincar\'e symmetry. We propose a deep learning algorithm that, given a known time evolution law and commutativity, deduces the corresponding spatial evolution law.
Rate-Distortion Optimized Quantization (RDOQ) has played an important role in the coding performance of recent video compression standards such as H.264/AVC, H.265/HEVC, VP9 and AV1. This scheme yields significant reductions in bit-rate at the expense of relatively small increases in distortion. Typically, RDOQ algorithms are prohibitively expensive to implement on real-time hardware encoders due to their sequential nature and their need to frequently obtain entropy coding costs. This work addresses this limitation using a neural network-based approach, which learns to trade-off rate and distortion during offline supervised training. As these networks are based solely on standard arithmetic operations that can be executed on existing neural network hardware, no additional area-on-chip needs to be reserved for dedicated RDOQ circuitry. We train two classes of neural networks, a fully-convolutional network and an auto-regressive network, and evaluate each as a post-quantization step designed to refine cheap quantization schemes such as scalar quantization (SQ). Both network architectures are designed to have a low computational overhead. After training they are integrated into the HM 16.20 implementation of HEVC, and their video coding performance is evaluated on a subset of the H.266/VVC SDR common test sequences. Comparisons are made to RDOQ and SQ implementations in HM 16.20. Our method achieves 1.64% BD-rate savings on luminosity compared to the HM SQ anchor, and on average reaches 45% of the performance of the iterative HM RDOQ algorithm.
In the framework of three-active-neutrino mixing, the charge parity phase, the neutrino mass ordering, and the octant of $\theta_{23}$ remain unknown. The Deep Underground Neutrino Experiment (DUNE) is a next-generation long-baseline neutrino oscillation experiment, which aims to address these questions by measuring the oscillation patterns of $\nu_\mu/\nu_e$ and $\bar\nu_\mu/\bar\nu_e$ over a range of energies spanning the first and second oscillation maxima. DUNE far detector modules are based on liquid argon TPC (LArTPC) technology. A LArTPC offers excellent spatial resolution, high neutrino detection efficiency, and superb background rejection, while reconstruction in LArTPC is challenging. Deep learning methods, in particular, Convolutional Neural Networks (CNNs), have demonstrated success in classification problems such as particle identification in DUNE and other neutrino experiments. However, reconstruction of neutrino energy and final state particle momenta with deep learning methods is yet to be developed for a full AI-based reconstruction chain. To precisely reconstruct these kinematic characteristics of detected interactions at DUNE, we have developed and will present two CNN-based methods, 2-D and 3-D, for the reconstruction of final state particle direction and energy, as well as neutrino energy. Combining particle masses with the kinetic energy and the direction reconstructed by our work, the four-momentum of final state particles can be obtained. Our models show considerable improvements compared to the traditional methods for both scenarios.
Matrix completion has received vast amount of attention and research due to its wide applications in various study fields. Existing methods of matrix completion consider only nonlinear (or linear) relations among entries in a data matrix and ignore linear (or nonlinear) relationships latent. This paper introduces a new latent variables model for data matrix which is a combination of linear and nonlinear models and designs a novel deep-neural-network-based matrix completion algorithm to address both linear and nonlinear relations among entries of data matrix. The proposed method consists of two branches. The first branch learns the latent representations of columns and reconstructs the columns of the partially observed matrix through a series of hidden neural network layers. The second branch does the same for the rows. In addition, based on multi-task learning principles, we enforce these two branches work together and introduce a new regularization technique to reduce over-fitting. More specifically, the missing entries of data are recovered as a main task and manifold learning is performed as an auxiliary task. The auxiliary task constrains the weights of the network so it can be considered as a regularizer, improving the main task and reducing over-fitting. Experimental results obtained on the synthetic data and several real-world data verify the effectiveness of the proposed method compared with state-of-the-art matrix completion methods.
In this paper, we present an experiment on using deep learning and transfer learning techniques for emotion analysis in tweets and suggest a method to interpret our deep learning models. The proposed approach for emotion analysis combines a Long Short Term Memory (LSTM) network with a Convolutional Neural Network (CNN). Then we extend this approach for emotion intensity prediction using transfer learning technique. Furthermore, we propose a technique to visualize the importance of each word in a tweet to get a better understanding of the model. Experimentally, we show in our analysis that the proposed models outperform the state-of-the-art in emotion classification while maintaining competitive results in predicting emotion intensity.
Video action recognition is one of the representative tasks for video understanding. Over the last decade, we have witnessed great advancements in video action recognition thanks to the emergence of deep learning. But we also encountered new challenges, including modeling long-range temporal information in videos, high computation costs, and incomparable results due to datasets and evaluation protocol variances. In this paper, we provide a comprehensive survey of over 200 existing papers on deep learning for video action recognition. We first introduce the 17 video action recognition datasets that influenced the design of models. Then we present video action recognition models in chronological order: starting with early attempts at adapting deep learning, then to the two-stream networks, followed by the adoption of 3D convolutional kernels, and finally to the recent compute-efficient models. In addition, we benchmark popular methods on several representative datasets and release code for reproducibility. In the end, we discuss open problems and shed light on opportunities for video action recognition to facilitate new research ideas.
A novel unsupervised deep learning method is developed to identify individual-specific large scale brain functional networks (FNs) from resting-state fMRI (rsfMRI) in an end-to-end learning fashion. Our method leverages deep Encoder-Decoder networks and conventional brain decomposition models to identify individual-specific FNs in an unsupervised learning framework and facilitate fast inference for new individuals with one forward pass of the deep network. Particularly, convolutional neural networks (CNNs) with an Encoder-Decoder architecture are adopted to identify individual-specific FNs from rsfMRI data by optimizing their data fitting and sparsity regularization terms that are commonly used in brain decomposition models. Moreover, a time-invariant representation learning module is designed to learn features invariant to temporal orders of time points of rsfMRI data. The proposed method has been validated based on a large rsfMRI dataset and experimental results have demonstrated that our method could obtain individual-specific FNs which are consistent with well-established FNs and are informative for predicting brain age, indicating that the individual-specific FNs identified truly captured the underlying variability of individualized functional neuroanatomy.
These are the notes for the lectures that I was giving during Fall 2020 at the Moscow Institute of Physics and Technology (MIPT) and at the Yandex School of Data Analysis (YSDA). The notes cover some aspects of initialization, loss landscape, generalization, and a neural tangent kernel theory. While many other topics (e.g. expressivity, a mean-field theory, a double descent phenomenon) are missing in the current version, we plan to add them in future revisions.
Password guessing approaches via deep learning have recently been investigated with significant breakthroughs in their ability to generate novel, realistic password candidates. In the present work we study a broad collection of deep learning and probabilistic based models in the light of password guessing: attention-based deep neural networks, autoencoding mechanisms and generative adversarial networks. We provide novel generative deep-learning models in terms of variational autoencoders exhibiting state-of-art sampling performance, yielding additional latent-space features such as interpolations and targeted sampling. Lastly, we perform a thorough empirical analysis in a unified controlled framework over well-known datasets (RockYou, LinkedIn, Youku, Zomato, Pwnd). Our results not only identify the most promising schemes driven by deep neural networks, but also illustrate the strengths of each approach in terms of generation variability and sample uniqueness.
The extent to which the arbitrarily selected L2 regularization hyperparameter value affects the outcome of semantic segmentation with deep learning is demonstrated. Demonstrations rely on training U-net on small LGE-MRI datasets using the arbitrarily selected L2 regularization values. The remaining hyperparameters are to be manually adjusted or tuned only when 10 % of all epochs are reached before the training validation accuracy reaches 90%. Semantic segmentation with deep learning outcomes are objectively and subjectively evaluated against the manual ground truth segmentation.
A well-known perceptual consequence of categorization in humans and other animals, called categorical perception, is notably characterized by a within-category compression and a between-category separation: two items, close in input space, are perceived closer if they belong to the same category than if they belong to different categories. Elaborating on experimental and theoretical results in cognitive science, here we study categorical effects in artificial neural networks. We combine a theoretical analysis that makes use of mutual and Fisher information quantities, and a series of numerical simulations on networks of increasing complexity. These formal and numerical analyses provide insights into the geometry of the neural representation in deep layers, with expansion of space near category boundaries and contraction far from category boundaries. We investigate categorical representation by using two complementary approaches: one mimics experiments in psychophysics and cognitive neuroscience by means of morphed continua between stimuli of different categories, while the other introduces a categoricality index that, for each layer in the network, quantifies the separability of the categories at the neural population level. We show on both shallow and deep neural networks that category learning automatically induces categorical perception. We further show that the deeper a layer, the stronger the categorical effects. As an outcome of our study, we propose a coherent view of the efficacy of different heuristic practices of the dropout regularization technique. More generally, our view, which finds echoes in the neuroscience literature, insists on the differential impact of noise in any given layer depending on the geometry of the neural representation that is being learned, i.e. on how this geometry reflects the structure of the categories.
Synthetic aperture radar (SAR) images are affected by a spatially-correlated and signal-dependent noise called speckle, which is very severe and may hinder image exploitation. Despeckling is an important task that aims at removing such noise, so as to improve the accuracy of all downstream image processing tasks. The first despeckling methods date back to the 1970's, and several model-based algorithms have been developed in the subsequent years. The field has received growing attention, sparkled by the availability of powerful deep learning models that have yielded excellent performance for inverse problems in image processing. This paper surveys the literature on deep learning methods applied to SAR despeckling, covering both the supervised and the more recent self-supervised approaches. We provide a critical analysis of existing methods with the objective to recognize the most promising research lines, to identify the factors that have limited the success of deep models, and to propose ways forward in an attempt to fully exploit the potential of deep learning for SAR despeckling.
Lung cancer is the leading cause of cancer death and morbidity worldwide. Many studies have shown machine learning models to be effective at detecting lung nodules from chest X-ray images. However, these techniques have yet to be embraced by the medical community due to several practical, ethical, and regulatory constraints stemming from the black-box nature of deep learning models. Additionally, most lung nodules visible on chest X-ray are benign; therefore, the narrow task of computer vision-based lung nodule detection cannot be equated to automated lung cancer detection. Addressing both concerns, this study introduces a novel hybrid deep learning and decision tree-based computer vision model which presents lung cancer malignancy predictions as interpretable decision trees. The deep learning component of this process is trained using a large publicly available dataset on pathological biomarkers associated with lung cancer. These models are then used to inference biomarker scores for chest X-ray images from two, independent data sets for which malignancy metadata is available. We mine multi-variate predictive models by fitting shallow decision trees to the malignancy stratified datasets and interrogate a range of metrics to determine the best model. Our best decision tree model achieves sensitivity and specificity of 86.7% and 80.0% respectively with a positive predictive value of 92.9%. Decision trees mined using this method may be considered as a starting point for refinement into clinically useful multi-variate lung cancer malignancy models for implementation as a workflow augmentation tool to improve the efficiency of human radiologists.
Topology optimization has emerged as a popular approach to refine a component's design and increase its performance. However, current state-of-the-art topology optimization frameworks are compute-intensive, mainly due to multiple finite element analysis iterations required to evaluate the component's performance during the optimization process. Recently, machine learning (ML)-based topology optimization methods have been explored by researchers to alleviate this issue. However, previous ML approaches have mainly been demonstrated on simple two-dimensional applications with low-resolution geometry. Further, current methods are based on a single ML model for end-to-end prediction, which requires a large dataset for training. These challenges make it non-trivial to extend current approaches to higher resolutions. In this paper, we develop deep learning-based frameworks consistent with traditional topology optimization algorithms for 3D topology optimization with a reasonably fine (high) resolution. We achieve this by training multiple networks, each learning a different step of the overall topology optimization methodology, making the framework more consistent with the topology optimization algorithm. We demonstrate the application of our framework on both 2D and 3D geometries. The results show that our approach predicts the final optimized design better (5.76x reduction in total compliance MSE in 2D; 2.03x reduction in total compliance MSE in 3D) than current ML-based topology optimization methods.
Phase contrast transmission electron microscopy (TEM) is a powerful tool for imaging the local atomic structure of materials. TEM has been used heavily in studies of defect structures of 2D materials such as monolayer graphene due to its high dose efficiency. However, phase contrast imaging can produce complex nonlinear contrast, even for weakly-scattering samples. It is therefore difficult to develop fully-automated analysis routines for phase contrast TEM studies using conventional image processing tools. For automated analysis of large sample regions of graphene, one of the key problems is segmentation between the structure of interest and unwanted structures such as surface contaminant layers. In this study, we compare the performance of a conventional Bragg filtering method to a deep learning routine based on the U-Net architecture. We show that the deep learning method is more general, simpler to apply in practice, and produces more accurate and robust results than the conventional algorithm. We provide easily-adaptable source code for all results in this paper, and discuss potential applications for deep learning in fully-automated TEM image analysis.
Automatic medical image segmentation plays a critical role in scientific research and medical care. Existing high-performance deep learning methods typically rely on large training datasets with high-quality manual annotations, which are difficult to obtain in many clinical applications. Here, we introduce Annotation-effIcient Deep lEarning (AIDE), an open-source framework to handle imperfect training datasets. Methodological analyses and empirical evaluations are conducted, and we demonstrate that AIDE surpasses conventional fully-supervised models by presenting better performance on open datasets possessing scarce or noisy annotations. We further test AIDE in a real-life case study for breast tumor segmentation. Three datasets containing 11,852 breast images from three medical centers are employed, and AIDE, utilizing 10% training annotations, consistently produces segmentation maps comparable to those generated by fully-supervised counterparts or provided by independent radiologists. The 10-fold enhanced efficiency in utilizing expert labels has the potential to promote a wide range of biomedical applications.
Using the optical camera in remote sensing is limited in various environmental conditions. This paper presents a system of combining deep learning and image transform algorithms to detect landslide location in satellite images. In the deep learning part, a convolution neural network is used to classify satellite images contain landslides. From landslide images classified, in order to accurately identify landslides under different lighting conditions, this paper proposes a transformation algorithm Hue - Bi-dimensional empirical mode decomposition (H-BEMD) to determine the landslide region and size. After the location of landslide is detected, we discover the size change of the landslide based on different time points. In this study, we record an accuracy of up to 96% in the classification process, and the accuracy of landslide location almost absolute.
Majorly classical Active Learning (AL) approach usually uses statistical theory such as entropy and margin to measure instance utility, however it fails to capture the data distribution information contained in the unlabeled data. This can eventually cause the classifier to select outlier instances to label. Meanwhile, the loss associated with mislabeling an instance in a typical classification task is much higher than the loss associated with the opposite error. To address these challenges, we propose a Cost-Based Bugdet Active Learning (CBAL) which considers the classification uncertainty as well as instance diversity in a population constrained by a budget. A principled approach based on the min-max is considered to minimize both the labeling and decision cost of the selected instances, this ensures a near-optimal results with significantly less computational effort. Extensive experimental results show that the proposed approach outperforms several state-of -the-art active learning approaches.