Efficient custom pooling techniques that can aggressively trim the dimensions of a feature map and thereby reduce inference compute and memory footprint for resource-constrained computer vision applications have recently gained significant traction. However, prior pooling works extract only the local context of the activation maps, limiting their effectiveness. In contrast, we propose a novel non-local self-attentive pooling method that can be used as a drop-in replacement to the standard pooling layers, such as max/average pooling or strided convolution. The proposed self-attention module uses patch embedding, multi-head self-attention, and spatial-channel restoration, followed by sigmoid activation and exponential soft-max. This self-attention mechanism efficiently aggregates dependencies between non-local activation patches during down-sampling. Extensive experiments on standard object classification and detection tasks with various convolutional neural network (CNN) architectures demonstrate the superiority of our proposed mechanism over the state-of-the-art (SOTA) pooling techniques. In particular, we surpass the test accuracy of existing pooling techniques on different variants of MobileNet-V2 on ImageNet by an average of 1.2%. With the aggressive down-sampling of the activation maps in the initial layers (providing up to 22x reduction in memory consumption), our approach achieves 1.43% higher test accuracy compared to SOTA techniques with iso-memory footprints. This enables the deployment of our models in memory-constrained devices, such as micro-controllers (without losing significant accuracy), because the initial activation maps consume a significant amount of on-chip memory for high-resolution images required for complex vision tasks. Our proposed pooling method also leverages the idea of channel pruning to further reduce memory footprints.
Formaldehyde (HCHO) is one one of the most important trace gas in the atmosphere, as it is a pollutant causing respiratory and other diseases. It is also a precursor of tropospheric ozone which damages crops and deteriorates human health. Study of HCHO chemistry and long-term monitoring using satellite data is important from the perspective of human health, food security and air pollution. Dynamic atmospheric chemistry models struggle to simulate atmospheric formaldehyde and often overestimate by up to two times relative to satellite observations and reanalysis. Spatial distribution of modelled HCHO also fail to match satellite observations. Here, we present deep learning approach using a simple super-resolution based convolutional neural network towards simulating fast and reliable atmospheric HCHO. Our approach is an indirect method of HCHO estimation without the need to chemical equations. We find that deep learning outperforms dynamical model simulations which involves complicated atmospheric chemistry representation. Causality establishing the nonlinear relationships of different variables to target formaldehyde is established in our approach by using a variety of precursors from meteorology and chemical reanalysis to target OMI AURA satellite based HCHO predictions. We choose South Asia for testing our implementation as it doesnt have in situ measurements of formaldehyde and there is a need for improved quality data over the region. Moreover, there are spatial and temporal data gaps in the satellite product which can be removed by trustworthy modelling of atmospheric formaldehyde. This study is a novel attempt using computer vision for trustworthy modelling of formaldehyde from remote sensing can lead to cascading societal benefits.
In recent years, deep learning has been at the center of analytics due to its impressive empirical success in analyzing complex data objects. Despite this success, most of the existing tools behave like black-box machines, thus the increasing interest in interpretable, reliable, and robust deep learning models applicable to a broad class of applications. Feature-selected deep learning has emerged as a promising tool in this realm. However, the recent developments do not accommodate ultra-high dimensional and highly correlated features, in addition to the high noise level. In this article, we propose a novel screening and cleaning method with the aid of deep learning for a data-adaptive multi-resolutional discovery of highly correlated predictors with a controlled error rate. Extensive empirical evaluations over a wide range of simulated scenarios and several real datasets demonstrate the effectiveness of the proposed method in achieving high power while keeping the false discovery rate at a minimum.
Reliable point cloud data is essential for perception tasks \textit{e.g.} in robotics and autonomous driving applications. Adverse weather causes a specific type of noise to light detection and ranging (LiDAR) sensor data, which degrades the quality of the point clouds significantly. To address this issue, this letter presents a novel point cloud adverse weather denoising deep learning algorithm (4DenoiseNet). Our algorithm takes advantage of the time dimension unlike deep learning adverse weather denoising methods in the literature. It performs about 10\% better in terms of intersection over union metric compared to the previous work and is more computationally efficient. These results are achieved on our novel SnowyKITTI dataset, which has over 40000 adverse weather annotated point clouds. Moreover, strong qualitative results on the Canadian Adverse Driving Conditions dataset indicate good generalizability to domain shifts and to different sensor intrinsics.
The brains of all bilaterally symmetric animals on Earth are divided into left and right hemispheres. The anatomy and functionality of the hemispheres have a large degree of overlap, but there are asymmetries, and they specialise in possesses different attributes. Other authors have used computational models to mimic hemispheric asymmetries with a focus on reproducing human data on semantic and visual processing tasks. We took a different approach and aimed to understand how dual hemispheres in a bilateral architecture interact to perform well in a given task. We propose a bilateral artificial neural network that imitates lateralisation observed in nature: that the left hemisphere specialises in local features and the right in global features. We used different training objectives to achieve the desired specialisation and tested it on an image classification task with two different CNN backbones: ResNet and VGG. Our analysis found that the hemispheres represent complementary features that are exploited by a network head that implements a type of weighted attention. The bilateral architecture outperformed a range of baselines of similar representational capacity that do not exploit differential specialisation, with the exception of a conventional ensemble of unilateral networks trained on dual training objectives for local and global features. The results demonstrate the efficacy of bilateralism, contribute to the discussion of bilateralism in biological brains, and the principle may serve as an inductive bias for new AI systems.
Adversarial extraction attacks constitute an insidious threat against Deep Learning (DL) models in-which an adversary aims to steal the architecture, parameters, and hyper-parameters of a targeted DL model. Existing extraction attack literature have observed varying levels of attack success for different DL models and datasets, yet the underlying cause(s) behind their susceptibility often remain unclear, and would help facilitate creating secure DL systems. In this paper we present PINCH: an efficient and automated extraction attack framework capable of designing, deploying, and analyzing extraction attack scenarios across heterogeneous hardware platforms. Using PINCH, we perform extensive experimental evaluation of extraction attacks against 21 model architectures to explore new extraction attack scenarios and further attack staging. Our findings show (1) key extraction characteristics whereby particular model configurations exhibit strong resilience against specific attacks, (2) even partial extraction success enables further staging for other adversarial attacks, and (3) equivalent stolen models uncover differences in expressive power, yet exhibit similar captured knowledge.
Deep learning training is an expensive process that extensively uses GPUs, but not all model training saturates modern powerful GPUs. Multi-Instance GPU (MIG) is a new technology introduced by NVIDIA that can partition a GPU to better-fit workloads that do not require all the memory and compute resources of a full GPU. In this paper, we examine the performance of a MIG-enabled A100 GPU under deep learning workloads containing various sizes and combinations of models. We contrast the benefits of MIG to older workload collocation methods on GPUs: na\"ively submitting multiple processes on the same GPU and utilizing Multi-Process Service (MPS). Our results demonstrate that collocating multiple model training runs may yield significant benefits. In certain cases, it can lead up to four times training throughput despite increased epoch time. On the other hand, the aggregate memory footprint and compute needs of the models trained in parallel must fit the available memory and compute resources of the GPU. MIG can be beneficial thanks to its interference-free partitioning, especially when the sizes of the models align with the MIG partitioning options. MIG's rigid partitioning, however, may create sub-optimal GPU utilization for more dynamic mixed workloads. In general, we recommend MPS as the best performing and most flexible form of collocation for model training for a single user submitting training jobs.
We develop a new formulation of deep learning based on the Mori-Zwanzig (MZ) formalism of irreversible statistical mechanics. The new formulation is built upon the well-known duality between deep neural networks and discrete dynamical systems, and it allows us to directly propagate quantities of interest (conditional expectations and probability density functions) forward and backward through the network by means of exact linear operator equations. Such new equations can be used as a starting point to develop new effective parameterizations of deep neural networks, and provide a new framework to study deep-learning via operator theoretic methods. The proposed MZ formulation of deep learning naturally introduces a new concept, i.e., the memory of the neural network, which plays a fundamental role in low-dimensional modeling and parameterization. By using the theory of contraction mappings, we develop sufficient conditions for the memory of the neural network to decay with the number of layers. This allows us to rigorously transform deep networks into shallow ones, e.g., by reducing the number of neurons per layer (using projection operators), or by reducing the total number of layers (using the decay property of the memory operator).
Activation Functions introduce non-linearity in the deep neural networks. This nonlinearity helps the neural networks learn faster and efficiently from the dataset. In deep learning, many activation functions are developed and used based on the type of problem statement. ReLU's variants, SWISH, and MISH are goto activation functions. MISH function is considered having similar or even better performance than SWISH, and much better than ReLU. In this paper, we propose an activation function named APTx which behaves similar to MISH, but requires lesser mathematical operations to compute. The lesser computational requirements of APTx does speed up the model training, and thus also reduces the hardware requirement for the deep learning model. Source code: https://github.com/mr-ravin/aptx_activation
Recently, there has been growing attention on an end-to-end deep learning-based stitching model. However, the most challenging point in deep learning-based stitching is to obtain pairs of input images with a narrow field of view and ground truth images with a wide field of view captured from real-world scenes. To overcome this difficulty, we develop a weakly-supervised learning mechanism to train the stitching model without requiring genuine ground truth images. In addition, we propose a stitching model that takes multiple real-world fisheye images as inputs and creates a 360 output image in an equirectangular projection format. In particular, our model consists of color consistency corrections, warping, and blending, and is trained by perceptual and SSIM losses. The effectiveness of the proposed algorithm is verified on two real-world stitching datasets.
Hazard and operability analysis (HAZOP) is the paradigm of industrial safety that can reveal the hazards of process from its node deviations, consequences, causes, measures and suggestions, and such hazards can be considered as hazard events (HaE). The classification research on HaE has much irreplaceable pragmatic values. In this paper, we present a novel deep learning model termed DLF through multifractal to explore HaE classification where the motivation is that HaE can be naturally regarded as a kind of time series. Specifically, first HaE is vectorized to get HaE time series by employing BERT. Then, a new multifractal analysis method termed HmF-DFA is proposed to win HaE fractal series by analyzing HaE time series. Finally, a new hierarchical gating neural network (HGNN) is designed to process HaE fractal series to accomplish the classification of HaE from three aspects: severity, possibility and risk. We take HAZOP reports of 18 processes as cases, and launch the experiments on this basis. Results demonstrate that compared with other classifiers, DLF classifier performs better under metrics of precision, recall and F1-score, especially for the severity aspect. Also, HmF-DFA and HGNN effectively promote HaE classification. Our HaE classification system can serve application incentives to experts, engineers, employees, and other enterprises. We hope our research can contribute added support to the daily practice in industrial safety.
FP8 is a natural progression for accelerating deep learning training inference beyond the 16-bit formats common in modern processors. In this paper we propose an 8-bit floating point (FP8) binary interchange format consisting of two encodings - E4M3 (4-bit exponent and 3-bit mantissa) and E5M2 (5-bit exponent and 2-bit mantissa). While E5M2 follows IEEE 754 conventions for representatio of special values, E4M3's dynamic range is extended by not representing infinities and having only one mantissa bit-pattern for NaNs. We demonstrate the efficacy of the FP8 format on a variety of image and language tasks, effectively matching the result quality achieved by 16-bit training sessions. Our study covers the main modern neural network architectures - CNNs, RNNs, and Transformer-based models, leaving all the hyperparameters unchanged from the 16-bit baseline training sessions. Our training experiments include large, up to 175B parameter, language models. We also examine FP8 post-training-quantization of language models trained using 16-bit formats that resisted fixed point int8 quantization.
Recent studies demonstrated that X-ray radiography showed higher accuracy than Polymerase Chain Reaction (PCR) testing for COVID-19 detection. Therefore, applying deep learning models to X-rays and radiography images increases the speed and accuracy of determining COVID-19 cases. However, due to Health Insurance Portability and Accountability (HIPAA) compliance, the hospitals were unwilling to share patient data due to privacy concerns. To maintain privacy, we propose differential private deep learning models to secure the patients' private information. The dataset from the Kaggle website is used to evaluate the designed model for COVID-19 detection. The EfficientNet model version was selected according to its highest test accuracy. The injection of differential privacy constraints into the best-obtained model was made to evaluate performance. The accuracy is noted by varying the trainable layers, privacy loss, and limiting information from each sample. We obtained 84\% accuracy with a privacy loss of 10 during the fine-tuning process.
Monkeypox virus is emerging slowly with the decline of COVID-19 virus infections around the world. People are afraid of it, thinking that it would appear as a pandemic like COVID-19. As such, it is crucial to detect them earlier before widespread community transmission. AI-based detection could help identify them at the early stage. In this paper, we aim to compare 13 different pre-trained deep learning (DL) models for the Monkeypox virus detection. For this, we initially fine-tune them with the addition of universal custom layers for all of them and analyse the results using four well-established measures: Precision, Recall, F1-score, and Accuracy. After the identification of the best-performing DL models, we ensemble them to improve the overall performance using a majority voting over the probabilistic outputs obtained from them. We perform our experiments on a publicly available dataset, which results in average Precision, Recall, F1-score, and Accuracy of 85.44\%, 85.47\%, 85.40\%, and 87.13\%, respectively with the help of our proposed ensemble approach. These encouraging results, which outperform the state-of-the-art methods, suggest that the proposed approach is applicable to health practitioners for mass screening.
Operational flare forecasting aims at providing predictions that can be used to make decisions, typically at a daily scale, about the space weather impacts of flare occurrence. This study shows that video-based deep learning can be used for operational purposes when the training and validation sets used for the network optimization are generated while accounting for the periodicity of the solar cycle. Specifically, the paper describes an algorithm that can be applied to build up sets of active regions that are balanced according to the flare class rates associated to a specific cycle phase. These sets are used to train and validate a Long-term Recurrent Convolutional Network made of a combination of a convolutional neural network and a Long-Short Memory network. The reliability of this approach is assessed in the case of two prediction windows containing the solar storm of March 2015 and September 2017, respectively.
Deep Learning has received increased attention due to its unbeatable success in many fields, such as computer vision, natural language processing, recommendation systems, and most recently in simulating multiphysics problems and predicting nonlinear dynamical systems. However, modeling and forecasting the dynamics of chaotic systems remains an open research problem since training deep learning models requires big data, which is not always available in many cases. Such deep learners can be trained from additional information obtained from simulated results and by enforcing the physical laws of the chaotic systems. This paper considers extreme events and their dynamics and proposes elegant models based on deep neural networks, called knowledge-based deep learning (KDL). Our proposed KDL can learn the complex patterns governing chaotic systems by jointly training on real and simulated data directly from the dynamics and their differential equations. This knowledge is transferred to model and forecast real-world chaotic events exhibiting extreme behavior. We validate the efficiency of our model by assessing it on three real-world benchmark datasets: El Nino sea surface temperature, San Juan Dengue viral infection, and Bj{\o}rn{\o}ya daily precipitation, all governed by extreme events' dynamics. Using prior knowledge of extreme events and physics-based loss functions to lead the neural network learning, we ensure physically consistent, generalizable, and accurate forecasting, even in a small data regime.
Artificial neural networks took a lot of inspiration from their biological counterparts in becoming our best machine perceptual systems. This work summarizes some of that history and incorporates modern theoretical neuroscience into experiments with artificial neural networks from the field of deep learning. Specifically, iterative magnitude pruning is used to train sparsely connected networks with 33x fewer weights without loss in performance. These are used to test and ultimately reject the hypothesis that weight sparsity alone improves image noise robustness. Recent work mitigated catastrophic forgetting using weight sparsity, activation sparsity, and active dendrite modeling. This paper replicates those findings, and extends the method to train convolutional neural networks on a more challenging continual learning task. The code has been made publicly available.
Fast and precise beam alignment is crucial to support high-quality data transmission in millimeter wave (mmWave) communication systems. In this work, we propose a novel deep learning based hierarchical beam alignment method that learns two tiers of probing codebooks (PCs) and uses their measurements to predict the optimal beam in a coarse-to-fine searching manner. Specifically, the proposed method first performs coarse channel measurement using the tier-1 PC, then selects a tier-2 PC for fine channel measurement, and finally predicts the optimal beam based on both coarse and fine measurements. The proposed deep neural network (DNN) architecture is trained in two steps. First, the tier-1 PC and the tier-2 PC selector are trained jointly. After that, all the tier-2 PCs together with the optimal beam predictors are trained jointly. The learned hierarchical PCs can capture the features of propagation environment. Numerical results based on realistic ray-tracing datasets demonstrate that the proposed method is superior to the state-of-art beam alignment methods in both alignment accuracy and sweeping overhead.
COVID-19 has affected more than 223 countries worldwide. There is a pressing need for non invasive, low costs and highly scalable solutions to detect COVID-19, especially in low-resource countries where PCR testing is not ubiquitously available. Our aim is to develop a deep learning model identifying COVID-19 using voice data recordings spontaneously provided by the general population (voice recordings and a short questionnaire) via their personal devices. The novelty of this work is in the development of a deep learning model for the identification of COVID-19 patients from voice recordings. Methods: We used the Cambridge University dataset consisting of 893 audio samples, crowd-sourced from 4352 participants that used a COVID-19 Sounds app. Voice features were extracted using a Mel-spectrogram analysis. Based on the voice data, we developed deep learning classification models to detect positive COVID-19 cases. These models included Long-Short Term Memory (LSTM) and Convolutional Neural Network (CNN). We compared their predictive power to baseline classification models, namely Logistic Regression and Support Vector Machine. Results: LSTM based on a Mel-frequency cepstral coefficients (MFCC) features achieved the highest accuracy (89%,) with a sensitivity and specificity of respectively 89% and 89%, The results achieved with the proposed model suggest a significant improvement in the prediction accuracy of COVID-19 diagnosis compared to the results obtained in the state of the art. Conclusion: Deep learning can detect subtle changes in the voice of COVID-19 patients with promising results. As an addition to the current testing techniques this model may aid health professionals in fast diagnosis and tracing of COVID-19 cases using simple voice analysis
Synthetic aperture sonar (SAS) imagery is crucial for several applications, including target recognition and environmental segmentation. Deep learning models have led to much success in SAS analysis; however, the features extracted by these approaches may not be suitable for capturing certain textural information. To address this problem, we present a novel application of histogram layers on SAS imagery. The addition of histogram layer(s) within the deep learning models improved performance by incorporating statistical texture information on both synthetic and real-world datasets.