Corun: Concurrent Inference and Continuous Training at the Edge for Cost-Efficient AI-Based Mobile Image Sensing

Medical image segmentation involves partitioning medical images into meaningful regions, with a focus on identifying anatomical structures and lesions. It has broad applications in healthcare, and deep learning methods have enabled significant advancements in automating this process. Recently, the introduction of the Segmentation Anything Model (SAM), the first foundation model for segmentation task, has prompted researchers to adapt it for the medical domain to improve performance across various tasks. However, SAM’s large model size and high GPU requirements hinder its scalability and development in the medical domain. To address these challenges, research has increasingly focused on lightweight adaptations of SAM to reduce its parameter count, enabling training with limited GPU resources while maintaining competitive segmentation performance. In this work, we propose MCP-MedSAM, a powerful and lightweight medical SAM model designed to be trainable on a single A100 GPU with 40GB of memory within one day while delivering superior segmentation performance. Recognizing the significant internal differences between modalities and the need for direct segmentation target information within bounding boxes, we introduce two kinds of prompts: the modality prompt and the content prompt. After passing through the prompt encoder, their embedding representations can further improve the segmentation performance by incorporating more relevant information without adding significant training overhead. Additionally, we adopt an effective modality-based data sampling strategy to address data imbalance between modalities, ensuring more balanced performance across all modalities. Our method was trained and evaluated using a large-scale challenge dataset, compared to top-ranking methods on the challenge leaderboard, MCP-MedSAM achieved superior performance while requiring only one day of training on a single GPU. The code is publicly available at <a href='https://github.com/dong845/MCP-MedSAM'>https://github.com/dong845/MCP-MedSAM</a>

Facial age recognition has been widely used in real-world applications. Most of current facial age recognition methods use deep learning to extract facial features to identify age. However, due to the high dimension features of faces, deep learning methods might extract a lot of redundant features, which is not beneficial for facial age recognition. To improve facial age recognition effectively, this paper proposed the deep manifold learning (DML), a combination of deep learning and manifold learning. In DML, deep learning was used to extract high-dimensional facial features, and manifold learning selected age-related features from these high-dimensional facial features for facial age recognition. Finally, we validated the DML on Multivariate Observations of Reactions and Physical Health (MORPH) and Face and Gesture Recognition Network (FG-NET) datasets. The results indicated that the mean absolute error (MAE) of MORPH is 1.60 and that of FG-NET is 2.48. Moreover, compared with the state of the art facial age recognition methods, the accuracy of DML has been greatly improved.

To achieve high accuracy when performing deep learning, it is necessary to use a large-scale training model. However, due to the limitations of GPU memory, it is difficult to train large-scale training models within a single GPU. NVIDIA introduced a technology called CUDA Unified Memory with CUDA 6 to overcome the limitations of GPU memory by virtually combining GPU memory and CPU memory. In addition, in CUDA 8, memory advise options are introduced to efficiently utilize CUDA Unified Memory. In this work, we propose a newly optimized scheme based on CUDA Unified Memory to efficiently use GPU memory by applying different memory advise to each data type according to access patterns in deep learning training. We apply CUDA Unified Memory technology to PyTorch to see the performance of large-scale learning models through the expanded GPU memory. We conduct comprehensive experiments on how to efficiently utilize Unified Memory by applying memory advises when performing deep learning. As a result, when the data used for deep learning are divided into three types and a memory advise is applied to the data according to the access pattern, the deep learning execution time is reduced by 9.4% compared to the default Unified Memory.

Optimizing makespan and resource utilization for multi-DNN training in GPU cluster

Reconstructing real-world 3D objects has numerous applications in computer vision, such as virtual reality, video games, and animations. Ideally, 3D reconstruction methods should generate high-fidelity results with 3D consistency in real-time. Traditional methods match pixels between images using photo-consistency constraints or learned features, while differentiable rendering methods like Neural Radiance Fields (NeRF) use differentiable volume rendering or surface-based representation to generate high-fidelity scenes. However, these methods require excessive runtime for rendering, making them impractical for daily applications. To address these challenges, we present EvaSurf, an Efficient View-Aware implicit textured Surface reconstruction method on mobile devices. In our method, we first employ an efficient surface-based model with a multi-view supervision module to ensure accurate mesh reconstruction. To enable high-fidelity rendering, we learn an implicit texture embedded with view-aware encoding to capture view-dependent information. Furthermore, with the explicit geometry and the implicit texture, we can employ a lightweight neural shader to reduce the expense of computation and further support real-time rendering on common mobile devices. Extensive experiments demonstrate that our method can reconstruct high-quality appearance and accurate mesh on both synthetic and real-world datasets. Moreover, our method can be trained in just 1-2 hours using a single GPU and run on mobile devices at over 40 FPS (Frames Per Second), with a final package required for rendering taking up only 40-50 MB.

Application of Face Recognition Method Under Deep Learning Algorithm in Embedded Systems

Every human has emotions in life at all phases of life’s moment or instance. These emotions keep changing ever so often and person to person because of the biological reflexes of human brain produced by the nervous system due to the neurophysiological changes which are closely associated with behavior, mood swings; thoughts, etc. are tending to be reflected as an expression on the human face. The application of study of these emotions of humans using facial expressions became very prominent in the fields of biomedical engineering, neuroscience, product screening, marketing, and advertisement, etc. In early works of emotion recognition is for the most part useful for analyzing product reviews or some methodologies by monitoring videos and images in real time. Imparting artificial intelligence to the future of the world where every industry thrives well. In recent years, progressive growth in artificial intelligence is seen especially in the field of machine learning with the predominant subset called deep learning. Mainly, deep learning algorithms made an overriding burgeon in image classification thereby avoiding the complex process in facial recognition. The framework proposed in this paper describes conceptual and theoretical knowledge for human facial emotion recognition with a labeled model and applying a Haar cascade classifier using CNN classification—a class in deep neural networks used in its implementation.KeywordsImage PreprocessingFacial Emotion RecognitionHaar Cascade ClassifierData AugmentationDeep LearningCNN—Convolutional Neural NetworkAdaboost TrainingPrediction

Recent breakthroughs in deep learning and artificial intelligence technologies have enabled numerous mobile applications. While traditional computation paradigms rely on mobile sensing and cloud computing, deep learning implemented on mobile devices provides several advantages. These advantages include low communication bandwidth, small cloud computing resource cost, quick response time, and improved data privacy. Research and development of deep learning on mobile and embedded devices has recently attracted much attention. This paper provides a timely review of this fast-paced field to give the researcher, engineer, practitioner, and graduate student a quick grasp on the recent advancements of deep learning on mobile devices. In this paper, we discuss hardware architectures for mobile deep learning, including Field Programmable Gate Arrays (FPGA), Application Specific Integrated Circuit (ASIC), and recent mobile Graphic Processing Units (GPUs). We present Size, Weight, Area and Power (SWAP) considerations and their relation to algorithm optimizations, such as quantization, pruning, compression, and approximations that simplify computation while retaining performance accuracy. We cover existing systems and give a state-of-the-industry review of TensorFlow, MXNet, Mobile AI Compute Engine (MACE), and Paddle-mobile deep learning platform. We discuss resources for mobile deep learning practitioners, including tools, libraries, models, and performance benchmarks. We present applications of various mobile sensing modalities to industries, ranging from robotics, healthcare and multimedia, biometrics to autonomous drive and defense. We address the key deep learning challenges to overcome, including low quality data, and small training/adaptation data sets. In addition, the review provides numerous citations and links to existing code bases implementing various technologies. These resources lower the user’s barrier to entry into the field of mobile deep learning.

The exponential growth in the number of complex datasets every year requires\nmore enhancement in machine learning methods to provide robust and accurate\ndata classification. Lately, deep learning approaches have achieved surpassing\nresults in comparison to previous machine learning algorithms. However, finding\nthe suitable structure for these models has been a challenge for researchers.\nThis paper introduces Random Multimodel Deep Learning (RMDL): a new ensemble,\ndeep learning approach for classification. RMDL solves the problem of finding\nthe best deep learning structure and architecture while simultaneously\nimproving robustness and accuracy through ensembles of deep learning\narchitectures. In short, RMDL trains multiple randomly generated models of Deep\nNeural Network (DNN), Convolutional Neural Network (CNN) and Recurrent Neural\nNetwork (RNN) in parallel and combines their results to produce better result\nof any of those models individually. In this paper, we describe RMDL model and\ncompare the results for image and text classification as well as face\nrecognition. We used MNIST and CIFAR-10 datasets as ground truth datasets for\nimage classification and WOS, Reuters, IMDB, and 20newsgroup datasets for text\nclassification. Lastly, we used ORL dataset to compare the model performance on\nface recognition task.\n

Hyperspectral images (HSI) classification has been a research hotspot in the remote sensing field. Deep learning methods such as Cube-CNN have been applied to address the HSI classification problem. However, mainstream frameworks exist a performance gap to train Cube-CNN, since they are not designed for processing high dimensional data like HSI. To close this gap, we propose a Multi-GPU Training Framework (MTFC) for Cube-CNN-based HSI classification. We first design a Parallel Neighbor Pixel Extraction (PNPE) algorithm for efficiently generating 3-dimensional cube samples from raw data. Then, to fully exploit massive GPU parallelism and realize unique characteristics of HSI and Cube-CNN, we employ optimizations in MTFC such as task division, fine-grained mapping between tasks and GPU thread blocks, shared memory usage reduction, etc. Finally, to further improve training speed, we take advantage of CUDA streams and multiple GPUs to train a mini-batches of data samples simultaneously. An extensive set of experiments highlights that MTFC constantly outperforms the two baselines Caffe and Theano for all measured metrics across all system configurations, while offering the same level of classification accuracy. The speedup is up to <inline-formula><tex-math notation="LaTeX">$3.6x$</tex-math></inline-formula> when using a single GPU and MTFC can achieve a rough linear scaling on multiple GPUs.

Although the existing traditional image classification methods have been widely applied in practical problems, there are some problems in the application process, such as unsatisfactory effects, low classification accuracy, and weak adaptive ability. This method separates image feature extraction and classification into two steps for classification operation. The deep learning model has a powerful learning ability, which integrates the feature extraction and classification process into a whole to complete the image classification test, which can effectively improve the image classification accuracy. However, this method has the following problems in the application process: first, it is impossible to effectively approximate the complex functions in the deep learning model. Second, the deep learning model comes with a low classifier with low accuracy. So, this paper introduces the idea of sparse representation into the architecture of the deep learning network and comprehensively utilizes the sparse representation of well multidimensional data linear decomposition ability and the deep structural advantages of multilayer nonlinear mapping to complete the complex function approximation in the deep learning model. And a sparse representation classification method based on the optimized kernel function is proposed to replace the classifier in the deep learning model, thereby improving the image classification effect. Therefore, this paper proposes an image classification algorithm based on the stacked sparse coding depth learning model-optimized kernel function nonnegative sparse representation. The experimental results show that the proposed method not only has a higher average accuracy than other mainstream methods but also can be good adapted to various image databases. Compared with other deep learning methods, it can better solve the problems of complex function approximation and poor classifier effect, thus further improving image classification accuracy.

Numerical simulation of underground CO2 storage requires accurate, fast and field-scale calculations of saturation and reservoir pressures, yet high-fidelity physics-based simulators are computationally expensive to do a rapid analysis. Recent advances in deep learning, especially Graph Neural Networks (GNNs), offer promising alternatives by approximating simulation outputs while dramatically reducing computational costs by minimizing the total reliance on traditional simulators. However, training GNN models on large-scale, spatiotemporal datasets remains a significant bottleneck when constrained to a single GPU. To address the computational bottleneck of training GNN on large-scale and high-resolution datasets, we propose to use multi-GPU scalable training of GNNs. The U-Net Enhanced Graph Neural Network (U-GCN) specifically was adopted to predict the spatiotemporal evolution of gas saturation and pressure buildup in CO2 storage in saline aquifers. The UGCN model was trained across multiple GPUs, allowing for concurrent gradient computations and parameter updates on separate GPU processes. The model operates natively on regular reservoir meshes to learn spatiotemporal evolution of pressure and saturation from thousands of simulation cases, and the training pipeline couples process-per-GPU data parallelism with a node-local caching strategy that deamplifies the source data file (in Hierarchical Data Format HDF5 format) Input-Output (I/O) by performing a single deserialization per node and broadcasting preprocessed graphs to sibling ranks. This eliminates multi-node startup contention and renders runs compute bound. Scaling experiments from 1 to 16 GPUs demonstrate substantial wall-clock reductions while preserving validation behavior. With per-GPU batch size 100, total training time decreased from 955.46 s (1 GPU) to 145.17 s (16 GPUs), a 6.58× speedup (≈41% parallel efficiency). With per-GPU batch size 8, time decreased from 5519.00 s to 1047.17 s, a 5.27× speedup (≈33% efficiency). The speedups are reported per epoch. Cross-node overheads relative to ideal doubling were ∼11–35%, more pronounced at small per-GPU batches where collective latency dominates. Training and validation losses rose modestly as GPU count increased under weak scaling—consistent with the larger global batch and fewer optimizer updates per epoch—yet validation tracked training across all settings, indicating stable generalization and no data leakage. These effects are mitigated by standard large-batch remedies (e.g., linear learning-rate scaling with warmup, longer schedules, optional gradient accumulation). By integrating domain-aware batching, multi-GPU data parallelism, mixed precision, and node-local caching with a GNN tailored to structured meshes, the proposed approach delivers a reproducible, high-throughput storage surrogate training regime that scales to multi-node clusters without compromising model validity. The resulting surrogates enable rapid scenario screening and outer-loop workflows (optimization and UQ) at orders-of-magnitude lower training cost compared to conventional GNN training on single a GPU baseline.

As the scientific exploitation of the Square Kilometre Array (SKA) approaches, there is a need for new advanced data analysis and visualization tools capable of processing large high-dimensional datasets. In this study, we aim to generalize the YOLO-CIANNA deep learning source detection and characterization method for 3D hyperspectral HI emission cubes. We present the adaptations we made to the regression-based detection formalism and the construction of an end-to-end 3D convolutional neural network (CNN) backbone. We then describe a processing pipeline for applying the method to simulated 3D HI cubes from the SKA Observatory Science Data Challenge 2 (SDC2) dataset. The YOLO-CIANNA method was originally developed and used by the MINERVA team that won the official SDC2 competition. Despite the public release of the full SDC2 dataset, no published result has yet surpassed MINERVA's top score. In this paper, we present an updated version of our method that improves our challenge score by 9.5%. The resulting catalog exhibits a high detection purity of 92.3%, best-in-class characterization accuracy, and contains 45% more confirmed sources than concurrent classical detection tools. The method is also computationally efficient, processing the full ∼1TB SDC2 data cube in 30 min on a single GPU. These state-of-the-art results highlight the effectiveness of 3D CNN-based detectors for processing large hyperspectral data cubes and represent a promising step toward applying YOLO-CIANNA to observational data from SKA and its precursors.

In this pilot study, we investigate the use of a deep learning (DL) model to temporally evolve the dynamics of gas accreting on to a black hole in the form of a radiatively inefficient accretion flow (RIAF). We have trained a convolutional neural network (CNN) on a data set that consists of numerical solutions of the hydrodynamical equations for a range of initial conditions. We find that deep neural networks trained on one simulation seem to learn reasonably well the spatiotemporal distribution of densities and mass continuity of a black hole accretion flow over a duration of 8 × 104GM/c3, comparable to the viscous time-scale at r = 400GM/c2; after that duration, the model drifts from the ground truth suffering from excessive artificial mass injection. Models trained on simulations with different initial conditions show some promise of generalizing to configurations not present in the training set, but also suffer from mass continuity issues. We discuss the caveats behind this method and the potential benefits that DL models offer. For instance, once trained the model evolves an RIAF on a single GPU four orders of magnitude faster than usual fluid dynamics integrators running in parallel on 200 CPU cores. We speculate that a data-driven machine learning approach should be very promising for accelerating simulations of accreting black holes.

Conventional protein-protein docking algorithms usually rely on heavy candidate sampling and reranking, but these steps are time-consuming and hinder applications that require high-throughput complex structure prediction, for example, structure-based virtual screening. Existing deep learning methods for protein-protein docking, despite being much faster, suffer from low docking success rates. In addition, they simplify the problem to assume no conformational changes within any protein upon binding (rigid docking). This assumption precludes applications when binding-induced conformational changes play a role, such as allosteric inhibition or docking from uncertain unbound model structures. To address these limitations, we present GeoDock, a multitrack iterative transformer network to predict a docked structure from separate docking partners. Unlike deep learning models for protein structure prediction that input multiple sequence alignments, GeoDock inputs just the sequences and structures of the docking partners, which suits the tasks when the individual structures are given. GeoDock is flexible at the protein residue level, allowing the prediction of conformational changes upon binding. On the Database of Interacting Protein Structures (DIPS) test set, GeoDock achieves a 43% top-1 success rate, outperforming all other tested methods. However, in the standard DIPS train/test splits, we discovered contamination of close homologs in the training set. After decontaminating the training set, the success rate is 31%. On the DB5.5 test set and a benchmark dataset of antibody-antigen complexes, GeoDock outperforms the deep learning models trained using the same dataset but falls behind most of the conventional methods and AlphaFold-Multimer. GeoDock attains an average inference speed of under 1 s on a single GPU, enabling its application in large-scale structure screening. Although binding-induced conformational changes are still a challenge owing to limited training and evaluation data, our architecture sets up the foundation to capture this backbone flexibility. Code and a demonstration Jupyter notebook are available at https://github.com/Graylab/GeoDock.