Abstract: Performance of a machine learning (ML) model in production, is heavily dependent on underlying distribution of data or underlying process generating labels from attributes. Any change in either one or both impacts the ML model performance heavily and inhibits knowledge of true labels. This in turn affects ML model uncertainty. Thus, performance monitoring of ML models in production becomes necessary. Embodiments of the present disclosure provide estimates operating model accuracy at production stage by constructing the correlations between the model accuracy, model uncertainty and deviation of the distributions in absence of ground truth. In the method of present disclosure, the model performance of the machine learning (ML) model deployed in production is estimated in absence of ground truths. Moreover, this can be done without retraining the model, thus saving computational costs and resources. The method of the present disclosure can be used and performed in real time.

Abstract: This disclosure relates generally to recommending an optimal VM instance. The increased use of Deep Learning (DL) models in several domains has resulted in an increased demand for hardware configurations to enable heavy computations and faster performance to support the DL techniques. However, the identification of the optimal hardware configuration for the DL requirement is challenging and requires a considerable amount of time and expertise, considering the highly configurable model configuration of DL techniques.

Abstract: Migrating application from on premise HPC cluster to serverless platform is tedious task and involves significant amount of human efforts as cloud infrastructure needs to be created, data along with libraries and application code need to be copied from on-premise to cloud, and application need to be made compliant for execution on cloud. Present disclosure provides method and system for performing automated migration of high performance computing application to serverless platform. The system first check cloud readiness of application based on operation qualification parameters of application. In case application is found to be cloud ready, the system determines whether application can be executed on serverless platform based on execution time of the application and permissible limits defined for application in service level agreements.

Abstract: This disclosure relates generally relates to method and system to process asynchronous and distributed training tasks. Training a large-scale deep neural network (DNN) model with large-scale training data is time-consuming. The method creates a work queue (Q) with a set of predefined number of tasks comprising a training data. Here, set of central processing units (CPUs) information and a set of graphics processing units (GPUs) information are fetched from the current environment to initiate a parallel process asynchronously on the work queue (Q) to train a set of deep learning models with optimized resources using a data pre-processing technique, to compute a transformed training data and training by using an asynchronous model training technique, the set of deep learning models on each GPU asynchronously with the transformed training data based on a set of asynchronous model parameters.

Abstract: State of art approaches independently use a Pruning-weight Clustering-Quantization (PCQ) or Knowledge Distillation (KD) for model optimization and require critical manual intervention. Embodiments of the present disclosure provide a method and system for the two-step hierarchical model optimization approach for generating optimized model DL model. The method comprises a AutoPCQ technique followed by conditional application of an automated KD (AKD) technique. The AutoPCQ technique formulates a problem of configuration selection of the DL model as an optimization problem by iteratively applying Bayesian optimization and Reinforcement Learning. Further, the AKD technique formulates automated search of a student model as the optimization problem with the DL model representing a teacher model. A search space for the student model is defined by a restricted Neural Network Architecture Search that restricts the search space.

Abstract: This disclosure relates generally to method and system for tuning graphics processing unit (GPU) parameters of a GPU kernel. The disclosure proposes a combination of both heuristic and deterministic techniques for tuning GPU parameters of a GPU kernel to achieve optimal configuration of the GPU parameters. The proposed method and a system for tuning GPU parameters is based on deterministic techniques and heuristic techniques that includes capturing behavior of the GPU application by monitoring several GPU hardware counters that comprise several hardware resources and performance counters. The proposed tuning GPU parameters also implements a set of heuristic techniques to decide course of the tuning for various GPU parameters based on the captured behaviour of the GPU hardware counters.

Abstract: State of the art predictive maintenance systems that generate predictions with respect to maintenance of High Performance Computing (HPC) systems have the disadvantage that they either are reactive, or the predictions are affected due to quality issues associated with the data being collected from the HPC systems. The disclosure herein generally relates to predictive maintenance, and, more particularly, to a method and system for predictive maintenance of High Performance Computing (HPC) systems. The system performs abstraction and cleansing on performance data collected from the HPC systems, and generates a cleansed performance data, on which a Machine Leaning (ML) prediction is applied to generate predictions with respect to maintenance of the HPC systems.

Abstract: This disclosure relates generally to method and system for tuning graphics processing unit (GPU) parameters of a GPU kernel. The disclosure proposes a combination of both heuristic and deterministic techniques for tuning GPU parameters of a GPU kernel to achieve optimal configuration of the GPU parameters. The proposed method and a system for tuning GPU parameters is based on deterministic techniques and heuristic techniques that includes capturing behavior of the GPU application by monitoring several GPU hardware counters that comprise several hardware resources and performance counters. The proposed tuning GPU parameters also implements a set of heuristic techniques to decide course of the tuning for various GPU parameters based on the captured behaviour of the GPU hardware counters.

Abstract: The disclosure generally relates to principal component analysis (PCA) computation and, more particularly, to concurrent PCA computation. In one embodiment, a plurality of concurrent PCA requests are received by a server. An input matrix for each of the concurrent PCA requests is computed using a general purpose-graphical processing unit (GP-GPU) by the server. Further, tridiagnolization on the input matrix is performed on each of the concurrent PCA requests by a general purpose-graphical processing unit (GP-GPU) in the server to generate a tridiagonal matrix for each of the concurrent PCA requests. Furthermore, a plurality of eigen values and corresponding eigen vectors are computed for the tridiagonal matrix of each of the concurrent PCA requests by the server and subsequently back transformation of the eigen values and the eigen vectors is performed by the server for each of the concurrent PCA requests to obtain associated principal components.

Abstract: This disclosure relates generally to correlation filters, and more particularly to designing of correlation filter. In one embodiment, a system for designing a correlation filter in a multi-processor system includes a multi-core processor coupled to a first memory and one or more co-processors coupled to one or more respective second memories. The multi-core processor partitions each of a plurality of frames associated with media content into a plurality of pixel-columns, and systematically stores said pixel-columns width-wise in a plurality of temporary matrices by a plurality of threads of the multi-core processor. The plurality of temporary matrices are transferred by the multi-core processor to one or more respective second memories in a plurality of streams simultaneously in an asynchronous mode. A plurality of filter harmonics of the correlation filter are computed by performing compute operations involving at least the plurality of temporary matrices, to obtain the correlation filter.

Abstract: The disclosure generally relates to principal component analysis (PCA) computation and, more particularly, to concurrent PCA computation. In one embodiment, a plurality of concurrent PCA requests are received by a server. An input matrix for each of the concurrent PCA requests is computed using a general purpose-graphical processing unit (GP-GPU) by the server. Further, tridiagnolization on the input matrix is performed on each of the concurrent PCA requests by a general purpose-graphical processing unit (GP-GPU) in the server to generate a tridiagonal matrix for each of the concurrent PCA requests. Furthermore, a plurality of eigen values and corresponding eigen vectors are computed for the tridiagonal matrix of each of the concurrent PCA requests by the server and subsequently back transformation of the eigen values and the eigen vectors is performed by the server for each of the concurrent PCA requests to obtain associated principal components.

Abstract: This disclosure relates generally to correlation filters, and more particularly to designing of correlation filter. In one embodiment, a system for designing a correlation filter in a multi-processor system includes a multi-core processor coupled to a first memory and one or more co-processors coupled to one or more respective second memories. The multi-core processor partitions each of a plurality of frames associated with media content into a plurality of pixel-columns, and systematically stores said pixel-columns width-wise in a plurality of temporary matrices by a plurality of threads of the multi-core processor. The plurality of temporary matrices are transferred by the multi-core processor to one or more respective second memories in a plurality of streams simultaneously in an asynchronous mode. A plurality of filter harmonics of the correlation filter are computed by performing compute operations involving at least the plurality of temporary matrices, to obtain the correlation filter.