KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Settings

Author: search.filters.author.Sangpetch, OrathaiSubject: search.filters.subject.Deep Learning

Search Results

Now showing 1 - 2 of 2
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    GPU Performance Tuning and Power Efficiency on the DGX A100 Cluster
    (2022-01-01)
    Udomchoksakul, Khanin
    ;
    Sangpetch, Orathai
    ;
    Sangpetch, Akkarit
    The complexity of current Deep learning has been growing rapidly nowadays. Such advancement allows various organizations such as private sectors and government to leverage intelligent systems on their use cases. High Performance Computing (HPC) infrastructure nowadays has pivoted to GPU-oriented systems, enabling developers and researchers to train complex models with large datasets unlike conventional clusters equipped only with CPU cores. However, focus on power efficiency on the HPC system has not been prevalent especially on the new system such as DGX A100 that does not have datapoints on how GPUs consumed power. Even though such HPC cluster can be powerful, always allowing it to run at the maximum capacity results to financial cost to the HPC provider at the end. Therefore, for any organization providing the system, it is crucial for them to balance the cluster capabilities while maintaining overall power consumption which can potentially be costly in the long term. This paper reveals A100 GPU metrics that are relevant to Power usage and explains GPU profiling applied to Deep learning workload on the cluster, saving up to 32% of the power usage while compromising only 11.5% of training time compared to a default profile. Then, the paper investigates literature review that could be learned further adopted to the current system at CMKL university as the next milestone.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    DASSL: Dynamic, AI-assisted, Scalable System for Labelling Used Bottle Images
    (2020-09-21)
    Daengphruan, Parnmet
    ;
    Sangpetch, Orathai
    ;
    Sangpetch, Akkarit
    To ensure sustainable consumption and production, one way is to reduce waste generation by increasing the reuse rate. We have been working with the bottle classification facility to enhance the efficiency and productivity. Many used bottles come in with unimaginable ways of dirty, defective conditions. To manage the sheer volume of used bottles, we create an AI-enabled, bottle classification system. However, it requires many labelled images for training to improve accuracy. Unfortunately, the traditional approach, having human label individual images, is very time consuming. Even worse, it is not effective for our dataset because conditions of used bottles are not well defined and studied. From our experiments, the human experts cannot agree on the same labelling for similar bottle conditions, especially when impurities or defects are not separable objects. For 42%-99% of images in certain subcategories, human experts assign different labels to bottles with similar conditions. With huge inconsistency in data labelling, it deteriorates the accuracy of our classification models. To alleviate this problem, we propose a Dynamic, AI-assisted, Scalable System for Labelling used bottle images, called DASSL. DASSL employs multiple algorithms to extract and/or quantize different features of used bottle images, and cluster the images into groups with the supervision of human. With DASSL, we can achieve labelling consistency and improve scalability by reducing the data labelling time by at least 10x. To enhance agility, we can dynamically adjust DASSL to adapt to changes of cleaning machines' capabilities or bottle demand.