Fayaz Moqueem Mohammed

Meta AI Video Similarity Challenge META AI and Boston University(Finished 3rd) Apr 2023 Participating in the competitive and invigorating Meta AI Challenge was a journey that pushed my boundaries and cultivated an environment for exceptional innovation. As a participant, my task was to create precise descriptors or embeddings for a vast repository of videos, with the ultimate goal of identifying manipulated portions within the query video set.Having an arsenal of 40,000 reference videos and 8,000 query videos at my disposal, I was tasked with crafting precise 32-bit… Show more Participating in the competitive and invigorating Meta AI Challenge was a journey that pushed my boundaries and cultivated an environment for exceptional innovation. As a participant, my task was to create precise descriptors or embeddings for a vast repository of videos, with the ultimate goal of identifying manipulated portions within the query video set.Having an arsenal of 40,000 reference videos and 8,000 query videos at my disposal, I was tasked with crafting precise 32-bit floating-point vectors of up to 512 dimensions, known as descriptors. The complexity of the task was compounded by the stipulation that multiple descriptors could be used per video, up to a limit of one descriptor per second of video.In executing this task, I had the opportunity to flex my skills in artificial intelligence, machine learning, computer vision, and data analysis. My approach was to employ the power of cutting-edge AI and ML techniques, especially in the generation and optimization of the descriptors. The project demanded a meticulous attention to detail, critical thinking, and innovative problem-solving techniques to successfully navigate the intricate landscape of the challenge.Ultimately, my solution, through the judicious application of AI and ML algorithms, generated descriptors that showcased high degrees of accuracy. The result of this work was an impressive fifth-place finish in the competition, tantalizingly close to the top. The evaluation script demonstrated the effectiveness of my approach by computing the inner-product distance between descriptor vectors in the query and reference sets. These calculations were then used to provide a confidence score that dictated the micro-average precision.This near-win underscores the potency of my solution, the culmination of the arduous but fulfilling journey that the Meta AI Challenge presented. Show less