The Machine Learning Developers Summit (MLDS) 2024 in Bengaluru witnessed an enlightening presentation by Ahmad Shaik, a distinguished Senior Data Scientist at KamerAI, who has been at the forefront of the industry’s evolution in computer vision and deep learning. With a Ph.D. in image processing and information security from VIT University, Ahmad’s expertise extends far beyond the technical realm, as he is also a passionate advocate for agriculture and aquaculture. In his talk, Ahmad set the stage by contextualizing the conference’s theme – generative AI and LLMs. He humorously acknowledged the prevailing trend of La La Land and the pursuit of happiness in scripted dialogues but steered the conversation towards a more Jason Bourne-like narrative. The crux of his presentation lay in the transformative journey of computer vision, particularly in the realm of video analytics.
Ahmad began by painting a vivid picture of the colossal data generated daily. He highlighted the fact that Google alone processes a staggering 8.5 billion search requests each day. However, he pivoted the narrative to the world of video analytics, where live video streams, when processed, generate an astronomical amount of data. Illustrating this, he emphasized that a single IP camera running at 20 frames per second would produce a mind-boggling 1.7 million frames in a day. Extrapolating this to 4,000 cameras, the result is an astonishing 8 billion images for inference. This underscored the magnitude of the challenges faced in the field of video analytics.
Transitioning into the core of his presentation, Ahmad provided an insightful overview of the current state of AI, drawing from his experiences both in academia and industry. He delved into the PwC AI index, which categorizes AI capabilities into automated intelligence, assisted intelligence, augmented intelligence, and autonomous intelligence. He emphasized the pivotal role of AI in various sectors and its evolution from descriptive or decision-based AI to the more advanced generative AI.
Ahmad shared intriguing statistics on the current status of AI, citing the Stanford AI index report of 2023. With a staggering 46,000 papers published, the field of AI has undergone a significant transformation. What once was purely theoretical has now found widespread application in diverse sectors. Ahmad particularly highlighted the increasing integration of AI in computer vision projects, constituting 34% of all AI projects in various industries.
While acknowledging the positive trajectory of AI, Ahmad did not shy away from addressing the challenges. Proving business value, choosing the right technologies, and managing related risks emerged as critical obstacles. He stressed the importance of obtaining the necessary data for model training, a challenge often faced in real-world applications.
Transitioning seamlessly into the main theme of his presentation, Ahmad explored the transformation of computer vision and its profound impact on video analytics. He shared his academic background in image processing and how AI has revolutionized the field. Ahmad’s overview encompassed the shift from traditional, rule-based video analytics pipelines to more dynamic, AI-driven approaches.
Ahmad elucidated the shift from reactive measures to proactive measures enabled by AI in video analytics. He emphasized the significance of AI in predicting and preventing incidents, particularly in manufacturing, logistics, and road safety. The ability to define algorithms that identify potential issues in advance represents a paradigm shift in the industry’s approach to safety and productivity.
Highlighting the evolving nature of AI models, Ahmad touched upon the shift from model-centric to data-centric development. He underscored the significance of developing highly optimized models that can operate on edge devices. In particular, Ahmad emphasized the need for models that can function efficiently with minimal data and time requirements, a departure from the resource-intensive models of the past.
As Ahmad concluded his presentation, he touched upon the future possibilities, introducing the concept of generative architectures. He proposed the integration of generative AI into the video analytics pipeline, combining the strengths of advanced models and generative models to optimize computation and enhance analytics capabilities.
In summary, Ahmad Shaik’s talk at MLDS 2024 provided a comprehensive journey through the evolution of computer vision and its impact on video analytics. From the overwhelming data challenges to the triumphs of AI in diverse industries, Ahmad navigated through complex themes with clarity and expertise. As the world of AI continues to unfold, Ahmad’s insights serve as a beacon, guiding the way towards a future where computer vision and video analytics play pivotal roles in shaping industries and ensuring a safer, more productive world.
