AI and oneAPI - A oneAPI Spotlight Discussion with AI Innovators

In the podcast episode of Intel’s "Code Together" series, AI and oneAPI - A oneAPI Spotlight Discussion with AI Innovators, the spotlight shines on the oneAPI Innovators participating in a oneAPI AI Hackathon. The discussion revolves around their insights into the application of AI, the significance of AI in today's world, and the role of oneAPI in delivering high-performance solutions to the AI landscape.

Motivation Behind the Hackathon Participation

The podcast features insights from three innovators: Mohan, Sudarshan, and Yuvraj, each bringing a unique perspective. Mohan highlights the dual motivation of contributing to open innovation in education and receiving support from senior managers at Intel. Sudarshan expresses how the hackathon aligns with their pre-existing idea of implementing object detection for autonomous vehicles. Yuvraj's team addresses the challenges of driving vehicles in rough and hilly terrain with a focus on using LED technology for object detection.

Problem Statements and Solutions

The innovators tackle diverse problem statements. Mohan's team focuses on developing a real-time Adaptive Driver Assistance System to address the challenges of night driving. Sudarshan's team works on object detection for autonomous vehicles, implementing an adaptive headlight mechanism to reduce glare. Yuvraj's team addresses the challenges of driving vehicles in rough and hilly terrain with a focus on using LED technology for object detection.

Integration of oneAPI

The podcast then delves into how the innovators integrate oneAPI into their projects. Mohan highlights initial reluctance but praises the ease of adapting to Intel's extensions. He specifically mentions using oneAPI toolkits like the Intel® Extension for PyTorch and Intel® Neural Compressor for efficient model quantization, resulting in a 4.4x improvement in inference speed.

Sudarshan emphasizes the seamless integration of oneAPI, especially with the ease of incorporating it into existing code with minimal changes. The team leverages oneAPI's libraries to achieve a significant speedup of 4.56x for both intents and training phases.

Yuvraj details how they utilize oneAPI's capabilities to compress their model, resulting in a 35.5% improvement in average execution time. He also commends the minimal lines of code changes required for integrating oneAPI.

Exploration of Intel Developer Cloud

The podcast touches upon the innovators' positive experiences with the new Intel® Developer Cloud. They highlight its seamless user experience, providing access to state-of-the-art Intel resources for model training and inference. The cloud's contribution to improved benchmark results is acknowledged, with significant speedups in both intents and training phases.

Future Plans and Applications

The innovators then share their future plans and the commercial viability of their projects. Sudarshan intends to create a minimum viable product for local implementation of their adaptive headlights. At the same time, Mohan emphasizes the potential market demand for their driver assistance system, particularly in hilly regions. Yuvraj outlines plans to test their online learning solution on busy traffic roads, aiming for commercial applicability.

Research Focus and Evolving in AI

The podcast also sheds light on the innovators' continuous learning and research regarding AI. Mohan discusses his focus on areas like causal analysis, knowledge graphs, and natural language processing, aiming for continuous improvement in the field.

The innovators collectively encourage developers to explore the oneAPI community, emphasizing its potential for diverse applications. They laud the community's support for college students and industry professionals, expressing gratitude for the opportunities and friendships developed through their involvement. The success stories and insights shared here are an inspiring narrative for developers entering the dynamic and evolving field of AI.

We encourage you to check out Intel’s other AI Tools and Framework optimizations and learn about the unified, open, standards-based oneAPI programming model that forms the foundation of Intel’s AI Software Portfolio.

The Speakers:

Kavita Aroor is a Developer Marketing Lead for Asia Pacific and Japan at Intel. She drives the developer outreach, engagements, and GPU, HPC, and AI programs.

Mohan Krishna R has around 13 years of experience providing comprehensive solutions to complex data problems across multiple industries like technology, FMCG, retail, etc. He is currently working as Principal AI Scientist at Course5 AI Labs. His work around predicting the Indian Elections winner using social data was published in mainstream media, including the Washington Post, Economic Times (digital & print), and Bloomberg.

Sudarshan Bandyopadhyay is pursuing his master's from the Indian Institute of Science, Bangalore, in Robotics and Autonomous Systems (Cyber-Physical Systems). His current areas of interest are Edge AI, Embedded Systems, and Machine Learning. He currently works under the guidance of Dr. Punit Rathore in developing an Intrusion Detection System using ML for Autonomous Vehicles.

Yuvraj Singh Deora is an enthusiastic tech pursuer on the path to an Integrated M.Tech in Software Engineering at VIT. His drive lies in bridging the gap between cutting-edge technology and real-world solutions. From AI and IoT to Blockchain, he is dedicated to tackling practical challenges that truly matter.

References

ChatGPT & Wolf, A. (2023, November 16). Summarize the main points from the following transcript. https://chat.openai.com