On 17th August, Team Nerds United Alpha (NUA) organized a meetup, based on AI and GPUs. The event named – ‘NUA Dev On Meetup v1.0’, in collaboration with KIIT School of Computer Engineering & Nvidia, was tailor-made for a limited audience, thus marking it the first AI & GPUs meet in Odisha. Around 50 participants comprising of both students, faculty and IT professionals from different organizations, attended the event.
The event was organized in School of Computer Science & Engineering, KIIT (Deemed to be University), Bhubaneswar. Kalinga Institute Of Industrial Technology, just recently gained the ‘Institute Of Eminence’ status. This makes it, one of the top seven private universities in India, to have received an acclaim of such stature.
Respected and honorable Dean (School of Computer Sc. & Engg.), Dr. Samaresh Mishra, delivered an awesome keynote, highlighting the achievements of KIIT, starting from its upbringing in 1992 as an ‘Industrial Training Institute’ to an Institute of Eminence in 2019.
In addition, he also honored the guest speakers at the meetup, with handcrafted fabric paintings, made by the tribal students of Kalinga Institute of Social Sciences (KISS).
Dr. Chittaranjan Pradhan (Professor, School of Computer Sc. & Engg), presented an initial talk about Deep Learning & Neural Networks. He primarily focused on the various types of ongoing R&D work on the same. His examples were centered on Deep-Learning in the healthcare sector.
Intro of NUA
Amlan Panigrahi (NUA Admin & KIIT Alumni) took the privilege, to introduce the audience about the community itself. Nerds United Alpha, established in 2016, is an initiative by Tech Legends, that primarily catered to connect with gaming and computer hardware enthusiasts. Not until this year, ‘NUA Dev On’, a spinoff community was initiated, to connect with developers.
Likewise, NUA’s methodology as a community has changed to “AI, GPUs & Gaming” – to gather such enthusiasts under one roof.
Nvidia Dev Talk
Mr. Sunil Patel (Senior Deep Learning Solutions Architect, Nvidia), was an honorary guest speaker in the meetup. The spokesperson delivered a talk about the various breakthroughs of GPU-Accelerated Deep-Learning made possible by the technologies of Nvidia.
Initiating his presentation with the latest Volta based GPUs primarily featuring 5120 CUDA cores, 640 Tensor cores and 21 billion transistors, Mr. Sunil explained the impact of GPU Computing in the contemporary world. In continuation, he elucidated the ease of GPU based application deployments with Nvidia GPU Cloud (NGC Containers), that can save the developers a considerable amount of time invested.
In addition he also shared the advancements in autonomous machines catering to domains like industrial, retail, logistics, etc. Starting from Xavier to Nano, with the Jetson platform, it has been made possible to deliver scalability and efficiency, in such fields.
Other key technologies, he discussed about were TensorRT for Deep-Learning inference, Clara for healthcare and use of RAPIDS in Data Science. For further information, one can opt to sign-up at Nvidia Developer Zone, to gain access to tons of free educational resources on Deep Learning & relevant technologies.
Embedded GPU Computing Demo With Jetson Nano
To make the audience, understand the magic of accelerated computing with GPUs, Amlan Panigrahi, gave a practical demonstration with Jetson Nano.
Jetson Nano is a tiny embedded SoC, tailor-made for portability and testing pre-trained AI models and various accelerated computing needs. The audience was first given a theoretical explanation with simple vector addition program, performed with GPUs in contrast to CPUs. The attendees learnt about the concept of using threads for parallel computing in GPUs.
Succeeding to the theoretical example, the audience got to witness the power of GPU computing with a sample program. We took the help of a sample multiplication program made by Mr. Avimanyu Bandopadhyay (Bioinformatics, PhD Scholar & GPU Computing Enthusiast). Execution on the CPU took around 28 seconds. In contrast, the GPU computation took a few milliseconds.
A video highlighting the implementation of YOLOv3 (a real-time object detection system) with Jetson Nano, was also shown to the participants.
Github Repository for sample multiplication code: NUA Dev On Repo
Hello AI World With Jetson Nano Repo: https://github.com/dusty-nv/jetson-inference
Ethics In AI
Many students, researchers, and developers have had been doing a considerable amount of work to train and optimize their AI models to achieve maximum accuracy. But how many are actually cross-verifying, whether their data collection and model training strategy is biased towards ethnicity? Manish Kumar Choudhary (Application Developer & Data Science Trainer, HighRadius) addressed such concerns with his talk on ‘Ethics In AI’.
Some of the key ethically biased AI examples highlighted by him were – ‘Black people tagged as apes by Google’, ‘Amazon recruitment tool biased against women (now scrapped)‘ and ‘Google Adsense displaying highly-paid jobs to men only‘. The spokesperson, highlighted the causes for such bias, being disparate treatment, ignorant data collection and preprocessing strategy.
Hence, to counter such issues, Manish elucidated key-concepts like disparate impact and examples of tools being used in practice like IBM AI Fairness 360. To gain further insight into the topic, you may refer to the mentioned deck link below.
Deck by Manish Choudhary: Ethics In AI.pptx
Other important guests from HighRadius: Sonali Nanda (VP, Product Management), Jeet Pati (Manager, Solutions Engineering) and Amitanshu Tripathy (Software Developer)
Quiz & Audience Connect
To connect with the audience, first a brainstorming session was initiated. Participants shared their ideas as AI projects. Some of them include ‘AI Based Sprinkler System For Irrigation’, ‘AI Based Inventory Stock Prediction System’, ‘AI Based Location Intelligence System’, ‘AI Based Media Player’, ‘Real-time AI Podcast Translation System’, etc.
Ankush Das (Tech Legends Founder & NUA Admin), asked a very important question to the audience. It was based on the recent GDPR implementation and the difficulties faced by business entities for AI model development with the lack of sufficient data. It was quite engaging and the attendees also presented innovative solutions to approach the problem.
Quiz questions related to AI & GPUs were also asked. Attendees with legible answers and active participation were rewarded with exclusive NUA posters. The session ended with a group photo and a closing note.
NUA Past Events & Roadmap
With a positive reception and feedback from audience, NUA shall come up with more events featuring rather more practical sessions and industrial use cases. To be a member of Nerds United Alpha, you may reach out to me or Ankush by dropping a mail. To know about our past events please refer the links as mentioned below:
Wonderful snaps taken by: Sangmen