NVidia Developer's Connect 2017

NVidia Developer's Connect 2017

NVidia is not just about gaming anymore.

Whenever I used to hear the word “NVidia” I used to think about all the processing power that it provides so that a game on my PC could work flawlessly. Back in school I used to think why isn’t NVidia doing more than this? With all the processing power to render such high quality graphics in realtime, why isnt it becomming more than just a gaming hardware manufacturing company? Seems like the GAME has finally changed.

NVidia organised a conference in various parts of India on AI, Machine Learning and Deep learning, about how everything we now use in daily life, from word predictions in keypad to a match on dating platform, how everything is weaved with AI or its subtrate. I got to attend the one which held at the hotel “The Suryaa” in New Friends Colony, Delhi on 29th of Nov’ 2017. What happened there? I’ll share everything in brief:

NV Welcome

The conference started with greetings and warm welcome, well the conference was about AI, so how could we expect a plain old welcome? A video was projected to extend the welcome and to give us an idea where is the world now and where we are going. I didn’t miss “Where can we go?” on purpose. With revolutionary hardware and Deep Learning algorithms, almost anything can be done in a computer model, no matter however hypthetical it may seem. That’s basically what Machine Learning does in first place. Taking a hypothesis (a computer model) and giving it more confidence according to the data fed to it (training data). With more data, the hypothesis acheives more accuracy. The video shows the potential (which is theoritically infinite) of AI and Deep Learning in less than 3 minutes:

A New Computing ERA

By Mr. Vishal Dhupar, MD, NVidia Sout Asia.

He gave a VR (Virtual Reality) demo of a collaboration work in which the demonstrator wore a VR gear and collaborated on a Car project (customising a car and reviewing it thorughly). The interface in the VR for customising tools and handling the settings was like straight out of a Sci-Fi movie. There was an expanded view which isolated every part of the car (from wind sheild to every nut in the gear box) in the 3D space (which is practically infinite in a VR world) by increasing 3D spaces in correct proportions between each part such that the geometry and alignment of any part is not changed. That gave the collaborator a view to inspect everything with precision.

Artificial Intelligence - The vision for economic growth

By Dr. Shantanu Choudhury, Director, CEERI.

Explaining how machine learning combined with denoising algorithms can help create super resolution algorithms. Generating a High Resolution image from low resolution image. How deep learning can be used to fill in the missing areas of an image by ‘learning’ from other areas of image, data from which can be gathered. Even with all the advancing technology in the world, ultrasound images couldn’t be made clearer they were speckled then, and are speckled now, but are they? Dr. Shantanu explained how deep learning algorithms are being developed to yeild a de-speckled image of the ultra sound. Though this is acheivable with remarkably high accuracy in the current date but the main challenge is to keep the speckles which are needed. With ambiguity in the speckles and various anomalies in it, doctors figure out any underlying condition/problem those rather unusual speckle correspond to. So the optimized algorithms need to figure out to not to de-speckle those areas.

Deep Learning and the future Enterprise

Dr. Lovekesh Vig, Sr. Scientist, TCS Research Labs

Explaining how enterprise work is being enhanced and optimized using deep learning. Giving the example of Amazon for general understanding. Amazon hires more than 50000 people per quarter to handle the manual picking which can be done only so much by the machines up to a certain level. These numbers double, sometimes triple in peak times like Christmas.

So, Amazon spoke about this problem to a wide audience (developers from all over the world) and came up with Amazon Pickup challenge where one has to show a remarkable and feasible solution to pickup items, label them (classifying according to the object) and put the to the assigned area of the label (with good accuracy). Though this too can be acheived with current technology but the problem becomes challenging when the item is in different angle and various items surround it which are similar to it but belong to a different label. Everything of this has to be done in realtime.

TCS (Tata consultancy services) developed Knadia, a virtual assistance bot which is hooked with their network as well as the physical robot. The software instance of Knadia provides with the assistance of all the tools and softwares related to TCS and the robot can converse with humans using Natural Language processing (NLP) to answer the querires related to TCS conducts.

Driver assistance in Indian Traffic context

By Dr. Brejesh Lall, Asst. Prof., IIT Delhi

The driving assistance models developed by tech giants like Google and Uber can’t exactly work in India as expected.

With pedestrians walking in the areas they’re not supposed to walk, animals crossing roads, sometimes sitting in between the roads, buses stopping abruptly to stop passengers, etc. Though however sad it may sound to accept the plight of the current road and traffic morals in India, a life is worth more than a rule. To ensure safety, driving assistance models need to be trained with India specific data. So a team assembled by Dr Brejesh Lall at IIT Delhi did it, though partially but it’s a remarkable acheivment given that they built the India Specific data set from scratch on their own by recording various road conditions and traffic around Delhi to simulate the most common scenarios in India.

Algorithms deciding life partners and careers

By Nitendra Rajput, Senior Vice President, Analytics at Infoedge

Not deciding life partners but helping you to not to choose the wrong one, that’s indeed brings you one step closer to choose the right one. Deep learning and AI is being used by websites such as www.jeevansathi.com to detect the fake profiles. Though the accuracy might not be so high to report the user directly but is enough to raise a suspicion and involve a human to inspect the profile thoroughly.

Resumes, job profiles and a large population seeking jobs makes it difficult to find the right match, both for interviewers and interviewees. So, deep learning models are trained such to sort out the candidates according to the companies code of conduct rather than plain filtering of required skills. www.naukri.com uses such algorithms.

Though India is doing great in this field but tech giants like Google and Facebook are way ahead by a large leap, so much that it’s okay to say that we have just started to scratch the surface.

These series of conferences were held to make people aware about what deep learning can do and why we should do it? Or stealing the words of Mr. Vishal Dhupar, “We need to get ahead in this field and we need to do it fast, otherwise there might come a time we might not even have a place on the table in which decisions about the future of our country are being made.”

Supporting this initiative by NVidia, let’s start a journey with Deep learning tools as our guides, NVidia provides all the Machine learning and deep learning tools for developers for free (no I’m not being paid to say that, it’s just a statement from NVidia fan), let’s THINK about the future we want to build and then build those machines which would help to build that future for us:

comments powered by Disqus