Naveen Rao

Naveen Rao

Co-founder and CEO of of Nervana

THIS CHAT HAPPENED ON July 12, 2016

Discussion

Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
Hi everyone. I am Naveen Rao, cofounder/CEO of Nervana (www.nervanasys.com), a tech startup that serves "AI on demand." I enjoy talking about the future of computation, robotics, and neuroscience - but I find a lot of other things interesting too.. Ask me anything!
Katie Fritsch
Katie Fritsch@katie_fritsch
My friends company is new to Deep Learning, and he wants to get started with AI. Why should he choose Nervana over Tensorflow?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@katie_fritsch Nervana Neon is a fully supported stack. We don't have to rely on the open source community to fix bugs or add features. This is important to providing a high-end experience to our customers.
Uhma
Uhma@deleted-594972 · Brr
Hello Naveen, Thank you for giving us an opportunity to ask you questions. Question: As it stands, humans are guided by three books: 1) What we owe is in the IRS book. 2) What we can not do is in the Law book. 3) What we can do is in the Constitution book. Are there any books being written for AI? Thank you!
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@victorpolyushko As in, what will guide AI? I'd say that AIs are built and trained by us...so they serve our needs.
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@victorpolyushko Therefore, there isn't a direct need to inform what AIs should be doing. Perhaps in the future we'll have books on how to best use your AI...
Theoharis Dimarhos
Theoharis Dimarhos@theo_dimarhos · Marketing+Biz Dev at AngelouEconomics
Hi Naveen! If you were to start from zero and try to become useful to the AI community without being technical, how would you go about it?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@theo_dimarhos The business side is really just starting to become clear. Go and understand where AI can make the most impact...and find those truly disruptive use cases.
Jacqueline von Tesmar
Jacqueline von Tesmar@jacqvon · Community at Product Hunt ⚡️
What are the first tabs you open on your computer/phone every morning?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@jacqvon I usually read the latest techblog (TC, Venturebeat, etc). Facebook is a good source of interesting articles..I have the good fortune of being connected with people in the AI space who post good stuff.
Andrew Ettinger
Andrew Ettinger@ettinger · 👟 @wearAtoms // ex @Twitter @ProductHunt
It seems that "AI" is such a buzzword these days. Where do we draw the line one what is AI and what isn't?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@andrewett That's a good question...it seems to be synonymous with deep learning these days. I'd say any system that learns and adapts from data is a form of intelligence. By that measure, more deep learning systems are intelligent, but only at specified intervals when the models are being retrained.
Tomer Borenstein
Tomer Borenstein@tborenst · Student @ Carnegie Mellon (tborenst.com)
Hi Naveen, thank you for taking the time to do this chat. I am interested in your thoughts regarding the apparent push towards Conversational UIs by several of the big players? Also regarding conversational AIs - is there any indication that what happened to computer vision with deep learning in the past several years will happen to NLU/NLP?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@tborenst Well, i'd say that's already happened. Most of the state-of-the-art NLP approaches are deep learning now. This transition happened fast just as it did in computer vision. I think anything that decreases the resistance of using a machine is a good technology. Arcane interfaces like a CLI are great for experts, but make technologies less accessible. By making machines interact more like we do, technologies can become more a part of our life.
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
Thanks everyone for the great questions!! To learn more about how Nervana can help address your biggest machine learning challenges please email us at info@nervanasys.com.
An Jason
An Jason@an_jason · student
I am a grad student in physics, building and numerical simulating all different models, but very limited experience in computer science/AI. Is there any advice to me to enter this field? What's the advantage for us who are not from CS major?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@an_jason Physics and the math you're learning lays the ground work for much of AI. I'd say, get some practical experience. Download Neon (from www.nervanasys.com) and get a model up and running. You'll get a sense really fast for what is easy and what is difficult. The advantage you have is a broader view on machine learning and the mathematical foundations. CS can be narrow with regards to this subject at some places.
Emily Snowdon (née Hodgins)
Emily Snowdon (née Hodgins)@emilyjsnowdon · Operations @ Product Hunt
Hi Naveen, what does the future of robotics look like?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@ems_hodge Robots are moving toward becoming real agents in the world. Currently, they are mostly fixed function...but once they can sense, process, and adapt, there will be a new world of how they can impact our lives.
Ben Tossell
Ben Tossell@bentossell · Makerpad.co
Hey - if you had to swap lives with a tech ceo/founder for a week, who would it be and why?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@bentossell Hmm...that's a tough one. I can't say Tony Stark can I? ;)
Emily Snowdon (née Hodgins)
Emily Snowdon (née Hodgins)@emilyjsnowdon · Operations @ Product Hunt
What one piece of advice that you've received during your career to date has had the most impact on you?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@ems_hodge Do something interesting. I quit a lucrative career to get a PhD in Neuroscience because I truly was interested in the topic. Find things that are interesting and pursue because you love them.
Andrew Ettinger
Andrew Ettinger@ettinger · 👟 @wearAtoms // ex @Twitter @ProductHunt
Who are the leaders in the AI space that you most respect?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@andrewett Yann LeCun, Bengio, Hinton are the big ones...they persevered through a time where their ideas weren't taken seriously. That takes belief and guts. I hope to have such convictions about my own work.
Mary Devincenzi
Mary Devincenzi@mary_devincenzi · Public Relations
Hi Naveen, What are the key factors that have enabled recent advancements in deep learning?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@mary_devincenzi I'd say there are 3 major points: 1) amount of data we have access to. This is crucial to being able to build an effective machine learning model and wasn't really the case until the last 10-15 years. 2) fast processing power. These models can take a long time to converge...10 years ago it would have take an entire PhD thesis to train one model. 3) business use case. Now we demand that things learn and adapt to become better. We want this in our internet and device experiences...and that wasn't true in the past (at least not at a large scale)
Jacqueline von Tesmar
Jacqueline von Tesmar@jacqvon · Community at Product Hunt ⚡️
Where is one place you want to visit that you haven't been before?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@jacqvon Antarctica :)
Sujith Perla
Sujith Perla@sujith_perla
Hello Naveen, Was just curious about the newly released iOS app called Prisma which turns the photos into artworks using Deep learning algos. Any thoughts on how these deep learning algos are used here? Thank you!
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@sujith_perla I think this is a cool use of models developed for a different purpose. It appears to reduce an image into some kind of feature space that is pleasing to our eyes...I'd say finding the use for deep nets is much more of an art than making them work these days
Matt Schwartz
Matt Schwartz@matt_schwartz · Product Manager, Intuitive Surgical
Hi Naveen, thanks for the opportunity! I'm a product manager for a robotic surgery company and very interested in the future role of AI/machine learning in making surgery better or eliminating it altogether. Are there any specific companies in this space that interest you? Does Nervana have anything cooking along these lines? Thanks!
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@matt_schwartz There are lots of interesting companies like Kindred.ai or others building robotics platforms. It's not clear how this space will sort out just yet though. At Nervana we're focused on enabling research and backend processing in the cloud for these and many other applications. We don't sell specific solutions, but more of a horizontal platform approach
Matt Schwartz
Matt Schwartz@matt_schwartz · Product Manager, Intuitive Surgical
@naveen_rao awesome, thanks!
Matt Schwartz
Matt Schwartz@matt_schwartz · Product Manager, Intuitive Surgical
@naveen_rao Any good articles on Kindred.ai that you would suggest? Their website seems to be a literal black box, and I'm not really finding anything about them elsewhere. Thanks again!
Sarav
Sarav@sarav_1n
Hi Naveen, Ive used neon and really love how powerful it is. However the whole community seems to have swayed towards tensorflow recently. With Sundar's recent announcement of TPUs, tensorflow looks like the way to go for something that has community support, proven cloud hosting domain expertise, continual updates, great tutorials and documentation. What differentiates you guys? In other words why should we build our AI projects around neon and what you guys have to offer? Thanks
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@sarav_1n Neon allows us to provide a fully supported experience to our customers without relying on a 3rd party like Google. We recognize that TF has a lot of mindshare now...and we're working on interoperability with TF in the next few months. Since we are also building our own hardware, we feel it is important to build the stack the right way for our path forward...our performance will be well beyond anything that TF supports. So, I'd say that our solutions will run TF, but we can offer a better, faster, more relaiable experience with our own stack
Gopal Mahadevan
Gopal Mahadevan@gopalmahadevan1
Good morning Naveen, With the advances in AI come many potential perils. For example, there is one scenario in which AI takes control of the planet away from the human race. Another is AI automating many human jobs out of existence (which is already happening). What is being done (and what should be done in the future) to mitigate these risks and others? What role should we humans play in designing these intelligent machines to minimize the negative and unintended consequences? Thank you, Gopal
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@gopalmahadevan1 There are efforts like OpenAI that are attempting to have some responsibility around AI. The thought is, if these technologies are available to all, there will always be a balancing force. At the moment, I don't believe we're headed down the path you describe...AIs are tools that allow us to do more with data. They do not have intent. Their intent is our intent. We should really only fear ourselves here as AI just give us more reach and influence
David Brann
David Brann@david_brann
@naveen_rao @gopalmahadevan1 I would add that automating human jobs should be regarded as a bonus. Is driving a large cargo truck for 8+ hours a day a good use of human abilities? NO! Automate it! Free up humans to do things only humans can do.
David Brann
David Brann@david_brann
@naveen_rao @gopalmahadevan1 Please also recall the history of other jobs that have been automated (e.g., typesetting for printing) only to end up creating a whole new set of related jobs (e.g., graphics designers
Ryan Loney
Ryan Loney@ryan_loney · systems engineer
Hi Naveen, It was recently discovered by my health care provider that I have a rare but life-threatening heart condition that may require surgery and implantation of a medical device in my body. The discovery was made by a cardiologist who examined an electrocardiogram (EKG/ECG) that was taken of my heart. A very clear pattern on my EKG was recognized by the cardiologist, which he happened to know was a clear indication that I have the rare heart condition. Unfortunately, a doctor who took two EKGs of my heart in 2012 showed the exact same pattern, but because he was not familiar with what to look for about my particular syndrome, it went completely overlooked and completely unnoticed until two weeks ago. How could AI be used by health care providers to recognize patterns on electrocardiograms, ultrasounds, CT scans, etc. to help detect conditions like mine? If so, how long until we start seeing this type of technology put into production? Thank you, Ryan
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@ryan_loney This is one of the areas I think will be huge in the future. We should have awareness of this kind on every scan, and this is what AI can do. It's not practical for human docs to be able to find every anomaly every time...but machines are tireless and cheap. They can check against the entire database of maladies on every scan, every time
Why did you choose San Diego as the headquarters for Nervana instead of conventional SF?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@aash_kaddi I like how you say 'conventional' SF. SF is expensive and crowded. We do have a bit more than half of the company in Palo Alto, but we enjoy the lifestyle in San Diego. Also, given the proximity to the bay area (only 1 hour flight) it hasn't been a huge issue to continue to engage with the investor community in the bay area. We see our San Diego/Palo Alto strategy of getting the best of both worlds
Julie Galla
Julie Galla@julie_galla · PR, Sparkpr
Hi Naveen - when many people think AI, they think about Watson and Go - how is AI going to get out of gaming and into the real world?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@julie_galla We're already seeing AI in most of our web and smartphone interactions. Facebook heavily relies on it for face tagging or suggestions. Google uses it in search and image search. Your phone uses it for speech to text. Bots are now becoming ever more prevailent
Ben Tossell
Ben Tossell@bentossell · Makerpad.co
Where do you think AI will have the biggest boom and biggest bust? Seems AI is being used just about anywhere!
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@bentossell I see this time as something akin to when the microprocessor came out...AI will literally change every interaction with have a a machine. Deep learning is a general technique and can be applied pretty much any data type and has many applications. Farming, finance, web, IoT, healthcare are some of the high impact areas.
Sarav
Sarav@sarav_1n
Hi Naveen, A few more questions. I've been following Nervana systems from the initial talk Deep Learning: Intelligence from Big Data hosted by Steve Jurvetson. Can you comment on the following? 1. Most help I can get now is on stackoverflow. Do you plan on opening up a forum so we can have a centralized knowledge center? 2. I think most people do no realize how awesome what you guys are doing is.Is there a plan to spend more on marketing, official tutorial videos etc. Generally speaking it makes sense to go after large enterprise customers. But i feel like this will help build a stronger independent developer community that will eventually see more content/benchmarks, tutorials etc, which should help you guys and all of us. I also see that only a small part of your staff are in developing products(most seem to be around algorithms and hardware development). Any idea of building industry specific solutions(example, HIPPA compliance for healthcare folk who want to build AI/Deep learning stuff) 3. Is there a plan to sell your architecture to independent developers or will it only be cloud hosted? Will there ever be an on-premise model? I still see it being more viable to own a bunch of TITAN X cards rather than using the cloud as of today until there is something production ready and we need to meet SLA's. Thanks Sarav
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@sarav_1n 1) Yes, we already have a support channel on Github under Neon. We are pretty quick to respond or someone else in the community might also get you an answer. Give it a shot! 2) We totally agree with you here. But, we are a small company still with limited resources. We tend to engage at academic conferences and give talks that contribute to the community. We do intend on having more video tutorials that will get people up and running. We're working with OpenAI on the RL Gym. This is just the beginning :) Our team is mostly working on engineering, but product dev touches every part of the org. All of our researches engage with users and customers. The cloud team works most directly on things that the customer touches, but every one is working toward building a better product 3) We wil have both a public cloud and on-prem version. We will not be selling chips or accelerator cards though...that's a bit of an artifact of graphics...people want solutions and that's what we sell. Our on-prem appliance will have most if not all of the functionality of the cloud in something that goes behind a user's firewall. People do believe they can go do a lot of this stuff on their own at first, but quickly realize that there is a lot of time spent getting things running an maintaining. We take that pain away, and let users focus on developing their kickass deep learning solutions!
Ronald Eng
Ronald Eng@ronaldeng5
Hi Naveen. Can you speak more about hardware acceleration and its impact (if any) on Nervana's high ranking in Kaggle competitions? How might hardware acceleration change the focus of future industries you are targeting?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@ronaldeng5 Speeding up deep learning training is very important. The cycle of innovation is essentially bounded by how many things you can try...if the turn around time is 1 month, that severely impacts innovation. Performance of models will just get better as training time decreases. What's more, is that current constraints on model size will be removed with our accelerators...there are lots of 'artifacts' of this in all published models out there. Better training will certainly change the way companies build products in the future.
Jessica Lauren
Jessica Lauren@jessica_lauren1
Hi Naveen, thank you for taking the time to answer questions. I was curious about your feelings regarding the recent Tesla autopilot accident.. I'm a really strong AI (and Tesla) advocate and was very saddened by this news. Is there a learning curve with machine learning? Should we expect more fatalities? Is there anything we can do prevent accidents in the interim?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@jessica_lauren1 That's a great question. I believe this is due to a really unfortunate set of circumstances with autopilot. But, if you look at the stats, the number of miles between accidents is better for autopilot as compared to humans. This is of course not comforting to the victim here, but does mean that we're already saving other lives with these technologies. This is a reminder that any technology is not flawless and that we, as the controller of the robot (the car in this case) need to keep some awareness about what is happening
Thomas Stöcklein
Thomas Stöcklein@tomstocklein · FoundersFundersFuture.com
What are the biggest challenges for Nervana right now? Where do you see the company in 3 years from now?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@tomstocklein Challenges are around getting the right product and pricing strategy for our users. We have a good handle on the technology part, but the market is evolving very fast here. Most of it has been going in the direction we have on...so that's good. We have great customers giving us great feedback, and have a good path to figuring out how best to add value to them. Our product will continue to improve with this virtuous cycle In 3 years we'd like to be the world's platform for doing deep learning in the cloud. We see cloud learning and model deployment as being in nearly every industry, and we're well positioned to supply these technologies.
Ryan Hoover
Ryan Hoover@rrhoover · Founder, Product Hunt
Naveen! The AI community has its contrarians and those with very strong opinions. What do you believe that many others in the space do not?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@rrhoover I'm a strong believer in neural approaches. I think deep learning has given us a small set of primitives that can solve a vast set of problems. What's more, we know how to distribute and parallelise those computations. We see this as the path forward for what a computer will become...it's no long just doing a set of calculations in sequence. It's finding useful inference in data. In addition, with the slowdown in silicon (end of Moore's law), we need to find new approaches to innovate on the current substrate. This is it.
Jessica Lauren
Jessica Lauren@jessica_lauren1
Any words of wisdom for people choosing between jobs at well-known, tech giant companies vs. less reputable startups?
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@jessica_lauren1 This comes down to personality really. If you thrive on being thrown into the fire, by all means, join a startup! You'll learn faster than ever. If you like the security and resources of a large company, that might be a better fit. I see advantages to both, but I'm clearly biased :)
Matt Schwartz
Matt Schwartz@matt_schwartz · Product Manager, Intuitive Surgical
Hi Naveen, follow up question on Kindred.ai. You mentioned them as an interesting company in AI/Healthcare. Are you able to share any details about what they're working on? Their website is a black box, and I can't seem to find any information about them anywhere. Thanks!
Naveen Rao
Naveen Rao@naveen_rao · CEO, Nervana
@matt_schwartz They are more in the robot platform space. I'll let them share the details they are comfortable sharing ;)
Matt Schwartz
Matt Schwartz@matt_schwartz · Product Manager, Intuitive Surgical
@naveen_rao ha fair enough - thanks for the great chat!