SMW '18: Cisco's Sangeeta Ramakrishnan Talks Machine Learning and eSports
Learn more about machine learning at Streaming Media's next event.
Read the complete transcript of this interview:
Tim Siglin: Welcome back to streaming media West 2018 I'm Tim Siglin. Contributing editor of streaming media magazine, and the founding Executive director of not for profit help me stream. Today I've got with me Sangeeta Ramakrishnan, a Distinguished Engineer in the CDO office at Cisco.
Sangeeta, you were on the SM advance forum with Dom Robinson and me. Two things you talked about there really fascinated me were sort of latency and eSports. Before we get to talking about those, tell me what you're doing at the conference here?
Sangeeta Ramakrishnan: I was on a panel yesterday on machine learning and AI. How that's leverage for video sentences, so that was exciting. Also, I'm also the founder of the Women in Streaming Media organization. We also had the meetup yesterday, so that was pretty exciting too.
Tim Siglin: Let’s talk about machine learning and AI, and the distinction between those two, because everybody talks about AI. Hardly anybody talks machine learning.
Sangeeta Ramakrishnan: I think the general public and the media talks about AI. The people working in the space, I suspect, use the term machine learning more.
I prefer to use the term “machine learning.” For a lot of people, it’s pushed too far. To them, AI is only when you truly can replicate human intelligence. Which we're not anywhere close to.
Machine learning is a smaller goal, but an achievable goal, and is delivering benefits already.
Tim Siglin: From a video standpoint, how does it differ? Say, machine learning from computer vision. Because, computer vision has been one of those topics over the years too. Is it primary metadata, or is it actually watching the scene to figure out who appears in it, et cetera?
Sangeeta Ramakrishnan: Good question. Computer vision is about replicating sort of vision that humans have, but from a computer perspective. Can a computer detect a cat versus a dog? Can it detect a soccer ball versus a basketball, those kinds of things?
Machine learning is broader. You can use machine learning to identify anomalies. If there is some problem with your traffic, problem with users abandoning content. Any kind of data. Traditionally, you would detect anomalies by simply rule base matters. If a certain thing exceeds 90%, or something drops below 10%, maybe even dropping going about 90%, or some cases some parameter would be under the range of of 10 and 90, but it's unusual, because it's 3 A.M. why have it even at 40% at 3 A.M.? That's a precursor to problems, because, maybe when their peak hour comes now you can detect it earlier. Those kinds of anomaly deduction are not really computer vision. Machine learning can be many more things.
Tim Siglin: I think Roger Pantos, this morning in his presentation, addressed one of those models of machine learning when he talked about using the gap term when there is discontinuity. To say yes, it's going away, rather than throwing a 404. Let's assume there may be a potential gap in here and it's coming back later.
Sangeeta Ramakrishnan: Right, right, exactly.
Tim Siglin: Okay, very good. Let's talk a little about eSports. You indicated when we did the SM Advance Forum, that eSports is a growing market for video. Is it growing just from a consumption standpoint, or a content production standpoint as well?
Sangeeta Ramakrishnan: I think, from both. If the consumption goes up it causes people to find out, how can we get into this space? How can we monetize this? Just at Streaming Media West on the opening night, I met someone from a university in Georgia who was telling me that they want to hold eSport competitions. For example, they don't have the network capacity to pull that off. There is a desire to do more in that space. I think the technology, and the infrastructure is not yet fully there to allow people to do as much as they would like to. In that sense to me that's depends on demand. Ithink it's good to grow.
Tim Siglin: Is that where the capacity issues need to be addressed to deal with latency in eSports?
Sangeeta Ramakrishnan: Exactly. eSports, pauses primarily two challenges, one is capacity and latency, and the reason I mention capacity is with eSports, like a Twitch broadcast, the person sending a lot of traffic in the upstream. Traditionally in a broadcast network, we’re used to people consuming a lot of bandwidth in the downstream. As a broadband provider, I'm sure operators are happy to see demands grow in the upstream, but then they need to be ready for it too. The capacity's there. These cable networks are getting closer to symmetric. They’re not quite there, but there are new technologies coming out in the next 12-24 months. We are gonna get to symettric bandwidth, even in cable networks.
Tim Siglin: I remember back as far as 2001--actually on 9/11--being at a show. They were debating the merits of SDSL versus ADSL. Of course, the phone companies wanted ADSL, but we knew from a consumer content creation standpoint, that eventually we would to have to go synchronous. It's interesting to hear we're finally getting close to parity at that point.
Sangeeta Ramakrishnan: Of course, besides the capacity, the latency is a big issue, because if you want interactivity. You have to be able to get the video across, and have your viewers interact with you in a timely fashion. So that is the other part that is challenging compared to live video streaming such as what we’re doing today, because there isn’t that interactive component in most of live streaming.
Tim Siglin: In those scenarios with interactivity, do you have to limit the number of hops to a certain number of hops from the initial sender and to the receiver and then back again, or does a number of hops that it goes through not really make a difference from an overall latency standpoint?
Sangeeta Ramakrishnan: The hops do matter. But a lot of the latency is probably driven by actually the transport and any kind of re-transmission that’s required. Because, traditionally for streaming media, we’ve used TCP--I mean, we’ve abused TCP, but we've used it for 30 years now. The TCP spec has been around for 30 years. Maybe it's time for something else, because as an example for live streaming, especially eGaming and things like that. Maybe it's much more important to give the content there on time, even if it's not perfect. Those kinds of things may change. Redundancy--we may need to look at how to improve redundancy, maybe add forward error correction. We may have to blend things, which come traditionally from the video conferencing world into the live streaming world. I see that as sort of mix of those two spaces.
Tim Siglin: Interesting. I started in video H.323, and H.263, before I was in streaming-
Sangeeta Ramakrishnan: Me too.
Tim Siglin: Our low latencyies were very low latencyies compared to streaming. How do we deal with it on an application level? Is there a way to speed things up on an application level as well? You mentioned new transports.Obviously we've had UDP and reliable UDP, is it essentially making sure the applications support those?
Sangeeta Ramakrishnan: Those would definitely be the place to start, to consider what transports to use. Unfortunately, today there isn't a clear winner in the transport that I could tell. Quick is framed to be better than TCP. Within TCP there's new congestion control protocols coming. There are a couple of universities working on some really interesting transports. One is in Israel and one is at Stanford. Of course google has BBR for congestion control. There's a lot of moving parts, a lot of changes. SRT is trying to address latency. There isn't a clear winner yet. I suspect a winner in the next couple of years. That space is going to perhaps settle. Because, right now it's still an open area.
Tim Siglin: Definitely. Well, thank you very much for your time. We will be right back after a brief break.
Related Articles
Streaming Media's Tim Siglin discusses key topics of Streaming Media West 2018, including microservices, interactive streaming, eSports, MR/AR (but not VR), low latency, and Women in Streaming Media.
14 Dec 2018
Streaming Media's Tim Siglin interviews nanocosmos CEO Oliver Lietz at Streaming Media West 2018.
13 Dec 2018
Streaming Media's Tim Siglin interviews Brightcove's Lexie Pike at Streaming Media West 2018.
12 Dec 2018
Streaming Media's Tim Siglin interviews Meet Group VP of Livestreaming Lauren Hallanan at Streaming Media West 2018.
11 Dec 2018
Streaming Media's Tim Siglin interviews Ellation VP Engineering Michael Dale at Streaming Media West 2018.
10 Dec 2018
Streaming Media's Tim Siglin interviews VideoLink Director of Production Kevin McCarthy at Streaming Media West 2018.
07 Dec 2018
Streaming Media's Tim Siglin interviews Twitch's Tarek Amara at Streaming Media West 2018.
06 Dec 2018
Streaming Media's Tim Siglin interviews THEO Technologies CTO Pieter-Jan Speelmans at Streaming Media West 2018.
05 Dec 2018
Streaming Media's Tim Siglin interviews Mobeon CEO Mark Alamares at Streaming Media West 2018.
04 Dec 2018
Streaming Media's Tim Siglin interviews Microsoft Principal Lead Program Manager Chris Knowlton at Streaming Media West 2018.
03 Dec 2018
Streaming Media's Tim Siglin interviews JD Fox, Director of Partnerships and Product Management at PrestoSports, at Sports Streaming Summit and Streaming Media West 2018.
30 Nov 2018
Streaming Media's Tim Siglin interviews IBM Watson Media and Weather Senior Solutions Engineer Scott Grizzle at Streaming Media West 2018.
29 Nov 2018
Streaming Media's Tim Siglin interviews Beamr VP Marketing Mark Donnigan at Streaming Media West.
27 Nov 2018
Streaming Media's Tim Siglin interviews First Tube Media CEO Andrew Beranbom at Streaming Media West 2018.
26 Nov 2018
Streaming Media contributing editor Tim Siglin interviews Mainstreaming Global VP, Solutions Engineering Mike Smith at Streaming Media West 2018.
23 Nov 2018
Streaming Media's Tim Siglin interviews MediaKind CTO & Head of Strategy Mark Russell at Streaming Media West 2018.
21 Nov 2018