Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Traffic Prediction with Advanced Graph Neural Networks

deepmind.com

Post reply on HN
Print view View original

21–30 of 45 posts

Re: Traffic Prediction with Advanced Graph Neural Networks

#21

Post by goodside »

rkagerer wrote: post #8
2bitencryption wrote: post #5

Google Map's traffic prediction has always led me to a very curious question: Clearly Google Maps has the ability to turn into a feedback loop. Traffic exists -> people use Google Maps to find better routes -> traffic is modified due to people taking alternate routes -> new traffic emerges. So my question is: what is Google Maps traffic optimizing for? The best traffic experience for User 3982274, or the best traffic…

If drivers discovered Google Maps was intentionally sending them down sub-optimal routes, they'd quickly switch to a different navigator. (Even if the Google way was better for the network overall).

Not if the sub-optimal routes happen only intermittently and the extent to which they’re sub-optimal (or whether they’re sub-optimal at all) is difficult to determine. If you take the same route every day you’ll notice when it changes, but who are you to say a one-time detour wasn’t the right decision given what you don’t know about traffic conditions?

Re: Traffic Prediction with Advanced Graph Neural Networks

#22

Post by datameta »

symplee wrote: post #17

Is it just me, or is it a little disingenuous to write the claim "...improve the accuracy of real time ETAs by up to 50% in places like Berlin, Jakarta, São Paulo, Sydney, Tokyo, and Washington D.C." When the actual numbers listed for those cities are: Berlin - 21% Jakarta - 22% São Paulo - 23% Sydney - 43% Tokyo - not listed Washington D.C. - 29%

It gets even more muddled when you consider they mention the following: "While Google Maps’ predictive ETAs have been consistently accurate for over 97% of trips, we worked with the team to minimise the remaining inaccuracies even further - sometimes by more than 50% in cities like Taichung."

Are they essentially saying that they lowered 3% inaccuracy to ~1.5% in Taichung? (And nevermind the fact that 51% is described as "more than 50%"...)

Of course this type of work is fascinating. Getting from 97 to 98.5% accuracy is far far more difficult than getting from 95.5 to 97%. But I don't enjoy the fudging of the perception of results.

Re: Traffic Prediction with Advanced Graph Neural Networks

#23

Post by frogblast »

paxys wrote: post #7
2bitencryption wrote: post #5

Google Map's traffic prediction has always led me to a very curious question: Clearly Google Maps has the ability to turn into a feedback loop. Traffic exists -> people use Google Maps to find better routes -> traffic is modified due to people taking alternate routes -> new traffic emerges. So my question is: what is Google Maps traffic optimizing for? The best traffic experience for User 3982274, or the best traffic…

When User 3982274 is on a busy road using the app, Google optimizes for that user's experience. If every user on that road is using the app at the same time, these algorithms should theoretically result in the optimal condition you described above. For example, if there are two roads leading up to the destination, one at 100% capacity and the other at 0%. The app will start routing people from road 1 to road 2. When…

I live in an area strongly impacted by Google making ‘individually optimal’ decisions for each driver, and actually leaving those drivers in a dramatically worse situation.

I live in a rural area, between a major population center and a major resort area, with one major highway and a few small back roads that provide alternate paths for part of the highways route. Every summer weekend the highway becomes highly congested.

Google quickly starts routing people down the back roads because of a 30 minute delay on the highway. A sudden crush of cars hits these back roads, and they end up gridlocked for 3-4 hours. Google then realizes traffic is literally stopped on these roads, and stops sending new traffic down those routes. But the people already on them are still stuck for hours.

It gets smelly when a bunch of drivers take a shit on the side of the road because they can’t go anywhere else, and leave it there.

All because Google simultaneously made an ‘individually optimal’ decision for a whole bunch of individual drivers at once.

——

Another example in the same area actually causes a backup on the highway itself. Google started suggesting one back road that required an unprotected left turn across oncoming traffic on the highway, to avoid a 10-15 minute delay further on the highway. Drivers dutifully followed directions by getting into the left turn lane.

The drain rate of the left turn rate is slow because oncoming traffic is also high. The left turn lane fills up, and one driver with directions to turn then stops in the traffic lanes to wait for room to get into the turn lane. And suddenly the highway is now encountering 2-3 hour delays that don’t clear for most of the day.

Re: Traffic Prediction with Advanced Graph Neural Networks

#24

Post by bumby »

frogblast wrote: post #23
paxys wrote: post #7

Earlier quoted context omitted.

When User 3982274 is on a busy road using the app, Google optimizes for that user's experience. If every user on that road is using the app at the same time, these algorithms should theoretically result in the optimal condition you described above. For example, if there are two roads leading up to the destination, one at 100% capacity and the other at 0%. The app will start routing people from road 1 to road 2. When…

I live in an area strongly impacted by Google making ‘individually optimal’ decisions for each driver, and actually leaving those drivers in a dramatically worse situation. I live in a rural area, between a major population center and a major resort area, with one major highway and a few small back roads that provide alternate paths for part of the highways route. Every summer weekend the highway becomes highly conge…

I wonder if this will be improved with more realistic cost functions that adequately model those localized effects.

Probably impossible at a global level but I wonder if it’s possible to eventually model and update those cost functions periodically to represent local maps

Re: Traffic Prediction with Advanced Graph Neural Networks

#25

Post by IAmGraydon »

paxys wrote: post #7
2bitencryption wrote: post #5

Google Map's traffic prediction has always led me to a very curious question: Clearly Google Maps has the ability to turn into a feedback loop. Traffic exists -> people use Google Maps to find better routes -> traffic is modified due to people taking alternate routes -> new traffic emerges. So my question is: what is Google Maps traffic optimizing for? The best traffic experience for User 3982274, or the best traffic…

When User 3982274 is on a busy road using the app, Google optimizes for that user's experience. If every user on that road is using the app at the same time, these algorithms should theoretically result in the optimal condition you described above. For example, if there are two roads leading up to the destination, one at 100% capacity and the other at 0%. The app will start routing people from road 1 to road 2. When…

This is the correct answer. Optimizing for the user is optimizing for the conglomerate. By giving each user the path of least resistance, the highest systemic throughput is achieved.

Re: Traffic Prediction with Advanced Graph Neural Networks

#26

Post by boloust »

datameta wrote: post #22
symplee wrote: post #17

Is it just me, or is it a little disingenuous to write the claim "...improve the accuracy of real time ETAs by up to 50% in places like Berlin, Jakarta, São Paulo, Sydney, Tokyo, and Washington D.C." When the actual numbers listed for those cities are: Berlin - 21% Jakarta - 22% São Paulo - 23% Sydney - 43% Tokyo - not listed Washington D.C. - 29%

It gets even more muddled when you consider they mention the following: "While Google Maps’ predictive ETAs have been consistently accurate for over 97% of trips, we worked with the team to minimise the remaining inaccuracies even further - sometimes by more than 50% in cities like Taichung." Are they essentially saying that they lowered 3% inaccuracy to ~1.5% in Taichung? (And nevermind the fact that 51% is describe…

It could mean that 97% of the trips had an predictive error of, say, 2 minutes, while the remaining 3% had an error of, say, 10 minutes, and they reduced that error to 5 minutes.

Re: Traffic Prediction with Advanced Graph Neural Networks

#27

Post by dqpb »

2bitencryption wrote: post #5

Google Map's traffic prediction has always led me to a very curious question: Clearly Google Maps has the ability to turn into a feedback loop. Traffic exists -> people use Google Maps to find better routes -> traffic is modified due to people taking alternate routes -> new traffic emerges. So my question is: what is Google Maps traffic optimizing for? The best traffic experience for User 3982274, or the best traffic…

I assume at some point Google will just start offering me a list of Pareto efficient activities that maximize the global action-value function, economic nirvana will be achieved for all of humanity, Sergey Brin will ascend beyond the physical plane, and then maybe, hopefully, they can stop serving me ads.

Re: Traffic Prediction with Advanced Graph Neural Networks

#28

Post by mc32 »

frogblast wrote: post #23
paxys wrote: post #7

Earlier quoted context omitted.

When User 3982274 is on a busy road using the app, Google optimizes for that user's experience. If every user on that road is using the app at the same time, these algorithms should theoretically result in the optimal condition you described above. For example, if there are two roads leading up to the destination, one at 100% capacity and the other at 0%. The app will start routing people from road 1 to road 2. When…

I live in an area strongly impacted by Google making ‘individually optimal’ decisions for each driver, and actually leaving those drivers in a dramatically worse situation. I live in a rural area, between a major population center and a major resort area, with one major highway and a few small back roads that provide alternate paths for part of the highways route. Every summer weekend the highway becomes highly conge…

This happens sometimes on I80 in the Winter (usually btwn Auburn and Tahoe). Snow conditions can wreak havoc on weekends. People are routed to backroads --but not everyone has chains on (so they may have to fumble and get them on while on a road without wide shoulders to pull over), or clearance that will enable to take the back roads --so they get stuck in even bigger snarl.

It's evident they (G) need to work with Traffic Engineers and not cowboy it.

Re: Traffic Prediction with Advanced Graph Neural Networks

#29

Post by namelosw »

2bitencryption wrote: post #5

Google Map's traffic prediction has always led me to a very curious question: Clearly Google Maps has the ability to turn into a feedback loop. Traffic exists -> people use Google Maps to find better routes -> traffic is modified due to people taking alternate routes -> new traffic emerges. So my question is: what is Google Maps traffic optimizing for? The best traffic experience for User 3982274, or the best traffic…

> in 30 years, if all cars are self-driving and self-navigating via systems like Google Maps, what is the system optimizing for?

If most users are connected to the same system, an obvious direction would be optimizing globally - if there are two routes to go, just load balance them.

I live in Beijing and the traffic is horrible sometimes. The Uber counterpart Didi mandates the routes, and sometimes counterintuitively nice - it seems to be a detour in a narrow valley but it's faster because there is no traffic jam there.

I'm not sure Uber or Didi is doing this already. At the end of the day, if most vehicles' GPS is connected to a single system, while the system is recommending routes to most users. Then it would be possible for the system to optimize for the whole population, rather than being greedy for individuals and create traffic problems.

Re: Traffic Prediction with Advanced Graph Neural Networks

#30

Post by nur0n »

2bitencryption wrote: post #5

Google Map's traffic prediction has always led me to a very curious question: Clearly Google Maps has the ability to turn into a feedback loop. Traffic exists -> people use Google Maps to find better routes -> traffic is modified due to people taking alternate routes -> new traffic emerges. So my question is: what is Google Maps traffic optimizing for? The best traffic experience for User 3982274, or the best traffic…

People seem to focus on the volume of cars. But traffic is not just a matter of volume but of friction between cars. Reducing the friction (naively) appears to be a simpler problem. You wouldn't even need full self driving, only enough tech for cars to merge/switch lanes without slowing down.

Not requiring a central point of control is an additional benefit. The reduction of traffic would be an "emergent" behavior.

Post reply on HN