Horizon: Facebook’s Open Source Applied Reinforcement Learning Platform(research.fb.com)
research.fb.com
Horizon: Facebook’s Open Source Applied Reinforcement Learning Platform
https://research.fb.com/publications/horizon-facebooks-open-source-applied-reinforcement-learning-platform/
21 comments
This kind of paper which describes a system is not uncommon in computer science. It's a way for future papers which use the system to cite it.
For example the scikit learn paper: http://www.jmlr.org/papers/volume12/pedregosa11a/pedregosa11...
For example the scikit learn paper: http://www.jmlr.org/papers/volume12/pedregosa11a/pedregosa11...
Indeed! This is an endemic issue in the computer sciences.
I don't think it's a bad way to approach knowledge dissemination by the way, it is however indicative of the problem of reprouducability and explainability in AI broadly.
I bring this point up simply to be another voice stating that we need more rigorous methodology in AI research if we are going to make advances that are focused first on knowledge, rather than primarily applicability of technology.
Way back before AI was cool, there was a good paper on this [1] that is very relevant to today.
Quoting from the abstract:
"There are two central problems concerning the methodology and foundations of Artificial Intelligence (AI). One is to find a technique for defining problems in AI. The other is to find a technique for testing hypotheses in AI. There are, as of now, no solutions to these two problems. The former problem has been neglected because researchers have found it difficult to define AI problems in a traditional manner. The second problem has not been tackled seriously, with the result that supposedly competing AI hypotheses are typically non-comparable."
[1]https://link.springer.com/chapter/10.1007/978-1-4471-3542-5_...
I don't think it's a bad way to approach knowledge dissemination by the way, it is however indicative of the problem of reprouducability and explainability in AI broadly.
I bring this point up simply to be another voice stating that we need more rigorous methodology in AI research if we are going to make advances that are focused first on knowledge, rather than primarily applicability of technology.
Way back before AI was cool, there was a good paper on this [1] that is very relevant to today.
Quoting from the abstract:
"There are two central problems concerning the methodology and foundations of Artificial Intelligence (AI). One is to find a technique for defining problems in AI. The other is to find a technique for testing hypotheses in AI. There are, as of now, no solutions to these two problems. The former problem has been neglected because researchers have found it difficult to define AI problems in a traditional manner. The second problem has not been tackled seriously, with the result that supposedly competing AI hypotheses are typically non-comparable."
[1]https://link.springer.com/chapter/10.1007/978-1-4471-3542-5_...
If they want to increase adoption, they really need to make this stuff easier to install. I mean zero friction. Either pip, or an Ubuntu PPA that "just works".
Caffe2 install page: "We only support Anaconda packages at the moment. If you do not wish to use Anaconda, then you must build Caffe2 from source." => We are a company with a 400B+ market cap but are too lazy to support more than one installation configuration. Good luck dealing with dependency hell, poor ML grad student researcher.
MXnet install page: "You can either upgrade your CUDA install or install the MXNet package that supports your CUDA version." => We welcome you with open arms regardless of your configuration! No matter your configuration we have an pre-built package for you!
Caffe2 install page: "We only support Anaconda packages at the moment. If you do not wish to use Anaconda, then you must build Caffe2 from source." => We are a company with a 400B+ market cap but are too lazy to support more than one installation configuration. Good luck dealing with dependency hell, poor ML grad student researcher.
MXnet install page: "You can either upgrade your CUDA install or install the MXNet package that supports your CUDA version." => We welcome you with open arms regardless of your configuration! No matter your configuration we have an pre-built package for you!
Hey! Have you tried our docker install? We tried to make it as simple as possible.
I agree that installing without docker is a pain, but you should follow our install guide. Particularly for Caffe2, it's included in PyTorch 1.0 so you don't have to install it separately :-).
I agree that installing without docker is a pain, but you should follow our install guide. Particularly for Caffe2, it's included in PyTorch 1.0 so you don't have to install it separately :-).
Github repo: https://github.com/facebookresearch/Horizon
Ah, it has the fabled BSD+ license: https://github.com/facebookresearch/Horizon/blob/master/PATE...
I guess that's not too unusual from their opensource contributions at this point.
I guess that's not too unusual from their opensource contributions at this point.
Hey! Thanks for your feedback. I have to admit that I'm a license noob and picked an option basically at random. I am changing Horizon to a BSD license (no PATENTS part).
Wow, that's... fantastic and generous, I wasn't expecting this response. The change makes this framework so much more inviting to use.
(I'm also surprised Facebook's legal / open source policy would allow such flexibility around licenses.)
(I'm also surprised Facebook's legal / open source policy would allow such flexibility around licenses.)
It’s terrible — they’ve removed it for other projects when put under pressure (e.g. React IIRC).
One of the most interesting part of the paper is how RL is used - especially here for Horizon where one of the goal seems to be problems where simulation isn't available. One such problem is push notifications:
Historically, we have used supervised learning models for predicting click through rate (CTR) and likelihood that the notification leads to meaningful interactions.
We introduced a new policy that uses Horizon to train a Discrete-Action DQN model for sending push notifications to address the problems above. The Markov Decision Process (MDP) is based on a sequence of notification candidates for a particular person. The actions here are sending and dropping the notification, and the state describes a set of features about the person and the notification candidate. There are rewards for interactions and activity on Facebook, with a penalty for sending the notification to control the volume of notifications sent. The policy optimizes for the long term value and is able to capture incremental effects of sending the notification by comparing the Q-values of the send and don’t send action.
Historically, we have used supervised learning models for predicting click through rate (CTR) and likelihood that the notification leads to meaningful interactions.
We introduced a new policy that uses Horizon to train a Discrete-Action DQN model for sending push notifications to address the problems above. The Markov Decision Process (MDP) is based on a sequence of notification candidates for a particular person. The actions here are sending and dropping the notification, and the state describes a set of features about the person and the notification candidate. There are rewards for interactions and activity on Facebook, with a penalty for sending the notification to control the volume of notifications sent. The policy optimizes for the long term value and is able to capture incremental effects of sending the notification by comparing the Q-values of the send and don’t send action.
This kind of AI applied at scale (Facebook) scares me. Mainly because humans are at the other side of the feedback loop, and not only is the AI adapting, but people are adapting too. Over time this kind of runaway feedback loop could lead into anything.
How does that differ substantially from humans interacting with the environment or other humans?
Minimally: rate of change > capacity to adapt, and lack of natural predators, right?
In case you want to see the actual paper: https://research.fb.com/wp-content/uploads/2018/10/Horizon-F...?
Thanks for the link!
I recommend that anybody who is interested in this read section 9 in this paper: "NOTIFICATIONS AT FACEBOOK". It brings into focus real ways that this technology is used and is useful.
I recommend that anybody who is interested in this read section 9 in this paper: "NOTIFICATIONS AT FACEBOOK". It brings into focus real ways that this technology is used and is useful.
So cryptic and hand wavy description in paper but hey I'm glad they have good results. Going from pedals to combustion engines clearly has efficiency benefits but they gloss over the real flesh of the fruit when they don't describe how their particular engine is better or tuned. Wish they had more talk of notifications as Markov chains and how they evaluate interaction
Discussion on Google's Dopamine - its Reinforcement Learning Framework https://news.ycombinator.com/item?id=15648746
Wow.... They actually call it dopamine? They're shameless.
That discussion is on Dopamine the startup, not the framework from Google.
Google's framework is at https://github.com/google/dopamine and I don't believe it's generated discussion on HN before.
Google's framework is at https://github.com/google/dopamine and I don't believe it's generated discussion on HN before.
Blog post: https://code.fb.com/ml-applications/horizon/ (would be a better link if that can be changed)
What this is, is documentation of how Facebook implemented a technology stack that uses reinforcement learning techniques to do something. Namely: "Notifications at Facebook"
So what can other developers and business owners take from this? I don't see anything about the down stream product impacts. Does it impact conversion to paid rate for users? Does it reduce human labor? How does it improve benefits to users. All I see them write are two things:
"We observed a significant improvement in activity and mean- ingful interactions by deploying an RL based policy for certain types of notifications, replacing the previous system based on supervised learning."
I'm sorry but there is absolutely nothing rigorous in that statement. How are "meaningful interactions" defined? Hopefully they aren't still arguing the formula (more interaction = makes users better off).
"After deploying the DQN model, we were able to improve daily, weekly, and monthly metrics without sacrificing notification quality."
Improve for who? Well obviously Facebook and how much activity people have. Not necessarily if the user is actually getting more value from it.
What's the Return on Investment for this system?
Listen, I'm a huge fan of being open with business practices, research etc...I'm also obsessive about RL and making progress in the field.
What I can't stand however is lack of rigorous and tangible proof of how we're making things better for users or the society broadly with RL yet, or even in most cases getting positive ROI for the effort we're putting into ML/DL.
I've built these tools at scale so it hurts to say this, but the economics just aren't yet lining up here across the entire ML/DL industry and that has me worried that another AI winter is coming.
[1]https://www.biorxiv.org/content/early/2018/11/01/422345