Loading ...
Sorry, an error occurred while loading the content.

Neuroevolution in the Age of Big Data and the Internet?

Expand Messages
  • Wesley Tansey
    Hi all, I d like to start a discussion on where NE fits into the modern world. We ve begun entering the age of Big Data -- 200M tweets/day, 700M Facebook
    Message 1 of 6 , Jul 1, 2011
    • 0 Attachment
      Hi all,

      I'd like to start a discussion on where NE fits into the modern world. We've begun entering the age of "Big Data"-- 200M tweets/day, 700M Facebook users, etc. We're also finding ourselves more immersed in computing technology as mobile phones become more ubiquitous. 

      I'm interested in your thoughts. Specifically:

      - Where does Neuroevolution hold the most potential for internet applications?

      - What kind of data will NE be best suited to leverage?

      - How does the ubiquity of mobile technology help NE-based applications?

      - As we now have over a billion people online, what kind of large-scale human feedback systems may be devised that rely on NE?

      - What do you generally see as the future of NE for consumer apps?

      I have my own thoughts on these, which I'm happy to share, but I'm really interested in what the community thinks. As the world has changed drastically even since the first NEAT paper was published, keeping up with the real world necessitates us periodically re-examining the practical application space for our research area. Other areas seem to have flourished in the new  world: NLP is regularly used to measure consumer sentiment in social media, graph analysis algorithms are used heavily at both Google and Facebook to determine which page to show or which friend's posts to display in your feed. Where does NE fit in here?


      Wesley
    • Ghulam Mubashar Hassan
      Hi, These are the good questions posted by Wesley. I am feeling them nowadays with NEAT. I am new to this but found that NE works well with small amount of
      Message 2 of 6 , Jul 2, 2011
      • 0 Attachment
        Hi,
         
        These are the good questions posted by Wesley. I am feeling them nowadays with NEAT. I am new to this but found that NE works well with small amount of data. I am trying to make it work with large amount of data from real world for image recognition and it is giving me tough time.
        I hope that everyone will be sharing his/her experiences.
         

         
        Ghulam Mubashar Jally.



        From: Wesley Tansey <tansey@...>
        To: neat@yahoogroups.com
        Sent: Sat, July 2, 2011 2:19:40 AM
        Subject: [neat] Neuroevolution in the Age of Big Data and the Internet?

         

        Hi all,


        I'd like to start a discussion on where NE fits into the modern world. We've begun entering the age of "Big Data"-- 200M tweets/day, 700M Facebook users, etc. We're also finding ourselves more immersed in computing technology as mobile phones become more ubiquitous. 

        I'm interested in your thoughts. Specifically:

        - Where does Neuroevolution hold the most potential for internet applications?

        - What kind of data will NE be best suited to leverage?

        - How does the ubiquity of mobile technology help NE-based applications?

        - As we now have over a billion people online, what kind of large-scale human feedback systems may be devised that rely on NE?

        - What do you generally see as the future of NE for consumer apps?

        I have my own thoughts on these, which I'm happy to share, but I'm really interested in what the community thinks. As the world has changed drastically even since the first NEAT paper was published, keeping up with the real world necessitates us periodically re-examining the practical application space for our research area. Other areas seem to have flourished in the new  world: NLP is regularly used to measure consumer sentiment in social media, graph analysis algorithms are used heavily at both Google and Facebook to determine which page to show or which friend's posts to display in your feed. Where does NE fit in here?


        Wesley
      • Colin Green
        Possibly not what you are looking for but one idea I have on the back burner is to evolve better ANN models, by which I mean the organisational primitives,
        Message 3 of 6 , Jul 2, 2011
        • 0 Attachment
          Possibly not what you are looking for but one idea I have on the back
          burner is to evolve better ANN models, by which I mean the
          organisational primitives, update rules, etc. rather than evolving
          topologies and weights for the traditional connection weights and
          activation function based models.

          I filled the idea out a bit more here:

          http://the-locster.livejournal.com/107223.html

          Colin.
        • Ken
          Hi Wesley, that s an interesting discussion topic. My take is that NE can play a constructive role in this big-data/many-users type of world. Or perhaps it
          Message 4 of 6 , Jul 3, 2011
          • 0 Attachment
            Hi Wesley, that's an interesting discussion topic. My take is that NE can play a constructive role in this big-data/many-users type of world. Or perhaps it might be more accurate to say that big-data/many-users can play a constructive role in NE. In any case, I think we're already seeing examples of this potential synergy in the form of crowd-sourced evolution. These examples include Picbreeder, Galactic Arms Race, and EndlessForms. While these examples do not rival the impact of something like big data in language translation, when you consider the resources/funding invested in something like Google, Picbreeder gets a lot of bang for its buck and really does leverage crowds in a unique way. The 8,000+ image phylogeny on Picbreeder is a great collection of evolutionary data whose study can lead to further innovations. In fact, novelty search was invented based on observations of the process of discovery on Picbreeder. There is more to be found there still, and crowd-sourced evolution remains a wide-open area with all kinds of unrealized possibilities

            GAR also shows that it can be implicit, so users don't even have to know what's going on. In a few days I'm actually heading off to an invited workshop on "crowd sourcing of research" and "scientific discovery games," where NE is standing alongside the various innovative ideas in crowdsourcing of recent years.

            Still, Ghulam makes the point that "big data" perhaps is still a challenge when we're talking about something like having many inputs into a neural network. That's a different issue for NE, but NE is taking steps in that direction with methods like HyperNEAT that don't really care how many inputs you give them. We still face a challenge in the massive CPU utilization that evolving populations of multimillion-connection networks requires, but perhaps the networked world offers or will offer novel opportunities for parallelization that can address the voracious need for CPU cycles in NE.

            At the same time, while it's true that there has been radical progress in e.g. foreign language translation, the promise of big data is sometimes over-hyped. For example, my Amazon recommendations are as uninspiring as they were 10 years ago. And the stuff Netflix suggests that I will like is usually somewhere between obvious and hilarious. Data-mining at its most grandiose rests on a kind of faith that there is some secret "truth" hiding deep in the masses of data that will reveal something so subtle about me and you (and humanity) that we never even knew it. While I'm sure that buying habits are helpful to analyze for companies at the margins, it's unclear that it is always an AI revolution. It's true there are 700M Facebook users, but you might look at that as a commercial opportunity more than an AI opportunity (crowd-sourced evolution and a few other application excepted).

            After all, let's not forget that one of the inspirations for NE is the evolution of brains, which predated all these buzzwords by millions of years. To some extent, the promise of NE is thus independent of the modern world even as it is realized within it. If we could learn to evolve brains, it won't be a question of how we fit in with the technologies of today, but how they fit in with us. Of course, that is more at a Holy Grail level, but still important to keep in mind as we watch modern technology evolve.

            However, I don't want to dismiss the idea that all these things can synergize, because they can. Like I said crowd-sourcing is a huge untapped opportunity for NE and the commercial opportunities presented by the modern world (e.g. with millions of users and mobile devices) are exciting in their own right, regardless of their centrality to AI. We are definitely going to see NE exploiting these opportunities. As NE methods improve, the potential for them to be commercially relevant increases. For example, improvements in representation open up whole new applications, like our work in music evolution. NE has the current disadvantage of significantly less investment compared to other AI technologies, but it has the advantage of the element of surprise.

            ken


            --- In neat@yahoogroups.com, Wesley Tansey <tansey@...> wrote:
            >
            > Hi all,
            >
            > I'd like to start a discussion on where NE fits into the modern world. We've
            > begun entering the age of "Big Data"-- 200M tweets/day, 700M Facebook users,
            > etc. We're also finding ourselves more immersed in computing technology as
            > mobile phones become more ubiquitous.
            >
            > I'm interested in your thoughts. Specifically:
            >
            > - Where does Neuroevolution hold the most potential for internet
            > applications?
            >
            > - What kind of data will NE be best suited to leverage?
            >
            > - How does the ubiquity of mobile technology help NE-based applications?
            >
            > - As we now have over a billion people online, what kind of large-scale
            > human feedback systems may be devised that rely on NE?
            >
            > - What do you generally see as the future of NE for consumer apps?
            >
            > I have my own thoughts on these, which I'm happy to share, but I'm really
            > interested in what the community thinks. As the world has changed
            > drastically even since the first NEAT paper was published, keeping up with
            > the real world necessitates us periodically re-examining the practical
            > application space for our research area. Other areas seem to have flourished
            > in the new world: NLP is regularly used to measure consumer sentiment in
            > social media, graph analysis algorithms are used heavily at both Google and
            > Facebook to determine which page to show or which friend's posts to display
            > in your feed. Where does NE fit in here?
            >
            >
            > Wesley
            >
          • Wesley Tansey
            Hi Ken, Thanks for the detailed response. If I may summarize your points, it seems in essence you see a few areas ripe for exploration: 1. Crowd-sourced
            Message 5 of 6 , Jul 4, 2011
            • 0 Attachment
              Hi Ken,

              Thanks for the detailed response. 

              If I may summarize your points, it seems in essence you see a few areas ripe for exploration:

              1. Crowd-sourced evolution.

              2. Games, either for opponent AI or some other component such as weapons in GAR.

              3. New representational enhancements.

              I believe these insights touch on a lot of the potential for NE in the next decade.

              I'll note two caveats to your points. First, I would certainly not limit the modern applications of AI in the big-data/many-users internet to simply translation and recommender systems. In fact, I would contend that a main area of application is in sentiment analysis on social networks like Facebook and Twitter. For instance, the guys at Dataminr gave an interesting talk at the Twitter developer conference where they showed how their system knew (and was able to alert oil traders) that Osama bin Laden had been killed 20 minutes before any news organization had reported it or the market had absorbed it. Understanding what people are saying collectively in order to make inferences about changes in the state of the world is quite powerful and revolutionary.

              Second, while I do appreciate the ultimate goal of evolving brains, I think it's crucial to ensure the smoothness of the landscape that NE research is exploring. If human brains are a global optimum (or at least a local one much higher than we currently have discovered), we are unlikely to discover it unless there is some gradually increasing path of rewards that leads us there. Scientific curiosity aside, one of the most rewarding reasons for pursuing this goal is the power an artificial brain has to better the world by being integrated into large-scale systems. Thus, as modern systems progressively grow more complex, it's important that we understand how to leverage the increase in information and ability, otherwise we may be unnecessarily limiting ourselves to a spikey landscape.

              I do agree on the fundamental insights about interactive evolution being a largely untapped opportunity. Here are a few ideas I came up with this morning that may be interesting and relevant:

              1. Instagram meets Picbreeder: Interactive evolution of image filters.

              2. Turntable.fm meets FSMC: Bringing interactive music evolution to a social setting. Enabling people to DJ their custom music or create it with a group.

              3. Readability/Diffbot via HyperNEAT: Looking at a webpage and identifying the article text (as opposed to ads, menus, etc.) seems very similar to the box discrimination task.

              In a more grandiose application: a commodity-style web service layer, similar to Amazon's AWS (EC2 hosting, S3 storage, etc), could be created around the algorithms to foster more wide-scale adoption. Right now we basically have the situation where thousands of developers are taking "Machine Learning" courses to learn how to use map-reduce to create basic correlation models and classifiers on big data. There is no real API solution for reinforcement learning without requiring the user to understand the difference between two esoteric algorithms. Google has started entering this area with the Prediction API, but it focuses mostly on recommendations and translation. A big advantage for NE as a service may be in areas such as mobile development, where battery life and computing power constraints make simulated evolution impractical, but evaluating a candidate model is feasible. Separately, if we are to reach millions of connections then a large-scale infrastructure will likely be required.

              New representations are interesting and it's an area I hadn't considered. Are there any specific kinds of representation improvements you had in mind? What kind of applications do you see as currently out of reach for NE that may be opened by an innovative representation?


              Wesley


              On Sun, Jul 3, 2011 at 7:13 AM, Ken <kstanley@...> wrote:
               



              Hi Wesley, that's an interesting discussion topic. My take is that NE can play a constructive role in this big-data/many-users type of world. Or perhaps it might be more accurate to say that big-data/many-users can play a constructive role in NE. In any case, I think we're already seeing examples of this potential synergy in the form of crowd-sourced evolution. These examples include Picbreeder, Galactic Arms Race, and EndlessForms. While these examples do not rival the impact of something like big data in language translation, when you consider the resources/funding invested in something like Google, Picbreeder gets a lot of bang for its buck and really does leverage crowds in a unique way. The 8,000+ image phylogeny on Picbreeder is a great collection of evolutionary data whose study can lead to further innovations. In fact, novelty search was invented based on observations of the process of discovery on Picbreeder. There is more to be found there still, and crowd-so! urced evolution remains a wide-open area with all kinds of unrealized possibilities

              GAR also shows that it can be implicit, so users don't even have to know what's going on. In a few days I'm actually heading off to an invited workshop on "crowd sourcing of research" and "scientific discovery games," where NE is standing alongside the various innovative ideas in crowdsourcing of recent years.

              Still, Ghulam makes the point that "big data" perhaps is still a challenge when we're talking about something like having many inputs into a neural network. That's a different issue for NE, but NE is taking steps in that direction with methods like HyperNEAT that don't really care how many inputs you give them. We still face a challenge in the massive CPU utilization that evolving populations of multimillion-connection networks requires, but perhaps the networked world offers or will offer novel opportunities for parallelization that can address the voracious need for CPU cycles in NE.

              At the same time, while it's true that there has been radical progress in e.g. foreign language translation, the promise of big data is sometimes over-hyped. For example, my Amazon recommendations are as uninspiring as they were 10 years ago. And the stuff Netflix suggests that I will like is usually somewhere between obvious and hilarious. Data-mining at its most grandiose rests on a kind of faith that there is some secret "truth" hiding deep in the masses of data that will reveal something so subtle about me and you (and humanity) that we never even knew it. While I'm sure that buying habits are helpful to analyze for companies at the margins, it's unclear that it is always an AI revolution. It's true there are 700M Facebook users, but you might look at that as a commercial opportunity more than an AI opportunity (crowd-sourced evolution and a few other application excepted).

              After all, let's not forget that one of the inspirations for NE is the evolution of brains, which predated all these buzzwords by millions of years. To some extent, the promise of NE is thus independent of the modern world even as it is realized within it. If we could learn to evolve brains, it won't be a question of how we fit in with the technologies of today, but how they fit in with us. Of course, that is more at a Holy Grail level, but still important to keep in mind as we watch modern technology evolve.

              However, I don't want to dismiss the idea that all these things can synergize, because they can. Like I said crowd-sourcing is a huge untapped opportunity for NE and the commercial opportunities presented by the modern world (e.g. with millions of users and mobile devices) are exciting in their own right, regardless of their centrality to AI. We are definitely going to see NE exploiting these opportunities. As NE methods improve, the potential for them to be commercially relevant increases. For example, improvements in representation open up whole new applications, like our work in music evolution. NE has the current disadvantage of significantly less investment compared to other AI technologies, but it has the advantage of the element of surprise.

              ken



              --- In neat@yahoogroups.com, Wesley Tansey <tansey@...> wrote:
              >
              > Hi all,
              >
              > I'd like to start a discussion on where NE fits into the modern world. We've
              > begun entering the age of "Big Data"-- 200M tweets/day, 700M Facebook users,
              > etc. We're also finding ourselves more immersed in computing technology as
              > mobile phones become more ubiquitous.
              >
              > I'm interested in your thoughts. Specifically:
              >
              > - Where does Neuroevolution hold the most potential for internet
              > applications?
              >
              > - What kind of data will NE be best suited to leverage?
              >
              > - How does the ubiquity of mobile technology help NE-based applications?
              >
              > - As we now have over a billion people online, what kind of large-scale
              > human feedback systems may be devised that rely on NE?
              >
              > - What do you generally see as the future of NE for consumer apps?
              >
              > I have my own thoughts on these, which I'm happy to share, but I'm really
              > interested in what the community thinks. As the world has changed
              > drastically even since the first NEAT paper was published, keeping up with
              > the real world necessitates us periodically re-examining the practical
              > application space for our research area. Other areas seem to have flourished
              > in the new world: NLP is regularly used to measure consumer sentiment in
              > social media, graph analysis algorithms are used heavily at both Google and
              > Facebook to determine which page to show or which friend's posts to display
              > in your feed. Where does NE fit in here?
              >
              >
              > Wesley
              >


            • Ken
              Hi Wesley, I m finally getting back to this thread after a couple weeks of travel including GECCO. You raise several good suggestions and the broader idea of
              Message 6 of 6 , Jul 21, 2011
              • 0 Attachment
                Hi Wesley, I'm finally getting back to this thread after a couple weeks of travel including GECCO. You raise several good suggestions and the broader idea of some kind of neuroevolution web service is also intriguing. It does seem that the field will gradually move towards greater public accessibility. I agree that we cannot measure all our pursuits directly by whether they will lead to a human brain and hence that some intermediate milestones are important. On the other hand, as in novelty search, the impetus for some advances will be from how they diverge from past conventions as opposed to how they facilitate future goals. So to really smooth out this space will require appreciating interesting ideas in their own right rather than only for what they clearly facilitate in the here and now. Nevertheless, achieving some killer applications is always good for the field, if for no other reason than diverting resources and support our way.

                Regarding innovative representation, there is clearly room for playing with representations in body-brain evolution, such as in the work of Bongard and Auerbach:

                http://www.cs.uvm.edu/~jauerbac/publications/auerbach_bongard_gecco_2011.pdf

                The field of evolutionary robotics in general also can benefit from innovations in representation, both in morphology and in neural networks alone. Natural brains control bodies and robots today move with nothing like the elegance of natural organisms, hinting at a big near-term opportunity.

                ken

                --- In neat@yahoogroups.com, Wesley Tansey <tansey@...> wrote:
                >
                > Hi Ken,
                >
                > Thanks for the detailed response.
                >
                > If I may summarize your points, it seems in essence you see a few areas ripe
                > for exploration:
                >
                > 1. Crowd-sourced evolution.
                >
                > 2. Games, either for opponent AI or some other component such as weapons in
                > GAR.
                >
                > 3. New representational enhancements.
                >
                > I believe these insights touch on a lot of the potential for NE in the next
                > decade.
                >
                > I'll note two caveats to your points. First, I would certainly not limit the
                > modern applications of AI in the big-data/many-users internet to simply
                > translation and recommender systems. In fact, I would contend that a main
                > area of application is in sentiment analysis on social networks like
                > Facebook and Twitter. For instance, the guys at Dataminr gave an interesting
                > talk at the Twitter developer conference where they showed how their system
                > knew (and was able to alert oil traders) that Osama bin Laden had been
                > killed 20 minutes before any news organization had reported it or the market
                > had absorbed it. Understanding what people are saying collectively in order
                > to make inferences about changes in the state of the world is quite powerful
                > and revolutionary.
                >
                > Second, while I do appreciate the ultimate goal of evolving brains, I think
                > it's crucial to ensure the smoothness of the landscape that NE research is
                > exploring. If human brains are a global optimum (or at least a local one
                > much higher than we currently have discovered), we are unlikely to discover
                > it unless there is some gradually increasing path of rewards that leads us
                > there. Scientific curiosity aside, one of the most rewarding reasons for
                > pursuing this goal is the power an artificial brain has to better the world
                > by being integrated into large-scale systems. Thus, as modern systems
                > progressively grow more complex, it's important that we understand how to
                > leverage the increase in information and ability, otherwise we may be
                > unnecessarily limiting ourselves to a spikey landscape.
                >
                > I do agree on the fundamental insights about interactive evolution being a
                > largely untapped opportunity. Here are a few ideas I came up with this
                > morning that may be interesting and relevant:
                >
                > 1. Instagram meets Picbreeder: Interactive evolution of image filters.
                >
                > 2. Turntable.fm meets FSMC: Bringing interactive music evolution to a social
                > setting. Enabling people to DJ their custom music or create it with a group.
                >
                > 3. Readability/Diffbot via HyperNEAT: Looking at a webpage and identifying
                > the article text (as opposed to ads, menus, etc.) seems very similar to the
                > box discrimination task.
                >
                > In a more grandiose application: a commodity-style web service layer,
                > similar to Amazon's AWS (EC2 hosting, S3 storage, etc), could be created
                > around the algorithms to foster more wide-scale adoption. Right now we
                > basically have the situation where thousands of developers are taking
                > "Machine Learning" courses to learn how to use map-reduce to create basic
                > correlation models and classifiers on big data. There is no real API
                > solution for reinforcement learning without requiring the user to understand
                > the difference between two esoteric algorithms. Google has started entering
                > this area with the Prediction API, but it focuses mostly on recommendations
                > and translation. A big advantage for NE as a service may be in areas such as
                > mobile development, where battery life and computing power constraints make
                > simulated evolution impractical, but evaluating a candidate model is
                > feasible. Separately, if we are to reach millions of connections then a
                > large-scale infrastructure will likely be required.
                >
                > New representations are interesting and it's an area I hadn't considered.
                > Are there any specific kinds of representation improvements you had in mind?
                > What kind of applications do you see as currently out of reach for NE that
                > may be opened by an innovative representation?
                >
                >
                > Wesley
                >
                >
                > On Sun, Jul 3, 2011 at 7:13 AM, Ken <kstanley@...> wrote:
                >
                > > **
                > >
                > >
                > >
                > >
                > > Hi Wesley, that's an interesting discussion topic. My take is that NE can
                > > play a constructive role in this big-data/many-users type of world. Or
                > > perhaps it might be more accurate to say that big-data/many-users can play a
                > > constructive role in NE. In any case, I think we're already seeing examples
                > > of this potential synergy in the form of crowd-sourced evolution. These
                > > examples include Picbreeder, Galactic Arms Race, and EndlessForms. While
                > > these examples do not rival the impact of something like big data in
                > > language translation, when you consider the resources/funding invested in
                > > something like Google, Picbreeder gets a lot of bang for its buck and really
                > > does leverage crowds in a unique way. The 8,000+ image phylogeny on
                > > Picbreeder is a great collection of evolutionary data whose study can lead
                > > to further innovations. In fact, novelty search was invented based on
                > > observations of the process of discovery on Picbreeder. There is more to be
                > > found there still, and crowd-so! urced evolution remains a wide-open area
                > > with all kinds of unrealized possibilities
                > >
                > > GAR also shows that it can be implicit, so users don't even have to know
                > > what's going on. In a few days I'm actually heading off to an invited
                > > workshop on "crowd sourcing of research" and "scientific discovery games,"
                > > where NE is standing alongside the various innovative ideas in crowdsourcing
                > > of recent years.
                > >
                > > Still, Ghulam makes the point that "big data" perhaps is still a challenge
                > > when we're talking about something like having many inputs into a neural
                > > network. That's a different issue for NE, but NE is taking steps in that
                > > direction with methods like HyperNEAT that don't really care how many inputs
                > > you give them. We still face a challenge in the massive CPU utilization that
                > > evolving populations of multimillion-connection networks requires, but
                > > perhaps the networked world offers or will offer novel opportunities for
                > > parallelization that can address the voracious need for CPU cycles in NE.
                > >
                > > At the same time, while it's true that there has been radical progress in
                > > e.g. foreign language translation, the promise of big data is sometimes
                > > over-hyped. For example, my Amazon recommendations are as uninspiring as
                > > they were 10 years ago. And the stuff Netflix suggests that I will like is
                > > usually somewhere between obvious and hilarious. Data-mining at its most
                > > grandiose rests on a kind of faith that there is some secret "truth" hiding
                > > deep in the masses of data that will reveal something so subtle about me and
                > > you (and humanity) that we never even knew it. While I'm sure that buying
                > > habits are helpful to analyze for companies at the margins, it's unclear
                > > that it is always an AI revolution. It's true there are 700M Facebook users,
                > > but you might look at that as a commercial opportunity more than an AI
                > > opportunity (crowd-sourced evolution and a few other application excepted).
                > >
                > > After all, let's not forget that one of the inspirations for NE is the
                > > evolution of brains, which predated all these buzzwords by millions of
                > > years. To some extent, the promise of NE is thus independent of the modern
                > > world even as it is realized within it. If we could learn to evolve brains,
                > > it won't be a question of how we fit in with the technologies of today, but
                > > how they fit in with us. Of course, that is more at a Holy Grail level, but
                > > still important to keep in mind as we watch modern technology evolve.
                > >
                > > However, I don't want to dismiss the idea that all these things can
                > > synergize, because they can. Like I said crowd-sourcing is a huge untapped
                > > opportunity for NE and the commercial opportunities presented by the modern
                > > world (e.g. with millions of users and mobile devices) are exciting in their
                > > own right, regardless of their centrality to AI. We are definitely going to
                > > see NE exploiting these opportunities. As NE methods improve, the potential
                > > for them to be commercially relevant increases. For example, improvements in
                > > representation open up whole new applications, like our work in music
                > > evolution. NE has the current disadvantage of significantly less investment
                > > compared to other AI technologies, but it has the advantage of the element
                > > of surprise.
                > >
                > > ken
                > >
                > >
                > > --- In neat@yahoogroups.com, Wesley Tansey <tansey@> wrote:
                > > >
                > > > Hi all,
                > > >
                > > > I'd like to start a discussion on where NE fits into the modern world.
                > > We've
                > > > begun entering the age of "Big Data"-- 200M tweets/day, 700M Facebook
                > > users,
                > > > etc. We're also finding ourselves more immersed in computing technology
                > > as
                > > > mobile phones become more ubiquitous.
                > > >
                > > > I'm interested in your thoughts. Specifically:
                > > >
                > > > - Where does Neuroevolution hold the most potential for internet
                > > > applications?
                > > >
                > > > - What kind of data will NE be best suited to leverage?
                > > >
                > > > - How does the ubiquity of mobile technology help NE-based applications?
                > > >
                > > > - As we now have over a billion people online, what kind of large-scale
                > > > human feedback systems may be devised that rely on NE?
                > > >
                > > > - What do you generally see as the future of NE for consumer apps?
                > > >
                > > > I have my own thoughts on these, which I'm happy to share, but I'm really
                > > > interested in what the community thinks. As the world has changed
                > > > drastically even since the first NEAT paper was published, keeping up
                > > with
                > > > the real world necessitates us periodically re-examining the practical
                > > > application space for our research area. Other areas seem to have
                > > flourished
                > > > in the new world: NLP is regularly used to measure consumer sentiment in
                > > > social media, graph analysis algorithms are used heavily at both Google
                > > and
                > > > Facebook to determine which page to show or which friend's posts to
                > > display
                > > > in your feed. Where does NE fit in here?
                > > >
                > > >
                > > > Wesley
                > > >
                > >
                > >
                > >
                >
              Your message has been successfully submitted and would be delivered to recipients shortly.