ibminsight

We are going to be come more and more dependent upon this..as data sets continue to grow with data coming from all things connected

not a popular question? :)

@theRab I think there is still a challenge with this can you trust it...what are your thoughts?

I really think #deepLearning will be the next extension of #deepAnalytics; we are in the golden age of #AI

A3 there's a lot of hype around deep learning, but underneath there's a lot of substance to it. deep learning is basically machine perception.

.@ibminsight A3 DL can basically help machines recognize patterns in raw sensory data. therefore, perception.

Ok Q3, what IS THE HYPE? What is it that people are believing? :)

.@ibminsight so, images, sound, text, time series data from sensors or economic tables, etc.

A3 As the data becomes more useable the ways that you can derive value also increase, so the incentives for deep learning grow. A virtuous adoption cycle.

.@ibminsight the more data you have, the better DL can perform.

I think we will also learn at levels we have never imagined...molecular, atomic, particle....we have human injectible machines already #reality

Deep learning, leveraging artificial neural networks and streaming media, is key to pattern-based sensemaking and predictive analysis in the era of all online media & entertainment, education, training, surveillance, etc.

A3 Look to Wall Street as example. When all data could be mined and analyzed all sorts of quants and hedges became possible. First the data, then the deep learning, then monetization

@topher920 bringing up trust is a good point. you should trust anything you can't test. but you can test deep learning algorithms, daily and continuously.

Deep learning algorithmic models are the essential component of artificial intelligence for video, voice, face, gesture, and other real-time recognition apps.

.@ibminsight deep learning isn't all of AI, though. i think we should be clear that there are other powerful algorithms out there, too.

I'm wondering if the data transfer volumes implied here are going to break the back of the existing internet.

.@ibminsight reinforcement learning helps agents navigate uncertain environments to maximize rewards.

genetic algorithms can be powerful ways to recognize patterns.

Deep learning is the foundation of computer vision, range-finding, collision avoidance, and other apps without which self-driven, automated, connected vehicles would be impossible.

More data doesn't necessarily mean better DL. If you're data are not appropriate for the problem at hand, are invalid..., then your prediction suffers.

.@ibminsight and traditional ML algorithms still need to be tacked onto to the end of a deep neural network in order to make predictions and classifications.

old standbys like regression, logistic regression etc. are still really powerful and can be used with neural networks.

because you can feed the output of the neural net into the other algorithms you might be familiar with. DL is just automatically extracting the features that used to take months and years to engineer.

The "hype" around deep learning isn't hype (i.e., exaggeration). The substance is real and it's fulfilling its promise in this new era of cognitive computing, which relies on a new generation of AI that auto-learns from big data.

@bobehayes sure. it's all ceteris paribus. if you have usable data, then having more usable data will make for better DL. if you don't have usable data, no amount of ML will produce valuable insights.

@deeplearning4j Do you think we need a new way of UX with AI to help with checking understanding or do you think what we have today can achieve that?

@SystemsandTech I think there are a lot of people in the world who fear AI and believe an intelligent computer could take over the world in our lifetimes, just look at some of the most recent blockbusters like Chappie and Ex Machina...

while DL is powerful, to avoid suspicions of hype we should recognize the usefulness of other algorithms.

@deeplearning4j Yes.

When you see DeepMind combine neural networks with reinforcement learning, you realize that one of the killer apps of DL is simply feeding vectors to other useful algorithms.

@topher920 the results that neural networks produce are transparent, but the way they arrive at their decisions is not. we can check their accuracy, but not necessarily their process. work is being done there to give them more explanatory power.

Ok, I see on #AI and Deep Learning fears, it's abt intrusiveness and Privacy

@topher920 as in: we can test their predictions and therefore believe in them, but we can't always say WHY they made them. lots of research there in visualizing networks, analyzing how they assign significance to one variable or another.

Deeper learning is a new level of social consciousness for all interacting and engaging - data surfaces the insight

@AyleeNielsen that's true. there's a lot of fear. it's not about deep learning necessarily, but it's about AI in general. are we going to produce machines that are generally smarter than us. if so, how will we control them?

@AyleeNielsen fear is just negative hype in a way. and most of the hype is the fearful kind.

@AyleeNielsen the fears expressed in those movies are not likely to be realized in the near future, but that doesn't really matter. because we want to know how humanity will deal with it regardless of when it happens.

AI has to go through the same process that other branches of science do. this is not the first time that humans have produced something so powerful that people get scared. nuclear energy, bio-engineering, there are other examples.

we've been through this before. there are risks. but historically, humans as a whole have been sane enough not to destroy themselves. there's a process the scientific community goes through to monitor itself.

@deeplearning4j Do you think if we could always understand the Why behind a AI we could have a better control of AI?

@SystemsandTech you're right. a lot of data has been collected, but we haven't really known how to extract insights from it. now that we have much more powerful tools to extract insight, people need to ask whether they want the data to be collected at all.

@topher920 the more you understand, the more you can control and improve something, so yes. don't get me wrong. we understand the algorithms we've built. but we've taught them to do things automatically, adjusting filters to recognize patterns.

and the way they do that isn't always human readable. it's just a bunch of vectors, long lists of numbers.

@deeplearning4j I think with the volumes of data we don't always know what questions to ask which can make it difficult that is what Watson Analytics is doing helping suggest questions.

chocolate without peanut butter

#BigData without #analytics is like Fred without Barney; Itchy without Scratchy

Fun. 'gasoline' 'match' :-)

"a day without sunshine" as in "dark data that remains unilluminated, hence not producing insights, hence not useful."

map without a compass

@EileenDSmith Making me hungry!

too dangerous for me to answer :-)

sand without a beach (and sunset)

Data without Analytics is like raw spaghetti and rare meatballs without a stove or chef..Would you say data is useless without the analytics? I think the spaghetti and meatballs would be... #deep #ponder #wonderwednesday

Data without Picard! #StarTrek

@theRab i'd go with picard without riker lol or data without geordi #startrek

@rsawyer42 i'd argue that Data learning his humanity from Picard but i can see Geordi too

@rsawyer42 learned i should say

Don't forget the counselor Deana empath ... Best neural net in the quadrant.