LessaWorld - Thoughts on AI, ML, Data Science, and Innovation
64 views | +2 today
Follow
Your new post is loading...
Your new post is loading...
Scooped by Andre Lessa
Scoop.it!

How Neural Networks Are Turning Human Brains Into AI

How Neural Networks Are Turning Human Brains Into AI | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
Neural networks can’t learn until you throw throw enough data at them. They need large quantities of information to consider, pass through their layers, and attempt to classify. Then, they can compare their classifications to the real answers, and either pat themselves on the back or try a little harder.
Andre Lessa's insight:

hmm... wondering how long we are from a self-taught Neural Network... 

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

The Father of the Emoticon Chases His Great White Whale

The Father of the Emoticon Chases His Great White Whale | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
What’s passing for human intelligence in computers most of the time is advanced computation, manipulation, and matching of symbols with no understanding of them.
Andre Lessa's insight:

Scott does have a lot of interesting thoughts on AI on his blog at http://scone1.scone.cs.cmu.edu/nuggets/

 

Wasn't aware of his research on AI until reading this CMU article.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

Deep Learning And The Future Of Search Engine Optimization

Deep Learning And The Future Of Search Engine Optimization | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
Previous to the last decade, the business industry had all but turned its back on the field of deep learning. The limitations of chip-processing abilities and the data sets used in the artificial neural networks that ran them made Hinton’s and his colleagues’ theories impractical and ahead of their time.
more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

Robot Masters New Skills Through Trial and Error

Robot Masters New Skills Through Trial and Error | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
In the world of artificial intelligence, deep learning programs create "neural nets" in which layers of artificial neurons process overlapping raw sensory data, whether it be sound waves or image pixels. This helps the robot recognize patterns and categories among the data it is receiving. People who use Siri on their iPhones, Google's speech-to-text program or Google Street View might already have benefited from the significant advances deep learning has provided in speech and vision recognition.

Applying deep reinforcement learning to motor tasks has been far more challenging, however, since the task goes beyond the passive recognition of images and sounds.
Andre Lessa's insight:

Unlike modeling sound waves and image pixels, modeling the robot's movements require environment feedback. To accomplish that, the model also uses a numeric score that points out how well each end-goal task is accomplished. The algorithm is called Reinforcement Learning.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

Planarian regeneration model discovered by artificial intelligence

Planarian regeneration model discovered by artificial intelligence | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
"While the artificial intelligence in this project did have to do a whole lot of computations, the outcome is a theory of what the worm is doing, and coming up with theories of what's going on in nature is pretty much the most creative, intuitive aspect of the scientist's job," Levin said. "One of the most remarkable aspects of the project was that the model it found was not a hopelessly-tangled network that no human could actually understand, but a reasonably simple model that people can readily comprehend. All this suggests to me that artificial intelligence can help with every aspect of science, not only data mining but also inference of meaning of the data."
more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

As We May Think

As We May Think | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
Consider a future device for individual use, which is a sort of mechanized private file and library. It needs a name, and, to coin one at random, "memex" will do. A memex is a device in which an individual stores all his books, records, and communications, and which is mechanized so that it may be consulted with exceeding speed and flexibility. It is an enlarged intimate supplement to his memory.
Andre Lessa's insight:

This is an amazing article. Hard to believe that it was written in July of 1945.  It's essentially the words of a visionary describing the technology of that era and his vision for the future.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

Shazam It! Music Processing, Fingerprinting, and Recognition

Shazam It! Music Processing, Fingerprinting, and Recognition | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
To make for easy lookup, this signature becomes the key in a hash table.
Andre Lessa's insight:

Really interesting and simple explanation of how the Shazam music recognition works. I didn't know it was based on hash-lookups ... I was surprised by that .... am wondering how successful people have been trying to re-think it using traditional Machine Learning classifiers.  Looked around and didn't really see many recent references showing progress (or interest) in that area.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

The Technology that Unmasks Your Hidden Emotions

The Technology that Unmasks Your Hidden Emotions | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
The 80-year-old psychologist pioneered the study of facial expressions in the 1970s, creating a catalog of more than 5,000 muscle movements to show how the subtlest wrinkling of the nose or lift of an eyebrow reveal hidden emotions.
Andre Lessa's insight:

Reading this article I learned about Mr. Paul Ekman's research on facial expression analysis, and a quick Google search took me to his website, and it looks like his findings are a great resource for those trying to apply AI/ML techniques to figure out people's emotions in videos and pictures.  I just wonder how incredible it would be to apply similar research to the animal realm, and include other non-facial features into the mix.  Take dogs for instance, a single picture can't tell whether they're happy, but a video showing their tail moving could certainly point in that direction. The point there is that temporal features would be important in that case so it might be a good research topic too... I haven't read the "Facial Action Coding System" doc, but I wonder if it addresses temporal cases like that. Either way, it would be a good project to unmask animal emotions.  Maybe there's a business case there. :-)

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

The Incredible AI That Can Watch Videos and Tell You What It's Seeing | WIRED

The Incredible AI That Can Watch Videos and Tell You What It's Seeing | WIRED | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
Using a database of 10,000 visual categories Clarifai has built over the past six months, the company’s software tracks the images that appear in the video, automatically describing it with words like “dog,” “female,” “eyes,” and even “cute.”
Andre Lessa's insight:

Writing a visual recog system ... yeah, many companies are working on that ... you train classifiers on thousands  of categories, tag images with the result categories, and hope that your end-user's goal is to find one of those 100K recorded concepts. What if it isn't? For AI startups, that's risk #1. What if your system is not good enough? As I commented in another article, you either get more data, which can become very costly, or you narrow your search domain.  For instance, what if instead of classifying thousands of objects for a general search solution you were to focus on creating a system where you know what exactly you're looking for.  For example, Ad companies could create a simple classifier capable of identifying general types of people  (man, woman, child, height, weight, approximate age, etc...) and create intelligent (classifier-backed) billboards capable of seeing its surroundings and display ads according to the type of people walking by the display. Replace people with cars, and additional opportunities emerge.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

AI won’t exterminate us. It will empower us. — Backchannel — Medium

AI won't exterminate us. It will empower us. - Backchannel - Medium
We need intelligent software that can answer questions such as, “what are the side effects of X drug in middle-aged women?” or at least identify a small number of relevant papers in response. We need software that can track new scientific publications and flag important ones, not based on keywords, but based on some level of understanding of the key information in the papers. That’s augmented expertise, and it’s a positive goal that I and other AI researchers are aiming at.
Andre Lessa's insight:

Many people are indeed reacting to AI advances just like others did during the Industrial Revolution. The fear of the unknown unknowns that might affect one's way of life will always be around whenever new technologies find their way to master tasks previously owned solely by humans.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

The idea maze for AI startups | cdixon blog

The idea maze for AI startups | cdixon blog | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
As a rule of thumb, you’ll spend a few months getting to 80% and something between a few years and eternity getting the last 20%.
Andre Lessa's insight:

I agree that getting those remaining 20% of edge cases is really hard, and many times it's more practical to narrow the domain and the scope of what you're trying to solve.  Building really comprehensive models require a ton of data, and since crawling, indexing, and modeling a gazillionbytes of data is insanely expensive, AI startups are definitely better off staying focused on narrowed domain missions.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

Bill Gates on dangers of artificial intelligence: ‘I don’t understand why some people are not concerned’

Bill Gates on dangers of artificial intelligence: ‘I don’t understand why some people are not concerned’ | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
Joining the likes of Stephen Hawking and Elon Musk, the Microsoft co-founder has sounded the alarm about AI.
Andre Lessa's insight:

Given Gates concern regard having too much AI, and the article statement claiming that "over a quarter of all attention and resources" at Microsoft Research are focused on artificial intelligence, my best guess is that Microsoft's priorities would likely be *very* different if Bill Gates were still part of its leadership.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

IBM supercomputer will now provide legal advice

IBM supercomputer will now provide legal advice | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
Thanks to the work of students at the University of Toronto, Watson will now aid legal cases, supporting a new law research service named Ross.
Andre Lessa's insight:

Interesting concept. In the demo, I see a Natural Language parser for the user's query, a search engine capable of indexing a ton of documents, and a recommendation engine that updates document scores based on the user's pos/neg feedback... should be an interesting project to re-create a similar demo without Watson.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

Memristors Mimic Brains For Massive Machine Learning

Memristors Mimic Brains For Massive Machine Learning | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
startup dedicated to using memristors at the heart of a new kind of analog computer
more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

How the Brain Forms New Memories

How the Brain Forms New Memories | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
The researchers showed each patient an assortment of 100 pictures of celebrities, animals and well known landmarks such as the Eiffel Tower and Leaning Tower of Pisa. In their first analysis, the team identified neurons that responded to one or more pictures. The team then photoshopped composite images, such as Clint Eastwood standing in front of the Tower of Pisa.
Andre Lessa's insight:

I also find it really interesting how our brain tries to predict the spatial area and movement of elements well-defined in this particular picture. Our previous knowledge of the Leaning Tower of Pisa tells us that it's not actually moving during the picture, but our brain is predicting that the person in the park is walking towards the right side and will continue moving its legs until it walks out of the picture. As for the spatial area of the Tower itself, we can't see it all, but without us even noticing, our Brain is actually predicting that the left-bottom corner is somewhere behind Clint's neck.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

AI. The new VC goldrush.

AI. The new VC goldrush. | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
Perhaps one of the most unique ingredients that continues to push the AI industry, is the emergence of pervasive computing principles. With the explosion of mobile and the Internet of Things (or “IOT”), the computing industry shifted from entering data into computers, to wearable devices that continuously observe us and learn.
This shift requires new techniques that are able to interpret this continuous flow of observational data and turn this into ambient intelligence. Achieving this next level of automation will provide companies a better understanding of the likings and preferences of users.
Andre Lessa's insight:

"Through all the hype, founders and investors seemingly forget that the true benchmark of success is not to build an algorithm that is capable of learning how to play Space Invaders, but how to turn AI into real-world solutions."


I sort of agree with the author in that the true benchmark of success is not building AI algorithms that can do random cool things but building AI algorithms that can deliver real-world solutions.  But to the particular example that was pointed out, I do see the learning of games as a step into learning how to interact with the real-world. Maybe it's not a solution for a problem, but an evolutionary step along the way.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

Ray Kurzweil: Humans will be hybrids by 2030

Ray Kurzweil: Humans will be hybrids by 2030 | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
"As I wrote starting 20 years ago, technology is a double-edged sword," he said. "Fire kept us warm and cooked our food but also burnt down our houses. Every technology has had its promise and peril."
Andre Lessa's insight:

2030 is a really optimistic view ... Not so sure the technology to support that is just 15 years away unless there's a bunch of scientists hidden in a vault somewhere working on it right now ... Either way, be it 2030 or 2050, the risk of something backfiring is extremely high. So we can just hope that by then, we'll have some tangible laws governing any abuses in the Artificial Intelligence area I don't believe Isaac Asimov's "Three Laws of Robotics" will be enough at that point.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

Rise of the machines

Rise of the machines | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
And deep learning is not restricted to images. It is a general-purpose pattern-recognition technique, which means, in principle, that any activity which has access to large amounts of data—from running an insurance business to research into genetics—might find it useful.
more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

SAPVoice: When Machines Replace Middle Management

SAPVoice: When Machines Replace Middle Management | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
I recently came across a provocative post by Peter Reinhardt where he illustrates numerous examples of companies that are creating software layers to manage people in industries that were traditionally purely based on human services. Reinhardt lists popular examples such as Uber and Lyft who created software layers in the taxi industry, 99designs in the visual design industry and Homejoy in the cleaning industry. These companies are now employing armies of human workers, optimizing their output, productivity and quality while driving prices down – all without managers to direct those who are performing the jobs. “Management” is almost completely done by machines and algorithms.
Andre Lessa's insight:

Interesting point. For a while now, I've thought of companies like Uber and Lyft as companies that had identified a niche marketplace to automate and be successful. But the perspective of looking at them as companies that are really automating a management layer does provide an interesting take on job automation.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

Google Brain’s Co-Inventor Tells Why He’s Building Chinese Neural Networks — Backchannel — Medium

Google Brain's Co-Inventor Tells Why He's Building Chinese Neural Networks - Backchannel - Medium
phonemes are a human construct — they’re not a fundamental fact of language. They are a description of language invented by humans. Many linguists vehemently disagreed with me, sometimes in public.

One of the things we did with the Baidu speech system was not use the concept of phonemes. It’s the same as the way a baby would learn: we show [the computer] audio, we show it text, and we let it figure out its own mapping, without this artificial construct called a phoneme.
Andre Lessa's insight:

Lots of interesting perspectives in this article about the current state of #AI.  

 

The phonemes argument reminds me of that famous text with a lot of letters missing or misplaced and, STILL, people can understand it and many times, not even realizing the words make no sense at all.

 

He also said "I think time is super important but none of us have figured out the right algorithms for exploring it." ... I was just talking about the possibility of analyzing emotions in animals ... unlike humans where a photo still can demonstrate some level of emotion, you can't easily look at a dog and figure out she is happy.  You need other features beyond facial features.  And even if you include the entire body of the dog in the classification model, you still need time-based features to be able to point out that the dog's tail is swinging from one side to another.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

Online Learning Perceptron | MLWave

Online Learning Perceptron | MLWave | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
The birth of artificial neural nets started with the 1943 paper “a
Logical Calculus of the Ideas Immanent in Nervous Activity”. Two researchers, McCulloch a neurologist, Pitts a logician, joined forces to sketch out the first artificial neurons.


McCulloch reasoned that nervous activity had an ‘all-or-nothing’ activity: A neuron would fire (output “1″) once it’s activation threshold was reached, or it wouldn’t fire (output “0″).
Andre Lessa's insight:

hmm...interesting results for simple linear problems, and it would be even better without the constraint of not being capable of outputting a probability.  It would be interesting to see an implementation where this single-node single-layer Perceptron could output a probability, without a need to move on to more complex ANN algorithms.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

Researchers use games to evolve AI brains | Games | Geek.com

Researchers use games to evolve AI brains | Games | Geek.com | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
that’s not far off from how our brains got to where they are today. They’re not getting bigger, but they’re evolving more integration to make them more functional and efficient.
Andre Lessa's insight:

According to the research, in the experiment, genetic algorithms were used on top of a simple neural network setup and after over 60,000 generations, it was clear that goals were met and very complex neural pathways were set in motion.  Imagine what will be possible the moment we can have near-infinite computing parallelism in the distributed computing world.  In my opinion, the next big computing leap won't be defined on a vertical scale, in terms of CPU speeds north of 10Ghz or Exabyte storage devices,  but it will be defined on a horizontal plain, as the number of parallel I/O connections that can be handled by a single CPU increases, making it easier for individual CPU cores to act as the elements of Artificial Neural Networks.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

The AI Revolution: Road to Superintelligence - Wait But Why

The AI Revolution: Road to Superintelligence - Wait But Why | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it

today’s Marty McFly, a teenager born in the late 90s, would be much more out of place in 1985 than the movie’s Marty McFly was in 1955.

Andre Lessa's insight:

Found it interesting the author's point that when it comes to history, most people think in straight lines ... when in reality we should be thinking exponentially.

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

How a supercomputer sees the State of the Union

How a supercomputer sees the State of the Union | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
See how Obama stacks up when you plug every State of the Union since 2001 into one of the world's smartest supercomputers.
Andre Lessa's insight:

IBM Watson was used to put together a very comprehensive sentiment analysis of all State of the Union addresses over the past 14 years.  Besides this analysis, there's a step-by-step on the IBM web site explaining how to create a similar analysis - https://developer.ibm.com/bluemix/2015/01/21/watson-sees-state-union/

more...
No comment yet.
Scooped by Andre Lessa
Scoop.it!

The Three Breakthroughs That Have Finally Unleashed AI on the World | WIRED

The Three Breakthroughs That Have Finally Unleashed AI on the World | WIRED | LessaWorld - Thoughts on AI, ML, Data Science, and Innovation | Scoop.it
Three recent breakthroughs have unleashed the long-awaited arrival of artificial intelligence
Andre Lessa's insight:

The author claims that 3 recent breakthroughs have unleashed the long-awaited arrival of artificial intelligence: Cheap parallel computation, Big Data, Better algorithms.  I would actually slightly change #1.  It should actually be "Cheap computation".  Period.  It's not just that cheap GPUs offer great parallel performance.  Cheap Storage/Memory/etc. enable a new class of algorithms to be written.  Algorithms that don't care about optimizing datasets to get rid of data redundancy.  Once there was a time where normalizing datasets played a role in keep data small in size but now, who cares? Just save everything, multiple times, and make it easy for the algorithms to do their jobs.

more...
No comment yet.