Back to all updates

about 3 years ago

Meet the Judges

Unlike previous hackathons, challenge participants have benefited from weeks of preparation and will showcase their demo applications to a live audience and judging panel. Get to know the judges that will be selecting the winners on-site!

We asked each judge a few questions and will be sharing their answers in the days leading up the challenge. Look out for our “Meet the Judges” posts here and on social media, and be sure to cast your vote during our public voting, which beings November 9th.

 

Meet Jeff Hawkins! Jeff is Co-founder of Numenta. See what Jeff shared:

1) How did you get involved in the Machine Intelligence field?

I got interested in how the brain works because understanding how we think and understand is arguably the most important intellectual endeavor of all time. Building intelligent machines is a natural outcome of understanding how brains work and will be the most important technology over the next 50 to 100 years.

2) What do you think is the most potentially useful or interesting application for Machine Intelligence?

In the short term: making use of the incredible amount of data we are collecting. In the long term: building super smart, never-tiring, machines to help us understand and explore the universe.

3) What excites you most about HTM Technology?

HTM technology is exciting because it is an entirely new way of understanding data streams. It is also exciting because, more than any other machine learning technology, it is an accurate model of how real neurons and neural tissue are working in the brain.

4) Where do you see the future of Machine Intelligence in the next 10 years?

In ten years machine intelligence will  be similar to personal computing was in 1985. The industry will be big, important, but with decades of growth ahead..

 

 -----------------------------------------------------------------------------------------------------------------------   

Meet Fergal Byrne! Fergal is an Independent HTM Consultant and a long-time member of the NuPIC Commmunity. Here are Fergal's thoughts: 

1) How did you get involved in the Machine Intelligence field?

I've always been interested in AI, since reading practically all of Asimov as a kid. Apart from dabbling in GOFAI and some NNs in college, I'd waited and watched as AI seemed to go nowhere for 20 years. I saw one of Jeff's talks in about 2010 and bought On Intelligence, but only really got involved again in 2013 when NuPIC was opened up. Since then I've been working full-time on HTM theory and applications, and helping to grow the community. 

2) What do you think is the most potentially useful or interesting application for Machine Intelligence?

In our end of the field, the biggest thing is understanding how the brain works. This will transform the future of humanity, even if we fail to build intelligent machines.  Apart from that, we'll be able to build really useful technologies based on what we learn about how natural intelligence works. Eventually, I'm sure, we'll figure out enough to build machines to help us finish the job. A future of collaborating humans and intelligent machines (envisioned in Iain M Banks' Culture novels) seems ideal as one to work towards.

3) What excites you most about HTM Technology?

It's the correct level to reason about cortical function, and it's the best basis for building real theories of cognition.   

4) Where do you see the future of Machine Intelligence in the next 10 years?

That depends on a) what we do next, and b) how people in currently disparate fields combine to do the work. In HTM, we have a huge amount of theory still to do, which I'm part of, and there's also a lot of work on improving and extending the technology. We can build links with others in both Neuroscience and Machine Learning, improving collaboration and learning across several fields.

The Deep Learning people have built a great ecosystem of academic and industrial teams, and they have excellent PR machinery. Every development builds on the last, and their progress is accelerating. HTM is not a competing field (as both sides seem to suggest), it's just a more brain-centric and structured version of something similar. The DL guys are not shy about recruiting previously competing ideas if it means succeeding, so we now have ConvNets, RNNs, autoencoders, LSTMs, and the fusion of these in things like DeepMind's Q-Networks. 

When the need arises (and I'm sure it will), HTM will join this family. Then we'll have many more people working on HTM, and we'll get to where we're all headed.

 

 -----------------------------------------------------------------------------------------------------------------------  

Meet Marion LeBorgne! Marion is a Senior Software Engineer at Numenta. Check out what Marion had to say: 

1) How did you get involved in the Machine Intelligence field?

I took machine learning classes in college and was instantly drawn to it.  Since then, I've often had side projects with machine learning models trained on publicly available data.  For example, I trained a model on Craiglist data to detect unusually priced cars.  That's how I bought my first car: it got detected as one of the anomalies that had an unusually low mileage and price compared to the main cluster.

After graduating, I worked for about a year in Venture Capital - where there was also lots of fun data - and then joined CloudWeaver (acquired by F5 Networks) to help build its analytics platform.  This is when I heard about Numenta.  The HTM was a revelation and I remember thinking "Finally, a machine learning algorithm that takes time into account".  Eager to learn more, I went to a Numenta hackathon where I experienced a combination of a very friendly team, a strong open-source stance, and cutting edge technology.  Fast forward; now I work for Numenta :-)

2) What do you think is the most potentially useful or interesting application for Machine Intelligence?

I think the internet of things and wearable computing are an area where Machine Intelligence will be extremely useful.  What's the point in collecting all this data if we are unable to make sense of it?  Anomaly detection, prediction, and classification of temporal data streams are going to be instrumental to extract meaningful patterns from all these sensors so that humans can interpret them. 

3) What three words would you use to describe Numenta?

  • Hierarchical 
  • Temporal 
  • Memory ;-)

4) What excites you most about HTM Technology?

The temporal and online nature of the HTM. As computing technologies move forward, the temporal data collected is closer to real-time.  The HTM is extremely well suited for these types of problems.

I'm also excited about the fact that you don't have to specifically engineer features for HTM models; the HTM is capable of learning the relevant spatial and temporal features.

5) Where do you see the future of Machine Intelligence in the next 10 years?

Sensor data is everywhere and is going to keep exploding. HTM models can help us make sense of this data. It's hard to predict the future, but I could definitely see a scenario where lots of little anomaly detection models are plugged into data streams coming from sensors - like wearables or other connected objects. Today, many cloud platforms collect wearables data, like Apple Health, Google Fit, Microsoft Health, just to name a few. In the future, having powerful models like the HTM that can find anomalies or classify this data is going to be key to build quality applications for the end user. 

 

 ----------------------------------------------------------------------------------------------------------------------- 

Meet Subutai Ahmad! Subutai is VP Research at Numenta. Here is what Subutai had to say: 

1) How did you get involved in the Machine Intelligence field?

I've been interested in it for a long time. As an undergraduate I studied Computer Science, and a little bit of Psychology and Linguistics. It was as a graduate student, when I studied Neural Networks and Neuroscience, that I really got interested in Machine Intelligence.  I then worked in startups for many years and learned about the challenges of creating real products using machine learning. But these techniques were pretty far removed from anything the brain does. At Numenta I can finally combine the best of both worlds.

2) What do you think is the most potentially useful or interesting application for Machine Intelligence?

There are many applications. One of my favorites is continuous personal healthcare.  Imagine if you had thousands of sensors integrated into your body. Everything a doctor might possibly measure, except they are measuring everything all the time. An intelligent system would be able to continuously monitor you, detect problems, and suggest specific actions. It would adjust to your specific body and living style. For difficult situations it might quickly try out a bunch of different remedies that are known.  Essentially it would plan out tiny medical experiments, and rapidly figure out what works for you specifically. Of course, it would also know when to alert specialists or emergency personnel (e.g. if you have a stroke).

3) What three words would you use to describe Numenta?

Hierarchical Temporal Memory

4) What excites you most about HTM Technology?

Anyone working on HTM technology can simultaneously create cutting-edge applications and contribute to one of the greatest scientific challenges of all time.

5) Where do you see the future of Machine Intelligence in the next 10 years?

Most of the fundamentals will be well understood. The main challenges in 10 years will be in building really large-scale practical systems.

 

----------------------------------------------------------------------------------------------------------------------- 

Meet Jeff Fohl! Jeff is a User Experience Designer, a long-standing member of the NuPIC community, and designed the HTM Challenge t-shirts. Here are Jeff’s thoughts:

1) How did you get involved in the Machine Intelligence field?

I read Jeff Hawkins' book, On Intelligence, which was a revelation to me. It profoundly altered the way I viewed the world, and how we perceive our own experiences. This led me to attend the Fall 2013 NuPIC Hackathon. I have been studying, learning, and trying to help in any way that I can ever since.

2) What do you think is the most potentially useful or interesting application for Machine Intelligence?

This is a tough one, since I find it difficult to narrow down the potential uses of HTMs. Whenever I do, the list just seems to get longer! That said, at the moment I am most interested in applications using sensors that are not native to humans (such as vision, hearing, touch, smell, etc). Instead, HTMs in which we can use non-human sensors will enable us to derive valuable intelligence about domains in which we humans do not normally operate. Examples might be in complex machine systems such as computer networks, electrical grid, energy plants, factories, or the inner workings of automobiles. Additionally, scientific research can greatly benefit from HTMs connected to a wide variety of environmental sensors that humans do not possess.

This will give us a greater level of intelligence about domains that are difficult for us to understand natively and intuitively.

3) What three words would you use to describe Numenta?

  • Visionary
  • Open
  • Persistent

4) What excites you most about HTM Technology?

I am most fascinated by how HTM is different from traditional approaches to computing. In fact, one may perhaps take the position that an HTM does not so much compute information, but instead responds, or is changed by information. The implications here are enormous. In traditional computing, someone, somewhere, has to come up with a set of rules in order to deal with any given situation. It tends to be very rigid. Generally, in order to create effective systems, the programmer must have a very firm grasp of the domain in which the computer is working, so that they can write rules that allow the computer to deal with the variety of situations the system is likely to experience. With an HTM, it is constantly changing and adapting to the information it is exposed to. This implies that we can build systems that can become extremely knowledgeable about a domain without needing to know very much about that domain ahead of time.

5) Where do you see the future of Machine Intelligence in the next 10 years?

Technology evolves very quickly, so it is very hard to see accurately into the future, especially when a specific thing will happen, but I will give it a shot. I foresee HTMs implemented in a variety of domains, but most importantly, also interconnected. This interconnectedness will allow the generation of very deep hierarchies, creating a kind of global intelligent resource of unprecedented scope. Imagine a scenario in which HTMs running in traffic monitoring equipment, cars, medical equipment, weather stations, financial institutions, home appliances, satellites, etc, are all able to draw on each other’s collective intelligence to make informed decisions about the challenges that they each face.