Interview with Tsvi Achler, CEO, Optimizing Mind - Speaker at Global Data Science Conf - March 2017 Posted on : Feb 20 - 2017

We feature speakers at Global Data Science Conference March 27 - 29, 2017 to catch up and find out what he or she is working on now and what's coming next. This week we're talking to Tsvi Achler, CEO, Optimizing Mind.

Interview with Tsvi Achler

1. Tell us about yourself and your background.
I have a unique background focusing on the neural mechanisms of recognition from a multidisciplinary perspective. I wanted to understand brain recognition from multiple perspectives so I have done extensive work in theory and simulations, human cognitive experiments, animal neurophysiology experiments, and clinical training. In the process I obtained an Electrical Engineering, Computer Science degree from UC Berkeley and advanced degrees in Neuroscience (PhD) and Medicine (MD) from University of Illinois at Urbana-Champaign.  I've also worked at Los Alamos National Labs, and IBM Research.

2.  What have you been working on recently?
I head Optimizing Mind whose goal is to provide the next generation of brain motivated machine learning algorithms.  We focus on solving the "black-box" problem of neural networks.  Feedforward neural networks have been the basis of Machine Learning methods such as Deep, Convolution, and Recurrent Networks.  However the internal decision processes of feedforward networks are difficult to explain: they are known to be a "black-box". We have developed a new type of neural network motivated by neuroscience.  This allows the network to be more updatable, and the internal decision process to be easier to understand. 

3. Tell me about the right tool you used recently to solve customer problem?
We are able to take any network such as: Convolutional networks, Recurrent Networks, or Deep Networks, and explain what each node is looking for.  Moreover given any test pattern, we can explain which inputs were most important in the network's decision process.  We were able to show to our customers, what their network is considering most important, and given a test pattern, which factors contributed network-wide to a confident decision and which did not. Afterward, the customer had less anxiety and could trust and verify their networks.

4. Where are we now today in terms of the state of Data Science, and where do you think we’ll go over the next five years?
The most successful data science applications have been those which do not have terrible consequences if the answer is wrong.  For example if  SIRI mistakes your speech you simply change it. As data science is being adopted more and more it is being applied to fields where mistakes have huge consequences, such as self-driving cars or medicine.  This means that the networks will have to be better understood and regulated, in other words they will need to be explained.

The other area of advancement is personalization of networks.  For example the network that is best suited to respond to me may not be the same as the one for you.  The networks need to be customize-able on the small devices they run on. Our technology enables active machine learning on small devices, personalized to each device. This has been an impossible goal for traditional methods.

We offer the most robust service to i) explain networks for customers and regulators and ii) to make modifications on the fly, without re-training!

6. What are some of the best takeaways that the attendees can have from your  "Explainable AI using brain-motivated neural networks" talk?
That although networks may have great performance after being trained and tested, oddities from the data set may creep in and the network may not be recognizing things as expected!  You must be able to explain and validate the network to make sure this does not happen.

7. What are the top 5 Data Science Use cases in enterprises?
Our technology can be applied to any machine learning application. The top five we are working on now include aspects of computer security, medical diagnostics, advanced fintech and self driving cars

8. Which company do you think is winning the global Data science race?
Companies with many resources continue to have a huge advantage. However those that haven't modernized must do so in a smart way. The largest companies have shown some promising results, but in reality they've hit a wall, the wall of the black-box.  With our breakthrough technology soon to be unleashed, even smaller companies can catch up to the biggest ones.

9. Any closing remarks
Ultimately we need to do less data science, and more automated systems that can do self introspection. This will bring us several steps closer to thinking machines and the ability to reach an artificial intelligence level to help us with the biggest problems humanity is facing. The conference audience will gain a fresh perspective on the explainability and editability of the blackbox nature of neural networks.