By Guruduth Banavar
With thousands of scientists, engineers, and business leaders focused on cognitive computing across IBM Research and the IBM Watson Group, IBM is pursuing the most comprehensive effort in the tech industry to advance into the new era of computing. Nobody has more people on it, a broader array of research and development projects nor deeper expertise in so many of the most significant fields of inquiry.
Yet we understand that to accelerate progress in cognitive computing, we can’t do this alone. That’s why IBM has been pursuing a strategy of forming deep collaborative partnerships with academic scientists who are among the leaders in their fields as well as opening Watson as a technology platform for others to build on.
Our newest research collaboration is with Yoshua Bengio and the Montreal Institute for Learning Algorithms (MILA), with its 60 researchers — five professors and a group of elite graduate students. Yoshua, a professor at the University of Montreal and the director of MILA, is recognized globally as one of the leading thinkers in the field of deep learning. Scientists and engineers from both IBM Research and Watson Group will work with him and his colleagues on a number of deep learning projects focusing on language, speech and images.
Deep learning is a branch of A.I. where, increasingly, machines learn from experience rather than requiring extensive manual training by humans. Deep learning is critical to fulfilling the promise of cognitive computing because as such systems acquire knowledge in this way, they’re able to learn more, faster, and with less expense. Those capabilities ultimately help people make better decisions that in turn help businesses, government and society work better.
At IBM, we already exploit deep learning techniques in our language understanding, speech, vision, language translation, dialog and other technologies, which are being put to use in a wide range of industries, from healthcare to retailing. We offer several cognitive computing services to entrepreneurs and developers on our Bluemix cloud development platform. And we also have teams working to scale machine learning applications on server clusters.
One of our research projects shows how deep learning can help transform healthcare. Scientists from IBM Research and Memorial Sloan Kettering Cancer Center have developed a system for detecting skin cancer. Rather than requiring scientists to write detailed instructions, we tell the computer what a cancerous lesion looks like and feed it thousands of images to continuously improve its knowledge. Using this technology, physicians or even individuals will be able to photograph spots on skin with smartphones and text them to a cloud-based diagnostics service where computers possessing deep learning capabilities will accurately flag the images that should be examined by a cancer specialist. This is a big deal because the survival rate for melanoma is 98% with early detection.
We already do a lot with deep learning innovation at IBM, but we want to do even more. We believe that the new collaboration with Yoshua Bengio and his group will accelerate progress in this critical area of cognitive computing.
Yoshua and his team have produced advances in the field in the past two years that have expanded the capabilities of deep learning far beyond what scientists would have predicted as recently as five years ago. We will now collaborate with him and his colleagues push this even further. In speech, for instance, we would like to extend the capabilities of neural networking architectures to model more aspects of speech recognition and generation processes. In the computer vision realm, we’re interested in building systems that better understand what’s happening in videos.
In addition to Yoshua’s expertise and the skills and talents of his colleagues, one of the things that attracted us to them is their commitment to the practice of “open science.” His team contributes all of their learning inventions to the Theano project and its derivatives on GitHub. At IBM, we have a proven track record of driving open innovation to accelerate progress, and we’ll continue to leverage these open innovations to design cognitive computing products and services capable of transforming industries and professions. We’ll also add the new capabilities to our IBM Watson open development platform, where students, startups, IBM business partners and clients can build their own cloud-based cognitive computing applications.
I’m convinced that this deep learning collaboration will help us achieve our ultimate goal: combining the strengths of humans and machines to achieve things that neither could do as well on their own.
To learn more about the new era, read Smart Machines: IBM’s Watson and the Era of Cognitive Computing.