Neural Networks and the Future of Machine Learning

Print Friendly, PDF & Email

In this special guest feature, Gary Baum, Vice President of Marketing at MyScript, talks about how handwriting recognition is enhancing machine (and human) learning. As an input method, handwriting recognition teaches machines to adapt to the user, adding in another layer to their evolving skill set. Those users can program systems simply by jotting down notes and in turn, build platforms most reflective of the human experience. Gary is a tech industry veteran with more than 20 years of executive marketing and product management experience. At MyScript, he oversees global marketing activities and educational efforts to build MyScript brand awareness and drive technology adoption, expand and commercialize digital ink technology offerings, and nurture strategic collaborations and partnerships within the digital writing ecosystem. He holds a Bachelor of Science in Electrical Engineering (BSEE) and graduate studies in computer architecture and advanced control theory. Gary holds six patents in the field of CPU and system design.

Not long ago, many would scoff at the notion that a machine is “learning,” “doing” or “knowing.” But neural networks and artificial intelligence (AI) technologies are layering those skillsets together to perform increasingly complicated, human-like functions. Google DeepMind, for example, is one of few very advanced neural networks that are driving the future of machine learning.

While machines have previously been able to read and answer our questions about news articles, for example, their knowledge was often limited by the length of a piece or driven to brute force computation. Newly-developed algorithms enable those systems to learn from experience and online data – leading to a more sophisticated understanding of topics and language. Researchers put this theory to the test by inputting hundreds of thousands of Daily Mail and CNN articles into a system with the goal of accurately detecting missing words or predicting a headline. The neural network correctly answered more than half of all queries, struggling only with those that featured more complex grammatical structures.

Countless researchers agree on the benefits of deep neural networks for applications ranging from acoustic modeling to speech recognition, and more recently even facial recognition. Deep neural networks with new training methods are even claimed to outperform the older Gaussian mixture models in audio representations. More recent applications are using the technology to predict human emotions by monitoring input methods.

Neural networks continue to push the boundaries of machine learning. While blockbuster hits such as “I, Robot” have made many consumers weary of artificial intelligence, these technologies help simplify our lives. From suggesting future vacations destinations based on a user’s travel history, to outlining steps a consumer can take to meet their weight loss goals, neural networks enable machines to be more helpful than ever before.

Implementing more natural input methods

When it comes to neural networks, input methods can make all the difference. By providing users with natural input methods that closely reflect their needs and habits, engineers and researchers can reduce the time it takes for machines to adapt to users instead of waiting for users to adapt to the device. Although input methods are becoming increasingly organic, the strength and ultimate success of a device’s user experience is linked to its input method, stressing the need for more natural and intuitive forms, such as handwriting recognition. Handwritten information has been pervasive in society for centuries but only comparatively recently has it been embraced in digital ink form on modern computing devices. The interpretation of that digital ink has remained a significant challenge for the industry, though that hasn’t prevented the advancement of a variety of neural network techniques.

Computer science researchers at the University of Toronto have met increased demands for handwriting recognition technology by teaching a neural network how to write by hand. Similarly, a former Spotify engineer collected 50,000 fonts across the Web to train a neural network to create characters and fonts of its own. At MyScript, neural networks have been at the core of our technology since the very beginning of the company, long before it was a popular point of discussion topic. The team at MyScript Labs has a deep experience developing, customizing and adapting this technology to build out handwriting recognition systems for applications ranging from text and math, to shapes and music. Our knowledge base includes a variety of neural net architectures including perceptron, deep architecture, convolutional networks and recurrent networks among others.

These natural, more intuitive innovations not only enhance the device’s ease of use but also enable deployment in applications, such as automotive, where handwriting was not possible only a few years ago. Automotive is a compelling example since it is far safer than most other input methods in a vehicle, according the AAA in a 2015 study. With handwriting, you can keep your attention on the road for minimum cognitive distraction to the driver and perform data entry while the vehicle is in motion – a key benefit to the driver. As an input method, handwriting recognition enables machines to adapt to the user’s natural written input and adds another layer to their evolving skill set. Wearables are another example where super imposing characters on top of one another allows for more efficient information processing, especially in situations where speaking to your device is not desirable.

Perhaps the most attractive handwriting use case is notetaking. Writing on a digital device with a high quality active stylus is becoming increasingly common. Handwriting for a wide variety of languages that includes 95 percent of the earth’s population can now be accurately interpreted into digital format through the use of neural networks. Even more impressive is recognition of diagrams, shapes, mathematic equations and more to provide for a profound transformation of most written notes into the digital environment. This now enables a wide range of users from students to company executives to easily share their handwritten notes in digital form with associates.

Machines that have learned how to read handwriting have previously been confined to science fiction novels. That, however, has changed due in large part to neural networks and AI technologies. By absorbing information and learning from detailed and intensive training, neural networks continue to add to the list of complex tasks machines can perform. As intuitive input methods such as handwriting recognition become more widely available, the value of neural networks will continue to skyrocket.

 

Sign up for the free insideBIGDATA newsletter.

Speak Your Mind

*