Share this post:
Yasunori Yamada, a computational neuroscientist at IBM Research – Tokyo, has long been fascinated by human intelligence, consciousness and robots. “We humans can learn autonomously, behave flexibly, and exploit and adapt to an environment. But understanding the mechanisms of intelligence is no easy task,” Yasunori said.
Yasunori Yamada, IBM Research-Tokyo
When Yasunori was in high school, he discovered Franz Kafka’s Die Verwandlung (The Metamorphosis). As he read through the novella, he imagined himself having multiple legs and eyes, and started wondering if he could adapt to such a new and different body.
“Then, I thought about how intelligence relates to our body,” Yasunori said.
After earning a PhD in Information Science & Technology from The University of Tokyo in 2015, where he published such papers as An Embodied Brain Model of the Human Foetus, he joined IBM to design intelligent machines to help us better understand ourselves. In our latest Q&A with an IBM scientist, we ask Yasunori about building intelligent machines, and how they can help improve our own intelligence.
Why did you decide to join IBM Research?
Yasunori Yamada: I joined IBM because I want to advance the research into cognitive computing. I found projects such as the Blue Brain project, the IBM Watson Jeopardy! Challenge, and the Synapse project fascinating. These and other projects stimulated my interest in being a scientist at IBM Research.
What I see in IBM Research is that IBM is not merely pursuing cognitive computing as an extension of conventional machine learning and artificial intelligence. Instead, like what the word “cognitive” represents, IBM is advancing this new research field in a multi-disciplinary way to better support us humans.
Cognitive computing has opened up a new academic field to understand and support human endeavors. And IBM Research pursues cognitive computing on a global scale across science, engineering, robotics, neuroscience – any and all disciplines and industries. I really wanted to be part of the exploration.
The paper An Embodied Brain Model of the Human Foetus that you and other colleagues from The University of Tokyo, Kurume University and Kyoto University wrote during your PhD program was recently published in Scientific Reports, an online, open access journal from the publishers of Nature. What is the paper about? What do you hope to accomplish based on the findings?
YY: I was fortunate to collaborate with six fantastic scientists with different expertise, including developmental science, medical science, robotics and neuroscience, to develop an embodied brain model. We used the model to simulate cortical learning through interactions between brain, body, and environment related to spontaneous bodily movements. It helped us explore the causal link between sensorimotor experiences and cortical learning – such as how a fetus explores moving one’s own body while still in the womb – and how sensorimotor experiences guide normal and abnormal brain development, as well as gave us insights into neuro-developmental disorders and potential treatment.
What have you been working on since you joined IBM Research a little more than a year ago?
YY: Day-to-day, I appreciate the flexible workstyle options at IBM Research. It lets me wisely choose how to use our limited time of 24 hours a day!
Right now, I am collaborating with the colleagues who continue to advance accessibility research on several different cognitive computing projects. I recently presented some of this work – cognitive fatigue detection, deep neural networks and cortical models – at three major workshop and conferences: Workshops at the Thirtieth AAAI Conference on Artificial Intelligence (AAAI-16), the 25th International Joint Conference on Artificial Intelligence (IJCAI-16), and The Organization for Computational Neurosciences (CNS 2016).
My presentation, “Toward assessment of mental fatigue in daily life from eye movements,”* at the AAAI-16 workshop was about a novel system we developed to detect mental fatigue in natural viewing conditions, for example, while watching TV. It could be used to help manage mental fatigue and improve cognitive and behavioral performance.
At the IJCAI-16, I presented “Weight features for predicting future model performance of deep neural networks.” Deep neural networks (DNNs) frequently require the careful tuning of model hyperparameters, such as the number of learning layers and the learning rate. In this study, to speed up the tuning, we proposed novel features extracted from network weights at an early stage of the learning process as explanation variables for predicting the eventual model performance of DNNs. Our proposed features can help construct prediction models with a smaller number of training samples, and terminate underperforming runs at an earlier learning stage than the conventional use of the learning curve.
And at CNS 2016, I presented “Large-scale cortical models towards understanding relationship between brain structure abnormalities and cognitive deficits.”*
Tell us more about the research on cortical models that you presented at CNS 2016.
YY: Brain imaging studies have revealed multiple kinds of structural abnormalities associated with brain diseases, such as Alzheimer’s disease. However, how each structural abnormality affects cognitive functions involved in brain diseases is still poorly understood. To deepen better understanding of this causal link, I took a computational approach using large-scale brain models. The computational approach allows us to systematically manipulate each factor and conduct detailed analysis – which is difficult to do in human studies – and help disentangle complex relationship between brain structure and cognitive functions.
In the study, I focused on APOE-4, one of the major genetic risk factor in developing Alzheimer’s, and built two cortical models with brain imaging data of the aging APOE-4 carriers and non-carriers. Through the computer experiments, I show the possibility that the structural connectivity abnormality observed in APOE-4 carriers might reduce cortical information propagation.
What’s next for your research?
YY: I would like to understand the underlying mechanisms of human intelligence enough to reconstruct it. At the same time, I would like to apply that knowledge to helping cognitive computing better support us humans.
Studying neuroscience has been known as, and positioned as, a fundamental research with few exceptions. Finding connections between neuroscience and industry to help solve business problems is still not quite reality. I know it won’t be easy, but I would like to take a big step forward in blending neuroscience and business to solve a societal issue.
Where do your ideas for these projects come from?
YY: When I read books or go for a jog. When I walk or jog, that seems to stimulate my brain to do a lot of thinking, as well as digest and sort out my thoughts. It is refreshing to me.
What is a word or phrase that you live by? Your motto?
YY: Seize the day. Even if I have a difficult day, I still try to enjoy every moment.
* This study was partially funded by the Japan Science and Technology Agency, under the Strategic Promotion of Innovative Research and Development Program.