I have always been fascinated by technology, probably because my father is an engineer. Eventually, I followed in his footsteps and became a computer engineer. I learned a lot about technology from my father. My favorite pastime as a kid was to play on computers with him on the weekends. I’ll never forget the time—at just six years of age—when I was walking through a shopping mall and my father showed me a toy robot holding a tray and told me that very soon there will be actual robot helpers in our home working for us. In that very moment, I decided I wanted to work with computers. Unfortunately, after over 30 years, I’m still waiting for a robot to move in. The best robot helpers w have are vacuums. Even though we are not quite there yet, I believe that with today’s AI technologies we have a clear path to lead us there.
My fascination with AI led me to computer science. I’ve found AI to be different from some other research fields in that we have a very efficient learning machine available to us—the human brain. We don’t want to replicate the human brain exactly as it is, but it does serve as a huge source of inspiration.
During my Ph.D., I focused on evolutionary algorithms, which are inspired by the evolutionary process to find a solution to problems. More specifically, I developed a new class of genetic algorithms that adaptively changed the search resolution to make it more efficient. The algorithms focused on areas of interest and avoided getting stuck in local best solutions.
When I joined IBM, I was given the opportunity to continue my work on machine learning (ML). In 2012, we saw a boom of deep learning start to take off and excite scientists as well as the public. Since then, deep learning has produced state-of-the-art results in many areas including image processing, speech recognition, natural language processing, and more. However, I strongly believe that deep learning alone is not a solution to reach a more human-like intelligent system. Humans are extremely good at understanding concepts without looking at actual examples of every possible combination. We also have common sense, can understand abstractions, and can reason our way through challenges. For example, a child can determine at first glance that elephants can’t fly. For a ML system, such common sense is not easy to obtain. A pure deep learning system may need to learn through lots of examples where it sees some animals, such as birds flying, while others stay on the ground.
Over the years I have led many projects related to ML in various application areas, including computer vision, robotics, and natural language processing. One area that is still in its infancy is reinforcement learning, which has the potential to learn from weak reward signals. However, deep learning-based reinforcement systems suffer from several problems, including sample inefficiency, lack of explainability, and reward engineering.
I strongly believe the answer to efficient learning, including reinforcement learning, is to combine deep learning with symbolic AI. The field is commonly known as neuro-symbolic AI. There’s a lot of knowledge embedded in the vast amount of literature that humans generate every day. A neuro-symbolic system will enable the algorithms to use that knowledge directly, which can then help AI reason and better understand abstract concepts. For example, it could reason only animals with wings can fly. The neuro-symbolic systems combine the best of both worlds—the stochastic data-driven learning and symbolic AI. I’m excited to work as program director for neuro-symbolic AI, as IBM is at the forefront of neuro-symbolic research. The team is working on the next generation of AI algorithms that can leverage the knowledge to learn explainable models with less data and compute.
AI is already transforming the world around us; however, I still believe this is just the start. I feel lucky to be able to work toward goals I set for myself years ago and can’t wait to see what AI’s future will bring. It’s important to reflect on where you would like to see yourself in the future but also remain flexible throughout the journey. Most importantly, never give up on your goals and dreams.
Sumber: ACM Magazine