CVPR Daily - Tuesday

I am a senior research scientist at Elemental Cognition, which is this fundamental AI research startup that is focused on pushing the boundaries of AI forward through to AI systems that can explain themselves, collaborate with humans via dialogue, and machine reading to have a better shared understanding of the world. You are here at CVPR to give a talk and just for one day. What is your relationship with computer vision? I would say that I have a complicated relationship with computer vision. We love complicated relationships! [ laughs ] I’m joking. [ laughs ] I’m not! [ both laugh ] I got exposed to working on vision and language when I was at Microsoft Research for about a year. There they had an ongoing research project on a line of research which was mine which was narrative structure understanding and story understanding, which they wanted that to be multi-modal. Hence, through that, I got involved with vision. Before that, I had never done any vision and language work. That was the start. After that, I built up on top of that line of work. I kept working on different vision and language projects, which involved deeper language understanding and common sense reasoning, which is the field of AI that is really close to my heart. I’m just super passionate about doing something on common sense reasoning. All the vision and language work that I’ve done has been trying to push that forward. That’s the complicated relationship. I’m not part of the vision community, I would say. I’m a part of the vision and language community. I was very happy to come to CVPR after all and become a part of the community. So that’s the complicated relationship. [ laughs ] You mentioned in your talk that ten years ago there were concepts that were difficult to explain to the community, and today it’s much easier. Can you tell our readers what that was and why you think that changed? So actually I made a point that the change happened and the kind of revolutionary effect that deep learning had on the community happened in other fields. I would characterize different AI challenges that we have in two categories. One being pattern recognition. The other being perception and reasoning. Deep natural language kind of falls beyond pattern recognition and into a task that requires reasoning and common sense. Anything like that, which deep natural language understanding is one of them, requires Nasrin Mostafazadeh Nasrin Mostafazadeh is a senior AI research scientist at Elemental Cognition where she works on the next generation of AI systems that not only comprehend stories, but also explain themselves. 16 Women in Science Tuesday