ECCV 2020 Daily - Wednesday

believe it, but it was many years ago. Over all those years, what do you think you have taught computers to do that they did not do before? It is not so much that we have taught them, we have just got a lot more data. I wouldliketosaythatwesomehowmoved closer to understanding intelligence, but I just think we got much better at fancy forms of memorization. So, data is the gold of the ‘20s? I think so. Everybody is talking about trying to do more with fewer labels – unsupervised learning, self-supervised learning, semi-supervised – but it is still the case that big data is the oil of modern deep learning, and we are trying to break our dependency on fossil fuels, so to speak. It sounds like the relationship between industry and academia is something that you value very much. What do you think this relationship is missing right now that could make it stronger and more effective? That is a good question. Right now, it is already powerful and effective because we are in an era of big data driven deep learning and the tech companies have so many computational resources and so much money. Really incredible advances are happening. The problem I think is that academics are either being plucked out of academia or partially plucked out of academia, which leaves very few professors in the halls to keep the lights on. We have substantial service requirements and committee work and it is hard to sustain that in the current 3 Serge Belongie 11 DAILY W e d n e s d a y