Daniela Massiceti is a DPhil student in computer vision in Oxford's Department of Engineering Science.
Areas of interest
My research focuses on using machine learning-based approaches to computer vision and natural language processing in order to build assistive technologies for visually impaired people.
Why is Oxford a good place to work in AI?
The field of AI, at its core, is highly cross-disciplinary, drawing together not just computer science, statistics and mathematics, but law, philosophy, neuroscience, economics and others. Oxford’s collegiate system brings together researchers working across these fields, and it is the resulting ecosystem of engaged experts symbiotically interacting with one another which makes Oxford a prime place to be working in the field. What better opportunity to ‘solve’ the conundrums of AI, moral and otherwise, than by having these experts sit at the same table for dinner every night?
What do you think is the biggest opportunity or challenge around AI?
Currently, our machine learning models heavily depend on the training data we provide them, meaning that if our datasets are biased in some way, or are only labelled for a specific purpose, then our models will not be adaptable in the real world. Think about an algorithm to detect people in images: if the algorithm is never shown images of people with a particular skin colour, then it will fail when presented with this scenario in the real world. Same, too, if we only provide the algorithm with images of people with their faces showing – how could we then expect it to identify a person whose face is not showing? The biggest challenges lie in building models that can work with smaller, unlabelled (or partially labelled) datasets collected on-the-fly in the real world, and models that can transfer what they’ve learned about one task to other related tasks. These challenges touch on unsupervised/low-shot learning paradigms, online learning, and transfer learning – all ‘lingo’ in the field of machine learning.
More about Daniela