Our Global BERT-based Transformer architecture fuses global and local information at every layer, resulting in a reading comprehension model that achieves a deeper understanding for long documents and enables flexibility for downstream tasks.
Visual localization is used by robots and self-driving cars to estimate their position and in AR apps to interact with the physical world both indoors and out in the open. This article gives an overview of current state-of-the-art methods and their advantages and drawbacks.
By dissecting the matching process of the recent ColBERT model, we make a step towards unveiling the ranking properties of BERT-based ranking models and show that ColBERT (implicitly) learns a notion of term importance that correlates with IDF.
Running into the unknown with RunAhead: finding pleasant running tours and intuitively navigating through them. The use of combinatorial optimization and UX design, together with head-tracking technology, provides a non-intrusive way of helping runners to discover new itineraries.
A novel framework which uses individual sample losses as error measures to determine the relative difficulty of samples in a dataset. Can be plugged on top of existing neural network models to implement curriculum learning for any task, even with noisy datasets.
A new approach to image search uses images returned by traditional search methods as nodes in a graph neural network through which similarity signals are propagated, achieving improved ranking in cross-modal retrieval.