27 February 2021
Agents are becoming more artificially intelligent but a number of problems must be solved before we can unleash them into our physical world (i.e. robots) and trust them in our digital one.
18 December 2020
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.
15 December 2020
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.
9 December 2020
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.
8 December 2020
How harder negatives facilitate better and faster contrastive self-supervised learning and ways of synthesizing harder negative features on-the-fly.
7 December 2020
Podcast and transcript on 3rd Robot Learning Workshop at NeurIPS 2020 with Julien Perez, group lead machine learning and optimization at NAVER LABS Europe and co-organiser of the workshop.
24 November 2020
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.
5 November 2020
A new greedy, brute-force solution improves fairness in ranking and encompasses realistic scenarios with multiple, unknown protected groups.
16 October 2020
By incorporating a set of reusable constructs that enable advanced functionality, Flow makes platform-agnostic application modelling, creation, distribution and maintenance easy.
24 August 2020
How mid-size sets of captioned images can rival with large-scale labelled image sets to learn generic representations
21 August 2020
A novel efficient model for whole-body 3D pose estimation (including bodies, hands and faces), that is trained by mimicking the output of hand-, body- and face-pose experts.
4 August 2020
A new method called FORCE achieves extreme sparsity in artificial neural networks by progressively removing up to 99.9% of parameters at initialization, making it a promising candidate for training networks on edge devices (like drones or smartphones).