Recent & Upcoming Talks

Aggregated data is commonplace in areas such as epidemiology and demography. For example, census data for a population is usually given as averages defined over time periods or spatial resolutions (cities, regions or countries). In this paper, we present a novel multi-task learning model based on Gaussian processes for joint learning of variables that have been aggregated at different input scales. Our model represents each task as the linear combination of the realizations of latent processes that are integrated at a different scale per task. We are then able to compute the cross-covariance between the different tasks either analytically or numerically. We also allow each task to have a potentially different likelihood model and provide a variational lower bound that can be optimised in a stochastic fashion making our model suitable for larger datasets. We show examples of the model in a synthetic example, a fertility dataset, and an air pollution prediction application.

From everyday apps to complex algorithms, technology has the potential to hide, speed, and deepen discrimination, while appearing neutral and even benevolent when compared to racist practices of a previous era. In this talk, Ruha Benjamin explores a range of discriminatory designs that encode inequity – what she terms the “New Jim Code.” This presentation takes us into the world of biased bots, altruistic algorithms, and their many entanglements, and provides conceptual tools to decode tech promises with historical and sociological insight. She will also consider how race itself is a kind of tool designed to stratify and sanctify social injustice and discuss how technology is and can be used toward liberatory ends. In doing so, Ruha challenges us to question not only the technologies we are sold, but also the ones we manufacture ourselves.

Probabilistic modeling is a core component of modern machine learning. Nevertheless, at least from our local perspective, educational barriers and practical implementation issues have undermined its applicability in both academy and industry. In practice, this vacuum is often filled by black-box approaches such as traditional neural networks. In such a scenario, our group (LOGIA) aims to provide some basis to enhance teaching, research and applications of probabilistic machine learning methods in our local community. In this talk we present some of our current work on the area, ranging from the proposition of new models to extensions and applications of consolidated approaches. We hope that our investigations will help to overcome limited local adoption of those methods, as well as contribute to the global ML community.

Recent advances in algorithmic fairness have largely omitted sexual orientation and gender identity. We explore the concerns of the queer community in privacy, censorship, language, online safety, health, and employment to study the positive and negative effects of artificial intelligence on queer communities. These issues highlight a multiplicity of considerations for fairness research, such as privacy preservation, context sensitivity and process fairness, to an awareness of sociotechnical impact and the increasingly important role of inclusive and participatory research processes. Sexual orientation and gender identity are often instances of unobserved characteristics, breaking the assumptions of many approaches to algorithmic fairness. This talk will summarize our recent paper on the importance of developing new approaches for algorithmic fairness that break away from the prevailing assumption of observed characteristics.

While reinforcement learning has shown great promise as a viable solution to many problems such as Go and Atari, it’s application to challenges like continuous control has been arguably less successful. Control and robotics impose real world constraints in terms of data sparsity and exploration in high dimensional continuous action spaces is challenging. Faced with this, methods that allow us to inject prior knowledge about the structure of the world become increasingly important. In this talk, I will discuss how one such method that we dub behavior priors - probabilistic models that capture common movement and interaction patterns which are shared across a set of related tasks or contexts. For example the day-to-day behavior of humans comprises distinctive locomotion and manipulation patterns that recur across many different situations and goals. I will discuss how these models can be integrated into reinforcement learning schemes to facilitate multi-task and transfer learning. I will then extend these ideas to latent variable models and consider a formulation to learn hierarchical priors that capture different aspects of the behavior in reusable modules. I will discuss how such latent variable formulations connect to related work on hierarchical reinforcement learning (HRL) thereby offering an alternative perspective on existing ideas. Finally I will discuss some exciting prospects and potential directions to extend this line of research.

When companies, whether start-ups or big corporations, talk about machine learning they usually mean some kind of neural network model. Not always though: I will talk about why instead we put a lot of our efforts on probabilistic models built using Gaussian processes. When a Machine Learning course briefly covers Gaussian processes, you might go away thinking they’re just basis function interpolation, only apply when the noise is Gaussian, and don’t scale to larger datasets. Here I will discuss why these are misconceptions and show why Gaussian processes are both interesting and useful in practical applications.

Computer graphics focuses on rendering high-quality 2D images from 3D scenes, with most research focusing on simulating elements of the physical world (such as light transport models or material simulation). However, the current physically-based rendering pipeline can be computationally expensive, and more importantly, not differentiable. Computer vision investigates the inference of scene properties from 2D images, and has achieved great success with the adoption of neural networks (NNs). However, NNs make few explicit assumptions about the physical world or how images are formed from it. Therefore, they still struggle in tasks that require 3D understanding such as novel-view synthesis, re-texturing or relighting. In this talk, I will present my recent work on combining the expressiveness of NNs and the knowledge of the physical world in the tasks of neural rendering and inverse rendering.

Our streets shape us in more ways than we can imagine. In this talk, I offer a unique view into the promising applications of machine learning for street networks. This talk starts with a motivation for the problems, then I describe our research efforts in this space. Particularly, I highlight our approaches to address some of the challenges associated with employing machine learning for problem solving on street networks. This talk concludes with a reflection on current progress and ideas for future exploration. Catch you there!

Inter-domain Gaussian processes (GPs) allow for high flexibility and low computational cost when performing approximate inference in GP models. They are particularly suitable for modeling data exhibiting global structure but are limited to stationary covariance functions and thus fail to model non-stationary data effectively. We propose Inter-domain Deep Gaussian Processes, an extension of inter-domain shallow GPs that combines the advantages of inter-domain and deep Gaussian processes (DGPs), and demonstrate how to leverage existing approximate inference methods to perform simple and scalable approximate inference using inter-domain features in DGPs. We assess the performance of our method on a range of regression tasks and demonstrate that it outperforms inter-domain shallow GPs and conventional DGPs on challenging large-scale real-world datasets exhibiting both global structure as well as a high-degree of non-stationarity.

As robots move from labs and factories into human-centric spaces, it becomes progressively harder to predetermine the environments, tasks, and human interactions they will need to be able to handle. Letting these robots learn from end users via natural language is an intuitive, versatile approach to handling novel situations robustly. Grounded language acquisition is concerned with learning the meaning of language as it applies to the physical world. At the same time, physically embodied agents offer a way to learn to understand natural language in the context of the world to which it refers. In this presentation, I will give an overview of our work on joint statistical models to learn the grounded semantics of natural language describing objects, spaces, and actions, and present recent work on using simulation-to-reality approaches to learn from unconstrained human-robot interactions.