A core ingredient of general artificial intelligence is the ability of AI to improve itself, including its own learning algorithm.In recent years meta-learning has made large progress in producing...

Everyone is now talking about Language Models (LMs), and arguing about whether they can reason or produce original content. Is there value in comparing the behavior of language models and humans?...

With the emergence of impressive and general-purpose chatbots, increasingly attention has concentrated on a specific training method for both Anthropic’s Claude and OpenAI's ChatGPT —...

Randomly masking and predicting word tokens has been a successful approach in pre-training language models for a variety of downstream tasks. In this work, we observe that the same idea also...

As artificial agents and robots become more advanced through modern machine learning and complex symbolic methods, testing these systems becomes increasingly difficult. Trained agents must be...

As interest in the field of open-endedness expands, ideas like continual discovery and increasingly complexity have gained significant attention. The aim of this talk is to bring attention to two...

While there has been major investment in developing robot learning algorithms, achieving true autonomy remains a wide-open research question. A key limitation of current learning approaches is the...

The capabilities of modern machine learning systems are to a large extent determined by their ability to effectively utilize large and diverse datasets. However, such systems typically focus on...

The world around us — and our understanding of it — is rich in compositional structure: from atoms and their interactions to objects and entities in our environments. How can we learn models of...

In this talk I will cover our recent publication "Open-Ended Learning Leads to Generally Capable Agents". In this work we turn our attention to how to create embodied agents in simulation that can...

Recently there has been a shift in computer vision research from static vision, e.g., object detection, to Embodied AI, e.g., robot navigation. In this talk, I will focus on the task of PointGoal...

Developing agents capable of learning complex human-like behaviors is a key goal of artificial intelligence research. Progress towards this goal has exciting potential for applications in video...

Learning agents in text-based games are faced with the challenge of understanding and generating language to accomplish various goals. Decomposing this primary challenge, I will highlight recent...

Video games have become an attractive testbed for evaluating AI systems, by capturing some aspects of real-world complexity (rich visual stimuli and non-trivial decision policies) while...

The benefit of multi-task learning over single-task learning relies on the ability to use relations across tasks to improve performance on any single task. While sharing representations is an...

In this talk, I'll give four good reasons to do embodied machine learning. Learning in an agent that can perceive and interact with its environment is fundamentally different from other ML...

Mental simulation—the capacity to imagine what will or what could be—is a salient feature of human cognition, playing a key role in a wide range of cognitive abilities. In artificial intelligence,...

The ability to cooperate through language is a defining feature of humans. As the perceptual, motory and planning capabilities of deep artificial networks increase, researchers are studying whether they can also develop a shared language to interact. In this talk, I will highlight recent advances in this field but also common headaches (or perhaps limitations) with respect to experimental setup and evaluation of emergent communication. Towards making multi-agent communication a building block of human-centric AI, and by drawing from my own recent work, I will discuss approaches on making...

How to learn good latent representations is an important topic in the modern era of machine learning. For reinforcement learning, using a good representation makes the decision-making process much...

Evolutionary algorithms are powerful black-box optimisers that find many applications in Game AI. They can be applied in real-time to provide robust policies across a range of games, or at design...

The AI's ability to answer questions that are grounded in context is interesting from both academic and practical perspectives. In this talk, I will present two projects that study question answering (QA) in visual and symbolic database context respectively. In the first project my collaborators and I show that a neuro-symbolic visual QA system can learn variable bindings from the top-down QA training signal only. The system successfully learns both conventional bindings (e.g. objects) and less trivial ones (e.g. groups). In our second project my colleagues at Element AI and I explore...

In recent years we have seen rapid progress on a number of zero-sum benchmark problems in artificial intelligence, e.g. Go, Poker and Dota. In contrast to these competitive settings, success in...

Central to tasks like instruction following and question answering is the ability to ground linguistic understanding in perception and action. Machine learning models for these tasks typically...

Julian Togelius will be giving a talk with the title: "Increasing generality in reinforcement learning through procedural content generation (or have fun trying)", exploring how to make...

Neural language models, as they grow in scale, continue to surprise us with utterly nonsensical and counterintuitive errors despite their otherwise remarkable performances on leaderboards. In this...

Social learning helps humans and animals rapidly adapt to new circumstances, and drives the emergence of complex learned behaviors. This talk focuses on Social Reinforcement Learning, developing...

Researchers in open-ended machine learning are inspired by natural and human processes of innovation (like biological evolution or science itself), and aim to uncover the engineering principles underlying boundlessly creative search algorithms, i.e. algorithms capable of continual production of useful and interesting innovations, The speculative promise for machine learning from such open-ended search includes automated discovery of curricula for reinforcement learning, new neural architectures, and most ambitiously, AGI itself. While its most ambitious potential is far from being...