As part of the reform of Part I undergraduate teaching, the department plans to deploy an information theory lecture that will be taken by all engineering students in the first year of their course. Introductions to information theory typically cover Shannon’s two theorems for source compression and for data transmission over noisy channels. While the former can easily be proved using elementary mathematics, the latter poses a challenge because it is usually derived using more advanced concepts of probability theory that are beyond the ability of first year students. We will show how a broad understanding of the noisy channel coding theorem can be achieved based on the simple example of linear coding for the binary erasure channel, giving students a solid intuition of why the theorem works, without getting bogged down in long derivations that are required to prove the theorem in its general form.

Robots are more than tools; they can mirror human design. By embedding principles of movement, perception, and interaction into machines, we externalise and test our own assumptions about intelligence, embodiment, and adaptation. This talk will explore how robots can serve as proxies to study how humans are designed—revealing both the constraints and the creative possibilities that shape us. Drawing on examples from soft robotics, embodied intelligence, and human–robot interaction, I will show how robotic systems expose hidden aspects of our own biology and behaviour, from sensorimotor coordination to social communication. By treating robots as experimental models, much like biologists use organisms, we generate new insights into the dynamics of human design.

Different neurons in the brain prefer different stimuli. In the absence of adaptation, when a given stimulus becomes more prevalent in the environment, neurons responsive to it will be more active on average. However, by adjusting the gain of neurons, adaptation can yield firing rate homeostasis and stabilise the average activity of neurons at fixed levels, despite changes in stimulus statistics. Such homeostatic adaptation has been observed in the primary visual cortex. I will show how the efficient coding hypothesis — the hypothesis that neural coding is optimized to minimize metabolic energy expenditure without compromising coding fidelity — provides a normative explanation for homeostatic adaptation.

Adaptive cognition relies on cooperation across anatomically distributed brain circuits. However, specialised neural systems are also in constant competition for limited processing resources. How does the brain’s network architecture enable it to balance these cooperative and competitive tendencies? Here we use computational whole-brain modelling to integrate multimodal structural and functional data, and examine the dynamical and computational relevance of cooperative and competitive interactions in the mammalian connectome. Across human, macaque, and mouse we show that the architecture of the models that most faithfully reproduce brain activity, consistently combines modular cooperative interactions with diffuse, long-range competitive interactions.

Robotic manipulation today often succeeds only when relying on accurate models or large datasets and heavy learning pipelines. However, are these approaches scalable? Can we achieve effective manipulation with partial information, limited data, and guaranteed safety and stability? In this talk, we present Virtual Model Control (VMC) as a practical and effective alternative. By embedding task-specific virtual components—such as mass, springs and dampers—VMC enables robots to exhibit adaptive and robust behaviors without explicit trajectory planning. We demonstrate its capability across a range of tasks, from simple reaching motions to contact-rich operations such as rock chopping, showing compliance and resilience under uncertainty. Finally, we explore new extensions including data-driven virtual components for enhanced performance and event-based variable stiffness mechanisms.

We present Free energy Estimators with Adaptive Transport (FEAT), a novel framework for free energy estimation -- a critical challenge across scientific domains. FEAT leverages learned transports implemented via stochastic interpolants and provides consistent, minimum-variance estimators based on escorted Jarzynski equality and controlled Crooks theorem, alongside variational upper and lower bounds on free energy differences. Unifying equilibrium and non-equilibrium methods under a single theoretical framework, FEAT establishes a principled foundation for neural free energy calculations. Experimental validation on toy examples, molecular simulations, and quantum field theory demonstrates improvements over existing learning-based methods.

Meeting today’s climate and energy goals requires reducing fossil fuel dependence and integrating renewable energy on a large scale. Wind power is central to this transition, but its connection to the grid brings unique challenges. Unlike traditional synchronous generators, wind turbines connect through power electronic converters, which allow flexible control but also introduce high-frequency switching harmonics. To suppress these, LCL filters are widely used. While effective, they also create high-frequency resonant frequencies that can destabilise the system if not properly controlled. Most turbines currently operate with grid-following control, where the converter simply injects available power without helping stabilise voltage or frequency. However, as conventional generators retire, wind turbines are increasingly required to provide these stability services. This has led to grid-forming control, in which converters actively set voltage and frequency like synchronous machines. Although promising, grid-forming control can interact unfavourably with LCL filters, amplifying resonance and making stable operation more difficult. Understanding these interactions and applying effective damping strategies is therefore essential. This presentation will first cover the fundamentals of wind power integration, control strategies, and LCL resonance. It will then explain how grid-forming control affects resonance, before discussing practical damping solutions to ensure stable, efficient, and grid-friendly wind power.

The pinnacle of international efforts on climate change, the Paris Agreement, is highly unlikely to succeed. But what would a credible solution have to look like? This talk has two parts; first I'll argue that an appropriate framing of the problem is in terms of managing a shared resource: cooperating over the atmosphere. Secondly, I'll present a concrete proposal for how this could be implemented in practice: the Themis Mechanism.