Fri 06 Jun 15:00: When is Multilinguality a Curse? Language Modeling for 350 Languages
NOTE THE UNUSUAL TIME FOR THIS SEMINAR
Language models work well for a small number of languages. For the other languages, the best existing language model is likely multilingual, still with the vast majority of the training data coming from English and a few “priority” languages. We show that in many cases, multilinguality leads to worse performance across many languages due to limited model capacity. We then train a suite of over 1,000 monolingual models for 350 languages, finding that these models can outperform multilingual models over ten times their size. However, multilinguality can also be a blessing: we train a small number of controlled bilingual models in order to study how crosslingual transfer happens. We aim to better understand transfer learning in order to better leverage multilinguality to improve language model performance for all languages.
- Speaker: Catherine Arnett and Tyler Chang (EleutherAI and UC San Diego)
- Friday 06 June 2025, 15:00-16:00
- Venue: ONLINE ONLY. Here is the Zoom link: https://cam-ac-uk.zoom.us/j/4751389294?pwd=Z2ZOSDk0eG1wZldVWG1GVVhrTzFIZz09.
- Series: NLIP Seminar Series; organiser: Suchir Salhan.
Fri 16 May 16:00: Title to be confirmed
Abstract not available
- Speaker: Prof Pedram Hassanzedeh, University of Chicago
- Friday 16 May 2025, 16:00-17:00
- Venue: MR2.
- Series: Fluid Mechanics (DAMTP); organiser: Professor Grae Worster.
Fri 30 May 16:00: PhD Students' talks
Abstract not available
- Speaker: Speakers listed in abstract in due course
- Friday 30 May 2025, 16:00-17:00
- Venue: MR2.
- Series: Fluid Mechanics (DAMTP); organiser: Professor Grae Worster.
Fri 06 Jun 16:00: Numerical simulations of multiphase flows with various complexities
Abstract not available
- Speaker: Prof Omar Matar, Imperial College London
- Friday 06 June 2025, 16:00-17:00
- Venue: https://cassyni.com/s/fmws.
- Series: Fluid Mechanics (DAMTP); organiser: Professor Grae Worster.
Fri 23 May 16:00: From Wall-Climbing Active Colloids to self-assembly of Magnetotactic Bacteria
The observation of flocks of birds, schools of fish, and swarms of bees reveals captivating examples of collective behavior in nature. Over the past decade, physicists have unveiled intriguing features in such systems, giving rise to both spectacular phenomena and fundamental questions. In this presentation, we will first explore active wetting phenomena in a suspension of self-propelled Janus colloids near a vertical wall. While classical capillary rise is governed by equilibrium surface tension, active fluids challenge this paradigm. We investigate whether analogous interfacial effects emerge in non-phase-separated active sediments, uncovering how self-propulsion modifies wetting behavior. By studying the interaction between a non-phase-separated active sediment and a wall, we uncover how self-propulsion alters wetting-like behavior, offering insights into the role of activity in interfacial processes. In the second part, we turn to magnetotactic bacteria— microswimmers equipped with intracellular magnetic nanoparticles, enabling directed motion along magnetic fields. These bacteria exhibit dual sensitivity, responding not only to magnetic fields (magnetotaxis) but also to oxygen gradients (aerotaxis), which drives them to form dense, dynamic bands. We demonstrate how the interplay of magnetic steering, chemical gradients, and hydrodynamic interactions leads to rich self-organization.
- Speaker: Prof Cottin-Bizonne, Université Lyon
- Friday 23 May 2025, 16:00-17:00
- Venue: MR2.
- Series: Fluid Mechanics (DAMTP); organiser: Professor Grae Worster.
Fri 30 May 13:00: Gravitational Wave Signatures of Dark Matter in Neutron Star Mergers
Binary neutron star mergers provide insights into strong-field gravity and the properties of ultra-dense nuclear matter. These events offer the potential to search for signatures of physics beyond the standard model, including dark matter. We present the first numerical-relativity simulations of binary neutron star mergers admixed with dark matter, based on constraint-solved initial data. Modeling dark matter as a non-interacting fermionic gas, we investigate the impact of varying dark matter fractions and particle masses on the merger dynamics, ejecta mass, post-merger remnant properties, and the emitted gravitational waves. Our simulations suggest that the dark matter morphology – a dense core or a diluted halo – may alter the merger outcome. Scenarios with a dark matter core tend to exhibit a higher probability of prompt collapse, while those with a dark matter halo develop a common envelope, embedding the whole binary. Furthermore, gravitational wave signals from mergers with dark matter halo configurations exhibit significant deviations from standard models when the tidal deformability is calculated in a two-fluid framework neglecting the dilute and extended nature of the halo. This highlights the need for refined models in calculating the tidal deformability when considering mergers with extended dark matter structures. These initial results provide a basis for further exploration of dark matter’s role in binary neutron star mergers and their associated gravitational wave emission and can serve as a benchmark for future observations from advanced detectors and multi-messenger astrophysics.
- Speaker: Violetta Sagun, University of Southampton
- Friday 30 May 2025, 13:00-14:00
- Venue: MR9/Zoom.
- Series: DAMTP Friday GR Seminar; organiser: Xi Tong.
Fri 16 May 13:00: TBC
Abstract not available
- Speaker: Benjamin Elder, Imperial College London
- Friday 16 May 2025, 13:00-14:00
- Venue: MR20/Zoom.
- Series: DAMTP Friday GR Seminar; organiser: Xi Tong.
Fri 09 May 13:00: TBC
Abstract not available
- Speaker: Robbie Hennigar, Durham University
- Friday 09 May 2025, 13:00-14:00
- Venue: MR9/Zoom.
- Series: DAMTP Friday GR Seminar; organiser: Xi Tong.
Fri 06 Jun 13:00: A Spacetime Interpretation of the Confluent Heun Functions in Black Hole Perturbation Theory
In Black Hole Perturbation Theory, confluent Heun functions appear as solutions to the radial Teukolsky equation, which governs perturbations in black hole spacetimes. While these functions are typically studied for their analytic properties, their connection to the underlying spacetime geometry has received less attention. In this talk, I will propose a spacetime interpretation of the confluent Heun functions, demonstrating how their behaviour near their singular points reflects the structure of key surfaces in Kerr spacetimes. By interpreting homotopic transformations of these functions as changes in the spacetime foliation, I will establish a connection between these solutions and various regions of the black hole’s global structure. I will also explore their relationship with the hyperboloidal formulation of the radial Teukolsky equation.
- Speaker: Marica Minucci, Bohr Inst., Copenhagen
- Friday 06 June 2025, 13:00-14:00
- Venue: Potter room/Zoom.
- Series: DAMTP Friday GR Seminar; organiser: Xi Tong.
Fri 30 May 14:00: Title to be confirmed
Abstract not available
- Speaker: Speaker to be confirmed
- Friday 30 May 2025, 14:00-15:00
- Venue: MR12, Centre for Mathematical Sciences.
- Series: Statistics; organiser: Qingyuan Zhao.
Wed 05 Nov 14:30: Title to be confirmed
Abstract not available
- Speaker: Professor Kim Jelfs, Imperial College London
- Wednesday 05 November 2025, 14:30-15:30
- Venue: Unilever Lecture Theatre, Yusuf Hamied Department of Chemistry.
- Series: Theory - Chemistry Research Interest Group; organiser: Lisa Masters.
Fri 13 Jun 14:00: Title to be confirmed
Abstract not available
- Speaker: Yining Chen (LSE)
- Friday 13 June 2025, 14:00-15:00
- Venue: MR12, Centre for Mathematical Sciences.
- Series: Statistics; organiser: Qingyuan Zhao.
Fri 06 Jun 14:00: Title to be confirmed
Abstract not available
- Speaker: Yuansi Chen (ETH Zurich)
- Friday 06 June 2025, 14:00-15:00
- Venue: MR12, Centre for Mathematical Sciences.
- Series: Statistics; organiser: Qingyuan Zhao.
Fri 16 May 14:00: Title to be confirmed
Abstract not available
- Speaker: Kosuke Imai (Harvard University)
- Friday 16 May 2025, 14:00-15:00
- Venue: MR12, Centre for Mathematical Sciences.
- Series: Statistics; organiser: Qingyuan Zhao.
Wed 28 May 15:05: Hard and soft equivariance priors via Steerable CNNs
Equivariance can enhance the data efficiency of machine learning models by incorporating prior knowledge about a problem. Thanks to their flexibility and generality, steerable CNNs are a popular design choice for equivariant networks. By leveraging concepts from harmonic analysis, these networks model symmetries through specific constraints on their learnable weights or filters. This framework facilitates the practical implementation of a wide variety of equivariant architectures – e.g. to most Euclidean isometries, including E(3), E(2) and their subgroups.
However, unknown or imperfect symmetries can sometimes lead to overconstrained weights and suboptimal performance. This challenge motivated the study of strategies to enforce softer priors into the models. In the second half of this talk, we will discuss a novel probabilistic approach to learning the degrees of equivariance in steerable CNNs. The method replaces the equivariance constraint on the weights with an expectation over a learnable distribution, which is analytically computed by leveraging its Fourier decomposition.
Link to join virtually: https://cam-ac-uk.zoom.us/j/87421957265
This talk is being recorded. If you do not wish to be seen in the recording, please avoid sitting in the front three rows of seats in the lecture theatre. Any questions asked will also be included in the recording. The recording will be made available on the Department’s webpage
- Speaker: Dr Gabriele Cesa - Qualcomm AI Research, Amsterdam
- Wednesday 28 May 2025, 15:05-15:55
- Venue: Lecture Theatre 1, Computer Laboratory, William Gates Building.
- Series: Wednesday Seminars - Department of Computer Science and Technology ; organiser: Ben Karniely.
Fri 02 May 08:45: A novel approach to the maxillary nerve: The Palatine Technique
Iliana graduated from the University of Thessaloniki in 2017. After undertaking 4 internships and spending a couple of years in first opinion practice, she decided to pursue her dream of becoming a veterinary anaesthesiologist. She started her residency at the University of Cambridge in 2023.
Chaired by Muriel Dresen and Olivier Restif
- Speaker: Iliana Antonopoulou , Department of Veterinary Medicine
- Friday 02 May 2025, 08:45-10:00
- Venue: LT2.
- Series: Friday Morning Seminars, Dept of Veterinary Medicine; organiser: Fiona Roby.
Wed 07 May 14:00: Synthesis RIG Postdoc Seminar - Dr Antti Lahdenpera and Dr Sona Krajcovicova
“Strategies for controlling enantioselectivity in radical reactions” and “Novel Synthetic Approaches for Next-Generation Therapeutics”
- Speaker: Dr Antti Lahdenpera and Dr Sona Krajcovicova
- Wednesday 07 May 2025, 14:00-15:00
- Venue: Dept. of Chemistry, Wolfson Lecture Theatre.
- Series: Synthetic Chemistry Research Interest Group; organiser: Dr. Robert J. Phipps.
Fri 30 May 13:00: Gravitational Wave Signatures of Dark Matter in Neutron Star Mergers
Binary neutron star mergers provide insights into strong-field gravity and the properties of ultra-dense nuclear matter. These events offer the potential to search for signatures of physics beyond the standard model, including dark matter. We present the first numerical-relativity simulations of binary neutron star mergers admixed with dark matter, based on constraint-solved initial data. Modeling dark matter as a non-interacting fermionic gas, we investigate the impact of varying dark matter fractions and particle masses on the merger dynamics, ejecta mass, post-merger remnant properties, and the emitted gravitational waves. Our simulations suggest that the dark matter morphology – a dense core or a diluted halo – may alter the merger outcome. Scenarios with a dark matter core tend to exhibit a higher probability of prompt collapse, while those with a dark matter halo develop a common envelope, embedding the whole binary. Furthermore, gravitational wave signals from mergers with dark matter halo configurations exhibit significant deviations from standard models when the tidal deformability is calculated in a two-fluid framework neglecting the dilute and extended nature of the halo. This highlights the need for refined models in calculating the tidal deformability when considering mergers with extended dark matter structures. These initial results provide a basis for further exploration of dark matter’s role in binary neutron star mergers and their associated gravitational wave emission and can serve as a benchmark for future observations from advanced detectors and multi-messenger astrophysics.
- Speaker: Violetta Sagun, University of Southampton
- Friday 30 May 2025, 13:00-14:00
- Venue: Potter room/Zoom.
- Series: DAMTP Friday GR Seminar; organiser: Xi Tong.
Tue 27 May 13:00: Title to be confirmed
Abstract not available
- Speaker: Giovanna Maria Dimitri, University of Siena (Italy)
- Tuesday 27 May 2025, 13:00-14:00
- Venue: Lecture Theatre 2, Computer Laboratory, William Gates Building.
- Series: Artificial Intelligence Research Group Talks (Computer Laboratory); organiser: Mateja Jamnik.
Tue 13 May 13:00: Explainable AI in Neuroscience: From Interpretability to Biomarker Discovery
Explainability plays a pivotal role in building trust and fostering the adoption of artificial intelligence (AI) in healthcare, particularly in high-stakes domains like neuroscience where decisions directly affect patient outcomes. While progress in AI interpretability has been substantial, there remains a lack of clear, domain-specific guidelines for constructing meaningful and clinically relevant explanations. In this talk, I will explore how explainable AI (XAI) can be effectively integrated into neuroscience applications. I will outline practical strategies for leveraging interpretability methods to uncover novel patterns in neural data, and discuss how these insights can inform the identification of emerging biomarkers. Drawing on recent developments, I will highlight adaptable XAI frameworks that enhance transparency and support data-driven discovery. To validate these concepts, I will present illustrative case studies involving large language models (LLMs) and vision transformers applied to neuroscience. These examples serve as proof of concept, showcasing how explainable AI can not only translate complex model behavior into human-understandable insights, but also support the discovery of novel patterns and potential biomarkers relevant to clinical and research applications.
- Speaker: Mike Mamalakis (University of Cambridge)
- Tuesday 13 May 2025, 13:00-14:00
- Venue: Lecture Theatre 2, Computer Laboratory, William Gates Building.
- Series: Artificial Intelligence Research Group Talks (Computer Laboratory); organiser: Mateja Jamnik.