Fri 21 Feb 11:00: Assessing language-specific capabilities of LLMs: Lessons from Swedish NLP
Abstract: In this talk, I discuss benchmarking and interpreting large language models in the context of Swedish. I present a selection of work from my PhD thesis, which analyze LLMs Swedish-specific capabilities in different areas: English-Swedish language transfer, multi-task benchmarking on Swedish NLU and targeted analysis of a specific case of Swedish linguistic variation. I outline some of the challenges that arise when trying to assess the language-specific capabilities of LLMs, some lessons I’ve learned throughout from my work and give a future outlook.
Bio: Felix Morger is a software engineer and computational linguist based in Gothenburg, Sweden. He recently defended his PhD thesis, titled “In the minds of stochastic parrots: Benchmarking, evaluating and interpreting large language models”. His main research interests are in benchmarking and model interpretability of large language models with a large focus on Swedish.
- Speaker: Felix Morger (University of Gothenburg)
- Friday 21 February 2025, 11:00-12:00
- 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.
Tue 25 Feb 14:30: Close fields and the local Langlands correspondence
There is an idea, going back to work of Krasner, that p-adic fields tend to function fields as absolute ramification tends to infinity. We will present a new way of rigorizing this idea, as well as give applications to the local Langlands correspondence of Fargues–Scholze.
- Speaker: Daniel Li Huerta (MPIM)
- Tuesday 25 February 2025, 14:30-15:30
- Venue: MR13.
- Series: Number Theory Seminar; organiser: Rong Zhou.
Thu 13 Mar 16:00: Algorithmic stability for regression and classification
In a supervised learning setting, a model fitting algorithm is unstable if small perturbations to the input (the training data) can often lead to large perturbations in the output (say, predictions returned by the fitted model). Algorithmic stability is a desirable property with many important implications such as generalization and robustness, but testing the stability property empirically is known to be impossible in the setting of complex black-box models. In this work, we establish that bagging any black-box regression algorithm automatically ensures that stability holds, with no assumptions on the algorithm or the data. Furthermore, we construct a new framework for defining stability in the context of classification, and show that using bagging to estimate our uncertainty about the output label will again allow stability guarantees for any black-box model. This work is joint with Jake Soloff and Rebecca Willett.
A wine reception in the Central Core will follow this lecture
- Speaker: Rina Foygel Barber (Chicago)
- Thursday 13 March 2025, 16:00-17:00
- Venue: Centre for Mathematical Sciences MR2.
- Series: Peter Whittle Lecture; organiser: HoD Secretary, DPMMS.
Thu 24 Apr 14:00: Gibbs state preparation on digital quantum simulators
State preparation is crucial for the simulation of quantum systems. In this talk, I will discuss recent advances in sampling Gibbs states through Lindbladian time evolution. I will highlight key challenges in implementing these techniques on quantum hardware and explore potential solutions. Finally, I will examine connections to driven dissipative time-dynamics, enabling implementation on near-term quantum devices.
- Speaker: Dominik Hahn, Oxford
- Thursday 24 April 2025, 14:00-15:30
- Venue: TCM Seminar Room.
- Series: Theory of Condensed Matter; organiser: Gaurav.
Tue 25 Feb 14:30: Close fields and the local Langlands correspondence
There is an idea, going back to work of Krasner, that p-adic fields tend to function fields as absolute ramification tends to infinity. We will present a new way of rigorizing this idea, as well as give applications to the local Langlands correspondence of Fargues–Scholze.
- Speaker: Daniel Li Huerta (MPIM)
- Tuesday 25 February 2025, 14:30-15:30
- Venue: MR13.
- Series: Number Theory Seminar; organiser: Rong Zhou.
Thu 06 Mar 14:00: The Rise of Density Functional Theory......or.....Millions of Lucky Flukes (and counting)!
In this talk I shall present a personal, partial, at times irreverent but, hopefully, occasionally insightful overview of Density Functional Theory.
- Speaker: Prof. Mike C. Payne, FRS (Cambridge)
- Thursday 06 March 2025, 14:00-15:30
- Venue: TCM Seminar Room.
- Series: Theory of Condensed Matter; organiser: Bo Peng.
Wed 07 May 14:30: Title to be confirmed
Abstract not available
- Speaker: Professor Bartomeu Monserrat, University of Cambridge
- Wednesday 07 May 2025, 14:30-15:30
- Venue: Unilever Lecture Theatre, Yusuf Hamied Department of Chemistry.
- Series: Theory - Chemistry Research Interest Group; organiser: Lisa Masters.
Unravelling the Atomic Structure of a Metal‐Covalent Organic Framework Assembled from Ruthenium Metalloligands
Metal-covalent organic frameworks (MCOFs) are novel porous materials that exhibit the advantages of covalent and metal-organic frameworks. However, it is extremely difficult to resolve their atomic structure and better understand their structure-properties relation. This work utilizes a synergistic combination of advanced microscopy, spectroscopic, and computational methods to unravel the atomic mapping of a photoactive ruthenium-MCOF.
Abstract
Covalent and metal-organic frameworks (COFs and MOFs) have shown great promise in light-driven processes mainly due to their ligand-to-metal charge-separation properties, as well as having access to a diverse range of photoactive metalloligands and organic linkers. However, both frameworks present individual drawbacks that can potentially be avoided by combining both systems (metal and covalent) to produce metal-covalent organic frameworks (MCOFs), exhibiting the advantages of both material types. Yet, due to their poor crystallinity, the understanding of the structure-properties relation of MCOFs remains unclear. Herein, we report photoactive linkers in the form of a [Ru(tpy)2]2+ (tpy: 2,2′,6,2″-terpyridine) complex which covalently binds to a luminescent pyrene core to yield a new, photoactive Schiff-base MCOF. The structure, thermal, electronic, and optical properties of this novel material have been exhaustively characterized by a wide range of microscopy, spectroscopic, and computational methods. This combined experimental and computational work represents a significant step toward the fundamental understanding of the photoactive units within the framework, their hierarchical arrangement and interactions with substrates, which is essential for the future design of efficient photocatalytic materials.
High‐Security Plastic with Integrated Holographic and Phosphorescent Images
The first case of orthogonal integration of holographic and phosphorescent images is demonstrated, providing a novel paradigm for high-security level information encryption and advanced anticounterfeiting.
Abstract
Organic room temperature phosphorescence (ORTP) polymer materials have sparked considerable research interests in recent years, but their optical function is still limited for multi-mode optical imaging. Herein, a feasible and universal approach is proposed to endow ORTP polymer materials with periodic refractive index modulation functions by holographic patterning. The key to this approach is to design a two-stage stepwise crosslinking. Stage-1, with low crosslinking density (≤0.75 mol L−1), is phosphorescence-silent but can provide greater free volume for monomer diffusion and thus facilitate the patterning of refractive index modulated holograms via photopolymerization-induced phase separation. The dense crosslinking at stage-2 can turn on phosphorescence with the intensity rising by 144% when the crosslinking density increases from 3.77 to 4.12 mol L−1. The enhanced phosphorescence is primarily ascribed to the increase of conformational distortion and spin-orbit coupling of organic phosphors based on theoretical calculations. Eventually, the first example is demonstrated of holographic plastic with the unique capability of independently displaying holographic andphosphorescent images. This work not only provides a novel paradigm to impart added optical functions to ORTP polymer materials but also paves the way for the development of high-security optical materials to combat counterfeiting.
High‐Strength Gelatin Hydrogel Scaffold with Drug Loading Remodels the Inflammatory Microenvironment to Enhance Osteoporotic Bone Repair
A high-strength gelatin hydrogel with a fracture strength exceeding 10 MPa is developed, combined with rapid photocross-linking capability. This hydrogel enables the fabrication of complex and precise scaffolds through 3D printing. When loaded with roxadustat, the scaffold facilitates sustained-release therapy, effectively remodeling inflammatory microenvironments. This approach presents a promising solution for the repair of osteoporotic bone defects.
Abstract
Osteoporosis is a widespread condition that induces an inflammatory microenvironment, limiting the effectiveness of conventional therapies and presenting significant challenges for bone defect repair. To address these issues, a high-strength gelatin hydrogel scaffold loaded with roxadustat is developed, specifically designed to remodel the inflammatory microenvironment and enhance osteoporotic bone regeneration. By incorporating minimal methacrylated hyaluronic acid (HAMA) into an o-nitrobenzyl functionalized gelatin (GelNB) matrix, a gelatin hydrogel with a fracture strength of 10 MPa is achieved, providing exceptional structural stability and enabling precise scaffold fabrication through digital light processing (DLP) 3D printing. Validated through cell experiments and animal studies, the hydrogel scaffold supports cell adhesion and migration, offers excellent tissue compatibility, and is fully degradable, meeting the requirements of a therapeutic scaffold. Including roxadustat further enhances the scaffold's functionality by regulating the inflammatory microenvironment via hypoxia-inducible factor-1α (HIF-1α) signaling, significantly improving bone defect repair in osteoporotic models. This drug-loaded scaffold effectively addresses inflammation-induced limitations and enhances the regenerative capacity of the affected area, paving the way for improved therapeutic outcomes in osteoporotic bone repair.
Enhanced Exciton Delocalization in Organic Near‐Infrared Photodetectors via Solid Additive‐Mediated J‐Aggregation
This study presents an innovative solid additive approach to enhance J-aggregation and exciton delocalization in NIR-OPDs, improving their sensitivity and specific detectivity across 695–860 nm, advancing self-powered organic photodetection applications.
Abstract
Near-infrared organic photodetectors (NIR-OPDs) have emerged as increasingly significant in optoelectronics, offering unparalleled advantages for applications in health monitoring and night vision. The development of self-powered devices with low dark currents and enhanced NIR sensitivity involves complex engineering that requires careful material selection, defect state density control, and environmental consideration. In this study, an innovative approach is introduced that utilizes solid additive (DIB) to induce improvements in the J-aggregation morphology and exciton delocalization in acceptor molecules. The goal is to broaden the response spectrum of the device and augment its detection capabilities. The key findings revealed that solid additive exhibit an electrostatic affinity for acceptors, which facilitates their orderly face-to-face stacking and controls the π–π stacking distance. These enhanced intermolecular interactions lead to the delocalization of electron–hole pairs, reduced exciton recombination, and increased charge separation efficiency. Consequently, the modified devices exhibited exceptional specific detectivity, exceeding 1014 Jones across the wavelength range of 695–860 nm, thereby establishing a new standard for NIR response in organic photodetection. Overall, this study successfully addressed the compatibility challenges associated with self-powered NIR-OPDs, thereby expanding their potential applications in various settings.
Acoustic Valley Filter, Valve, and Diverter
The discovery of valley degrees of freedom in electronic and classical waves opened the field of valleytronics. The implementation of valley-based devices remains challenging in practice. Utilizing the flexibility of phononic crystals in design and fabrication, the long-desired valley devices, such as filters, valves, and diverters for acoustic waves are reported. The designed devices may serve as a basis for developing advanced valley-based devices.
Abstract
The discovery of valley degrees of freedom in electronic and classical waves opened the field of valleytronics and offered the prospect for new devices based on valleys. However, the implementation of valley-based devices remains challenging in practice. Here, by taking advantage of the flexibility of phononic crystals in design and fabrication, the realizations of valley devices, or filters, valves, and diverters for acoustic waves are reported. All the devices are configured as the structures of input and output ports bridged by channels. The phononic crystals serving as ports allow the propagation of both valley polarizations, whereas the phononic crystals serving as channels, as they are narrow, only allow the propagation of single polarizations. For valley filters that achieve valley-polarized currents, the bridge channel is simply a straight single phononic crystal, but for valley valves that can turn off the valley-polarized currents, the channel consists of two parts, allowing the propagation of opposite valley polarizations. The valley diverters have one input port, and two output ports, and thus a branched channel, and the three parts in the channel allow the propagation of the same valley polarizations, so that the energy flow can be partitioned. The results may serve as a basis for developing advanced acoustic valley devices.
Strain‐Induced Decoupling Drives Gold‐Assisted Exfoliation of Large‐Area Monolayer 2D Crystals
Au substrate induces a strain in the adhered bottommost MoS2 layer. The strain weakens the coupling at the first MoS2-MoS2 interface, making it the weakest point in the system, therefore MoS2 crystal preferentially cleaves at this interface, facilitating the selective exfoliation of large-area monolayers with size comparable to that of the parent crystal.
Abstract
Gold-assisted exfoliation (GAE) is a groundbreaking mechanical exfoliation technique that produces centimeter-scale single-crystal monolayers of 2D materials. Such large, high-quality films offer unparalleled advantages over the micron-sized flakes typically produced by conventional exfoliation techniques, significantly accelerating the research and technological advancements in the field of 2D materials. Despite its wide applications, the fundamental mechanism of GAE remains poorly understood. In this study, using MoS₂ on Au as a model system, ultra-low frequency Raman spectroscopy is employed to elucidate how the interlayer interactions within MoS2 crystals are impacted by the gold substrate. The results reveal that the coupling at the first MoS2-MoS2 interface between the adhered layer on the gold substrate and the adjacent layer is substantially weakened, with the binding force being reduced to nearly zero. This renders the first interface the weakest point in the system, thereby the crystal preferentially cleaves at this junction, generating large-area monolayers with sizes comparable to the parent crystal. Biaxial strain in the adhered layer, induced by the gold substrate, is identified as the driving factor for the decoupling effect. The strain-induced decoupling effect is established as the primary mechanism of GAE, which can also play a significant role in general mechanical exfoliations.
Polyamine‐Depleting Hydrogen‐Bond Organic Frameworks Unleash Dendritic Cell and T Cell Vigor for Targeted CRISPR/Cas‐Assisted Cancer Immunotherapy
A microfluidic-assisted hydrogen-bond organic framework (HOF) nanoplatforms depletes polyamines in the tumor microenvironment, reversing T-cell immunosuppression and enhancing dendritic cell activity. By releasing CRISPR-Cas9 and inducing neoantigen generation via carbonyl stress and mutational burden elevation, this system enables precision cancer immunotherapy through dual-action mechanisms targeting both immune vigor and tumor vulnerability.
Abstract
Polyamines have tantalized cancer researchers as a potential means to rein in the rampant growth of cancer cells. However, clinical trials in recent decades have disappointed in delivering notable progress. Herein, a microfluidic-assisted synthetic hydrogen-bond organic framework (HOF) as a polyamine-depleting nanoplatforms designed to unleash the vigor of both dendritic cells (DCs) and T cells for precision cancer immunotherapy is reported. Upon internalization by tumor cells, the loaded plasma amine oxidase (PAO) in HOF efficiently depletes polyamines, remolding the tumor microenvironment and alleviating T-cell immunosuppression. This process also generates acrolein and H2O2, triggering CRISPR-assisted neoantigen generation. Specifically, Acrolein induces carbonyl stress, increasing mutational burdens. Simultaneously, HOF leverages the energy from the bis[2,4,5-trichloro-6-(pentyloxycarbonyl)phenyl] oxalate (CPPO)-H2O2 reaction for CRET-triggered singlet oxygen production, leading to thioether bond cleavage and release CRISPR-Cas9. Once released, CRISPR-Cas9 knocks out the DNA mismatch repair (MMR)-related MLH1 gene, further elevating mutational burdens and generating neoantigens, ideal targets for DCs. This dual-action strategy not only corrects T-cell immunosuppression but also enhances DC efficacy, presenting a powerful approach for tumor immunotherapy.
“Water‐In‐Oil” Electrolyte Enabled by Microphase Separation Regulation for Highly Reversible Zinc Metal Anode
A microphase separated “water-in-oil” (W/O) electrolyte is proposed to simultaneously regulate the water molecules in both electrolyte bulk and electrode interface. Benefiting from the directed movement and reversible demulsification of micelles under the electric field, the as-developed zinc metal batteries using W/O electrolyte exhibits extended cyclability, improve interface dynamics and negligible H2 evolution.
Abstract
The sustained hydrogen evolution and zinc (Zn) dendrite growth greatly impede the practical application of low-cost aqueous Zn metal batteries (ZMBs). Herein, for the first time, a microphase separation strategy is proposed to construct a ″water-in-oil (W/O) electrolyte to endow durable ZMBs. As validated by theoretical modeling and experimental characterizations, the unique reverse micelle structure within the electrolyte not only disrupts water hydrogen bonding and efficiently inhibits the water consumption at Zn anode, but also undergoes directed movement and reversible demulsification under electric field, thus enhancing the anode desolvation kinetics and inhibiting the interfacial side reactions. Owing to the simultaneous regulation of water molecules in both electrolyte bulk and anode interface, this W/O electrolyte achieves a high Zn plating/stripping Coulombic efficiency of 99.76% over 6000 cycles, and maintains an extend lifespan in Zn||V10O24·12H2O (VOH) cells with negligible hydrogen evolution and dendrite formation. These key findings are expected to promote the electrolyte engineering toward reversible ZMBs.
Efficient Harvesting Waste Heat by Zn‐Ion Battery Under Thermally Regenerative Electrochemical Cycles
Herein, an effective strategy is proposed to efficiently harvest waste heat by Zn-ion battery under thermally regenerative electrochemical cycles. A record high temperature coefficient of 2.944 mV K−1 is obtained, leading to the high relative Carnot efficiency of 26.08% and energy efficiency of 104.11% when the battery is charged at 50 °C and discharged at 5 °C.
Abstract
Typical technologies that can convert waste heat into electricity include thermoelectrics, thermionic capacitors, thermo-cells, thermal charge cells, and thermally regenerative electrochemical cycles. They have small thermal-to-electrical conversion efficiency or poor stability, severely hindering the efficient recovery of waste heat. Herein, a thermally regenerative electrochemical Zn-ion battery to work under Carnot-like mode to efficiently harvest waste heat into electricity is successfully developed. Through introducing Layered Double Hydroxides to modify the battery's anode reaction, a record absolute high temperature coefficient of 2.944 mV K−1 is achieved in NiHCF/Zn battery, leading to a high thermal-to-electrical conversion efficiency of 26.08% of the Carnot efficiency and an extraordinary energy efficiency of 104.11% when the battery is charged at 50 °C and discharged at 5 °C. This work demonstrates that thermally regenerative electrochemical batteries can effectively harvest waste heat to provide a powerful energy conversion technology.
Antibody‐level Bacteria Grabbing by “Mechanic Invasion” of Bioinspired Hedgehog Artificial Mesoporous Nanostructure for Hierarchical Dynamic Identification and Light‐Response Sterilization
Through the de novo design of bioinspired hedgehog artificial mesoporous nanostructure with virus-like spiky topography (mAPt), the antibody-level bacteria grabbing attributed to the “mechanic invasion”-induced hierarchical dynamic identification ranging from rough surface contact to penetration fixation is verified. Further integrated with spiky-induced near-superblack characteristic and plasmonic “hot spot”, mAPt enables advanced applications in immunoassay and photothermal sterilization.
Abstract
The interactions exploration between microorganisms and nanostructures are pivotal steps toward advanced applications, but the antibody-level bacteria grabbing is limited by the poor understanding of interface identification mechanisms in small-sized systems. Herein, the de novo design of a bioinspired hedgehog artificial mesoporous nanostructure (core–shell mesoporous Au@Pt (mAPt)) are proposed to investigate the association between the topography design and efficient bacteria grabbing. These observations indicate that virus-like spiky topography compensates for the obstacles faced by small-sized materials for bacteria grabbing, including the lack of requisite microscopic cavities and sufficient contact area. Molecular dynamics simulation reveals that spiky topography with heightened mechano-invasiveness (6.56 × 103 KJ mol−1) facilitates antibody-level bacteria grabbing, attributed to the “mechanic invasion”-induced hierarchical dynamic identification ranging from rough surface contact to penetration fixation. Furthermore, light reflectance and finite element calculation confirmed that mAPt exhibits near-superblack characteristic and plasmonic hot spot, facilitating enhanced photothermal conversion with power dissipation density at 2.04 × 1021 W m−3. After integrating the hierarchical dynamic identification with enhanced light response, mAPt enables advanced applications in immunoassay with 50-fold sensitivity enhancement and over 99.99% in vitro photothermal sterilization. It is anticipated that this novel biomimetic design provides a deeper understanding of bacteria grabbing and a promising paradigm for bacteria combating.
Tribology in Nature: Inspirations for Advanced Lubrication Materials
The advancement of lubricating materials is essential across multiple fields, including biomedicine and mechanical manufacturing. Insights drawn from natural lubrication are invaluable for designing innovative lubrication materials. This review examines the lubrication mechanisms in natural organisms and highlights recent advancements in biomimetic soft matter lubricating materials, providing a valuable resource for research in advanced lubrication technologies.
Abstract
Friction-induced energy consumption is a significant global concern, driving researchers to explore advanced lubrication materials. In nature, lubrication is vital for the life cycle of animals, plants, and humans, playing key roles in movement, predation, and decomposition. After billions of years of evolution, natural lubrication exhibits remarkable professionalism, high efficiency, durability, and intelligence, offering valuable insights for designing advanced lubrication materials. This review focuses on the lubrication mechanisms of natural organisms and significant advancements in biomimetic soft matter lubrication materials. It begins by summarizing common biological lubrication behaviors and their underlying mechanisms, followed by current design strategies for biomimetic soft matter lubrication materials. The review then outlines the development and performance of these materials based on different mechanisms and strategies. Finally, it discusses potential research directions and prospects for soft matter lubrication materials. This review will be a valuable resource for advancing research in biomimetic lubrication materials.
Nanocrystalline Perovskites for Bright and Efficient Light‐Emitting Diodes
Nanocrystalline perovskites enable efficient, high-color-purity future displays through advanced material engineering. This review highlights their roles in improving PeLED efficiency and stability, focusing on three types: polycrystalline perovskites, quasi-2D perovskites, and perovskite nanoparticles. It discusses material design strategies, exciton recombination dynamics, development trends, challenges, and pathways for commercialization.
Abstract
Nanocrystalline perovskites have driven significant progress in metal halide perovskite light-emitting diodes (PeLEDs) over the past decade by enabling the spatial confinement of excitons. Consequently, three primary categories of nanocrystalline perovskites have emerged: nanoscale polycrystalline perovskites, quasi-2D perovskites, and perovskite nanocrystals. Each type has been developed to address specific challenges and enhance the efficiency and stability of PeLEDs. This review explores the representative material design strategies for these nanocrystalline perovskites, correlating them with exciton recombination dynamics and optical/electrical properties. Additionally, it summarizes the trends in progress over the past decade, outlining four distinct phases of nanocrystalline perovskite development. Lastly, this review addresses the remaining challenges and proposes a potential material design to further advance PeLED technology toward commercialization.
Author Correction: Predesigned perovskite crystal waveguides for room-temperature exciton–polariton condensation and edge lasing
Nature Materials, Published online: 20 February 2025; doi:10.1038/s41563-025-02179-w
Author Correction: Predesigned perovskite crystal waveguides for room-temperature exciton–polariton condensation and edge lasing