Nanoscience News (<front>)

Metal-Polyphenol Meat Freshness Intelligent Monitoring Platform

In article number 2503246, Yunfei Xie, Tiancong Zhao, and co-workers propose for the first time a multi-level competitive coordination chromogenic mechanism between metal, polyphenol, and amine. The metal-polyphenol network colorimetric sensor array (MPN-CSA) developed based on this has excellent stability, specificity, and economic environmental benefits. Combined with convolutional neural network technology, it can achieve sensitive, accurate, and real-time online intelligent monitoring of meat freshness.

Residue-Free Fabrication of 2D Materials

In article number 2418669, Minyoung Lee, Changho Kim, Jae Hun Seol, and co-workers report a residue-free fabrication technique for 2D materials using van der Waals interactions. This approach allows for the isolation and precise manipulation of residue-free 2D materials while preserving their intrinsic properties. The technique enhances the performance and versatility of 2D material-based electronic and optoelectronic devices.

Stretchable Electronics

The stretchable circuit can resist compression and autonomously repair circuit cracks by filling a micropillar-embedded channel with a biphasic liquid-solid metal. More details can be found in article number 2420469 by Jian Lv, Jinyou Shao, and co-workers.

Nature Energy, Published online: 27 May 2025; doi:10.1038/s41560-025-01788-8

Na and Ti share roles

Nature Energy, Published online: 27 May 2025; doi:10.1038/s41560-025-01789-7

Contamination control

Title to be confirmed

Abstract not available

• Speaker: Dr Zac Goodwin (Oxford)
• Wednesday 04 June 2025, 16:00-17:30
• Venue: Seminar Room 3, RDC.
• Series: Theory of Condensed Matter; organiser: Bo Peng.

Add to your calendar or Include in your list

Modularity of certain trianguline Galois representations

An unpublished result of Emerton states that every trianguline representation of the absolute Galois group of Q, satisfying certain conditions, arises as a twist of the Galois representation attached to an overconvergent p-adic cuspidal eigenform of finite slope. I will outline a new approach to prove this result by patching trianguline varieties and eigenvarieties for modular forms on GL2 to establish an “R=T” theorem in the setting of rigid analytic spaces. There are several nice consequences to such a theorem, including a new approach to deduce the classicality of overconvergent eigenforms of small slope, as well as applications to the Fontaine-Mazur conjecture.

• Speaker: James Kiln (Queen Mary)
• Tuesday 27 May 2025, 14:30-15:30
• Venue: MR13.
• Series: Number Theory Seminar; organiser: Jef Laga.

Add to your calendar or Include in your list

This work systematically summarizes the development of synaptic devices with negative photoconductivity (NPC) phenomena. Material systems, device structures, and mechanisms of NPC effect-based devices are summarized for designing high-performance neuromorphic electronics. The prospect and challenge are deeply discussed for advanced application scenarios, which provides valuable guidance for next-generation optoelectronic in-sensor neuromorphic computing devices.

Abstract

The emerging optoelectronic devices with positive photoconductivity (PPC) and negative photoconductivity (NPC) have promoted the development of high-performance photodetectors, non-volatile photoelectric memory, and neuromorphic computing. With advantages of high bandwidth, low power consumption, and parallel computing, NPC effect-based optoelectronic devices show great application potential in logic gates, in-sensor computing, and artificial visual systems. Material systems, device structures, and mechanisms of NPC effect-based devices are summarized for designing high-performance neuromorphic electronics. The evaluation parameters of the photoelectric properties, memory capabilities, and synaptic plasticity of optoelectronic devices are discussed for the realization of high-efficiency neuromorphic computing. Hardware and software operation of in-sensor computing for neuromorphic computing using NPC effect-based devices are systematically summarized to provide insights into future applications. The prospect and challenge are deeply discussed for advanced application scenarios, which provides valuable guidance for next-generation optoelectronic in-sensor neuromorphic computing devices.

A self-healing, recyclable polyurethane-based conductor (DACPU/100Li) is engineered via supramolecular and dynamic covalent networks, exhibiting ionic conductivity (1.23 × 10− 3 S cm−1), tensile strength (7.62 MPa), 1200% stretchability, and tear resistance (45.6 kJ m− 2). Its sensor enables machine-learning-assisted gesture recognition (sensitivity 5.89, strain 0.1–1000%), while its triboelectric nanogenerator harvests energy (3.87 W m− 2) and supports machine-learning-assisted object recognition.

Abstract

Flexible ionic conductors hold potential for wearable sensors and energy harvesting. However, most gel-based conductors suffer from solvent evaporation and liquid leakage, limiting practical applications. Although solid-state ionic conductors mitigate these issues, achieving strong mechanics, high conductivity, self-healing, and stability remains challenging. Here, by integrating supramolecular engineering and dynamic covalent adaptive networks, a self-healing polyurethane-based solid-state ion-conductive elastomer (DACPU/100Li) with outstanding overall properties is successfully synthesized. DACPU/100Li exhibits ultrahigh ionic conductivity (1.23 × 10− 3 S cm−1) and high tensile strength (7.62 MPa), along with an elongation at break of 1200%. Additionally, it exhibits excellent tear resistance and a fracture energy of 45.6 kJ m− 2, along with 96% self-healing efficiency (after self-healing at 120 °C for 24 h), good recyclability, and stability under extreme conditions. The DACPU/100Li-based sensor has high sensitivity (5.89) and a wide strain range (0.1–1000%). Integrated with machine learning, it enables precise gesture recognition and human–machine interaction. Furthermore, the triboelectric nanogenerator based on DACPU/100Li achieves a high power density of 3.87 W m− 2. It harvests energy from body motion to power small devices and aids object recognition via machine learning. It is believed that these solid-state ion-conductive elastomers provide new opportunities for wearable electronics, energy harvesting, and ionotronics.

Advanced Materials, EarlyView.

Solar-driven interfacial water evaporation is a sustainable water treatment technology for desalination. This review highlights the advantages of biomass materials in solar evaporators, discussing their unique structures, light absorption, and thermal conductivity. It covers design principles, performance enhancement strategies, and recent advancements in biomass-based evaporators. Challenges and future directions for biomass-based evaporators are also summarized.

Abstract

Solar-driven interfacial water evaporation is a green and energy-efficient water treatment technology with diverse applications in desalination, steam power generation, and agricultural irrigation. Biomass materials have gained significant attention in solar evaporator engineering due to their unique structure, low cost, and ease of adjustment. With enhanced light absorption and high thermal conductivity, biomass materials can improve evaporation efficiency substantially, thus providing opportunities in solar evaporation applications. Therefore, in this critical review, the operating principles and design concepts of solar evaporators are first briefly discussed in terms of the photothermal conversion mechanism. Subsequently, the superiority of biomass materials in solar evaporator design is described in detail from the types of biomass and their structural properties at micro/macro scales. The design principles and corresponding performance enhancement strategies for biomass-based evaporators are also highlighted, including material selection, structural design, and thermal management techniques. Meanwhile, recent advances in biomass-based evaporators for several cutting-edge applications are comprehensively discussed. This review can provide a comprehensive reference for the relevant researchers to advance the research and application of biomass-based solar evaporators and to promote their wide application in the field of green technology.

A heterogeneous liquid-crystal superstructure is proposed to activate higher optical dimensions in soft matter. Multilevel optical information is precisely encoded into the superstructure through a new strategy called multidimensional photopatterning. Consequently, near-field full-color printing and far-field full-color holography are achieved simultaneously with brightness controllability and spin selectivity. This facilitates the exquisite construction of LC superstructures and enlightens higher-dimensional optics.

Abstract

Soft matter, featuring superior flexibility and intriguing tunability, has shown enormous potential in sensors, soft robots, and light tailoring. However, limited by its inherent structural complexity, soft matter remains uncompetitive in multidimensional and high-density information optics. Herein, a heterogeneous liquid-crystal (LC) superstructure composed of interlocked nematic and chiral LCs is designed to achieve higher-dimensional light control. Optically multidimensional photopatterning with programmable UV polarization and dosage is proposed to precisely customize both transverse and longitudinal LC arrangements, bringing in a wide range of light-matter interactions within a single micrometer-thick film. The constructed heterogeneous LC superstructure not only enables simultaneous near-field full-color printing and far-field full-color holography but also boasts brightness controllability and polarization selectivity. This low-cost photonic structure enables a high information density of ≈1.6 million hybrid-dimensional optical data per square millimeter, unlocking new capabilities in optical storage, display, and encryption. This work creates an ingenious bond between advanced photopatterning technologies and higher-level optical informatics, and pioneers soft-matter-mediated full-dimensional optics.

Latent Concepts in Large Language Models

Large Language Models (LLMs) have achieved remarkable fluency and versatility—but understanding how they represent meaning internally remains a challenge. In this talk, we explore the emerging science of latent concepts in LLMs: the semantic abstractions implicitly encoded in their internal activations.

We examine how concepts—such as truthfulness, formality, or sentiment—can be represented as low-dimensional structures, discovered through training dynamics, and understood through the lens of linear algebra and associative memory. We discuss the implications for interpretability, robustness, and control, including how concepts can be steered at test time to adjust model behavior without retraining. Specifically, we explore empirical and theoretical evidence supporting the linear representation hypothesis, where such concepts correspond to vectors or affine subspaces, emerging naturally from training dynamics and next-token prediction objectives. We further show that LLMs behave as associative memory systems, retrieving outputs based on latent similarity rather than logical inference. This behavior underlies phenomena such as context hijacking, where semantically misleading prompts can bias the model’s response.

We introduce formal latent concept models that unify these ideas, describe conditions under which concepts are identifiable, and propose learning algorithms for extracting interpretable, controllable representations. We argue that such latent concept modeling offers a principled framework for bridging representation learning with interpretability and model alignment, and offers a promising path toward safer, more controllable, and more trustworthy AI.

• Speaker: Prof. Pradeep Ravikumar, Carnegie Mellon University
• Tuesday 10 June 2025, 14:00-15:00
• Venue: JDB Seminar Room, CUED.
• Series: Signal Processing and Communications Lab Seminars; organiser: Prof. Ramji Venkataramanan.

Add to your calendar or Include in your list

Nature Nanotechnology, Published online: 26 May 2025; doi:10.1038/s41565-025-01927-y

Ferroelectric membranes of BaTiO3 can form centre-convergent polar topology domes that couple with light to generate circularly polarized beams.

Title to be confirmed

Abstract not available

• Speaker: Amila Jayasinghe, University of Cambridge
• Friday 13 June 2025, 15:00-16:00
• Venue: CivEng Seminar Room (1-33) (Civil Engineering Building).
• Series: Engineering Department Structures Research Seminars; organiser: Shehara Perera.

Add to your calendar or Include in your list

Title to be confirmed

Abstract not available

• Speaker: Keith Seffen, University of Cambridge
• Friday 20 June 2025, 15:00-16:00
• Venue: CivEng Seminar Room (1-33) (Civil Engineering Building).
• Series: Engineering Department Structures Research Seminars; organiser: Shehara Perera.

Add to your calendar or Include in your list

Title to be confirmed

Abstract not available

• Speaker: Jelena Ninic, University of Birmingham
• Friday 17 October 2025, 15:00-16:00
• Venue: CivEng Seminar Room (1-33) (Civil Engineering Building).
• Series: Engineering Department Structures Research Seminars; organiser: Shehara Perera.

Add to your calendar or Include in your list

Statistical Finite Elements via Interacting Particle Langevin Dynamics

In this work, we develop a class of interacting particle Langevin algorithms to solve inverse problems for partial differential equations (PDEs). We leverage the statistical finite elements formulation to obtain a finite-dimensional statistical model, where the parameter is that of the forward map and the latent variable is the discretised solution of the PDE , assumed to be partially observed. We then adapt a recently proposed expectation-maximisation like scheme, the interacting particle Langevin algorithm (IPLA), for this problem and obtain a joint estimation procedure for the parameters and the latent variables. The estimation of an unknown source term is demonstrated for linear and nonlinear Poisson PDEs, as well as the diffusivity parameter for the linear Poisson PDE . We provide computational complexity estimates for forcing estimation in the linear case, including comprehensive numerical experiments and preconditioning strategies that significantly improve the performance.

• Speaker: Alex Glyn-Davies, University of Cambridge, UK
• Friday 30 May 2025, 15:00-16:00
• Venue: CivEng Seminar Room (1-33) (Civil Engineering Building).
• Series: Engineering Department Structures Research Seminars; organiser: Shehara Perera.

Add to your calendar or Include in your list

Title to be confirmed

Abstract not available

• Speaker: Matthew Santer, Imperial College London, UK
• Friday 10 October 2025, 15:00-16:00
• Venue: CivEng Seminar Room (1-33) (Civil Engineering Building).
• Series: Engineering Department Structures Research Seminars; organiser: Shehara Perera.

Add to your calendar or Include in your list

Unveiling the Secret Sauce: A Causal Look at Data Memorisation and Tokenisation in Language Models

While model design gets much of the spotlight, subtle data choices, such as which documents are seen and how they’re represented, can profoundly shape the behaviour of language models. Nowadays, training data is the secret sauce behind a language model’s success, yet it remains relatively understudied. In this talk, I will discuss how training data influences a model’s behaviour via two key phenomena: memorisation and tokenisation bias. First, I’ll present our work on memorisation, asking: To what extent does a model remember specific documents it was trained on? Directly answering this question is computationally expensive. Instead, we frame memorisation as a causal question and introduce an efficient method to estimate it without re-training. This reveals how memorisation depends on factors such as data order and model size. Next, I’ll discuss how subword tokenisation, often seen as a preprocessing detail, systematically biases model predictions. We ask: How would a model’s output change if a piece of text were tokenised as one subword instead of two? Using tools from econometrics, we estimate this counterfactual question without re-training the model using a different vocabulary. We show that when a piece of text is tokenised into fewer subwords, it consistently receives a higher probability. Together, these results show that training data profoundly shapes a model’s behaviour. Causal methods let us efficiently estimate and understand these phenomena, offering insight into how to better train language models.

Bio: Pietro Lesci is a final-year PhD student in Computer Science at the University of Cambridge, working with Prof Andreas Vlachos. His research explores how training data shape a model’s behaviour, focusing on memorisation, tokenisation, and generalisation. To study this question, he draws on causal methods from econometrics. His work has been presented at major machine learning conferences such as ICLR , ACL, NAACL , and EMNLP . He has received the Best Paper Award at ACL 2024 , the Paper of the Year Award from Cambridge’s Department of Computer Science and Technology, and funding from Translated’s Imminent Research Grant. Pietro’s experience spans academia and industry, including 3+ years working in research labs, consulting firms, and international institutions. He holds an MSc in Economic and Social Sciences from Bocconi University.

• Speaker: Pietro Lesci (University of Cambridge)
• Friday 30 May 2025, 12:00-13:00
• Venue: Room FW26 with Hybrid Format. Here is the Zoom link for those that wish to join online: https://cam-ac-uk.zoom.us/j/4751389294?pwd=Z2ZOSDk0eG1wZldVWG1GVVhrTzFIZz09.
• Series: NLIP Seminar Series; organiser: Suchir Salhan.

Add to your calendar or Include in your list