Nanoscience News (<front>)

An Introduction to Transformer Neural Processes

Neural processes (NPs) have significantly improved since their inception. A principal factor in their effectiveness has been advancements in the architecture of permutation-invariant set functions—a notable example of which being transformer-based architectures. In this reading group session, we will introduce participants to Transformer Neural Processes (TNPs). We do not assume prior knowledge of NPs or transformers.

Useful background reading: 1) Transformer Neural Processes: Uncertainty-Aware Meta Learning Via Sequence Modelling. Nguyen and Grover (2022). https://arxiv.org/abs/2207.04179 2) Set Transformer: A Framework for Attention-based Permutation-Invariant Neural Networks. Lee et al. (2018). https://arxiv.org/abs/1810.00825 3) Latent Bottlenecked Attentive Neural Processes. Feng et al. (2022). https://arxiv.org/abs/2211.08458 4) Attentive Neural Processes. Kim et al. (2019). https://arxiv.org/abs/1901.05761.

Zoom link available upon request (it is sent out on our mailing list, eng-mlg-rcc [at] lists.cam.ac.uk). Sign up to our mailing list for easier reminders via lists.cam.ac.uk.

• Speaker: Matt Ashman and Cristiana Diaconu
• Wednesday 28 February 2024, 11:00-12:30
• Venue: Cambridge University Engineering Department, CBL Seminar room BE4-38..
• Series: Machine Learning Reading Group @ CUED; organiser: Isaac Reid.

Add to your calendar or Include in your list

Representing Data Collections for Analysis and Transformation

To fully leverage the capabilities of modern computer hardware, compilers become a necessity. By automatically applying hundreds of varied optimizations, compilers generate binaries that make efficient use of these capabilities. The intermediate representation (IR) of an optimizing compiler is the most important decision in its design, clearly laying out the scope and goal of optimizations. For example, the now ubiquitous use of static single assignment (SSA) forms enhances the compiler’s ability to perform data flow analysis both accurately and efficiently. However, modern optimizing compilers have turned a blind eye to the representation of memory in the program, leaving optimizations on the layout and usage of memory as a secondary consideration. This decision hands developers the arduous task of describing both the semantics and layout of their data in memory. As a result, high-level concepts like data collections are prematurely lowered to low-level memory representations. From this, the compiler can only glean conservative information about the memory in a program—e.g., alias analysis—and can make little to no strong guarantees for data collections more complex than arrays. This poses a barrier to optimization. To remedy this issue, we propose MEMOIR , a language-agnostic SSA IR for irregular data collections. At the core of MEMOIR is a decoupling of the memory used to store data from that used to logically organize data. With this decoupling, compiler optimizations can reason about the state of a collection at any given program point without ambiguity. Through its SSA form, MEMOIR generalizes a wealth of scalar optimizations to operate on data collections at the element granularity. Additionally, it opens the door for an entirely new class of memory layout and reuse optimizations that are impossible in modern compilers.

• Speaker: Tommy McMichen, Northwestern University
• Friday 08 March 2024, 10:45-11:45
• Venue: SS03, Computer Laboratory, William Gates Building.
• Series: Computer Laboratory Computer Architecture Group Meeting; organiser: Timothy M Jones.

Add to your calendar or Include in your list

Abstract

Nanotechnology offers significant advantages for medical imaging and therapy, including enhanced contrast and precision targeting. However, integrating these benefits into ultrasonography has been challenging due to the size and stability constraints of conventional bubble-based agents. Here we describe bicones, truly tiny acoustic contrast agents based on gas vesicles, a unique class of air-filled protein nanostructures naturally produced in buoyant microbes. We show that these sub-80 nm particles can be effectively detected both in vitro and in vivo, infiltrate tumors via leaky vasculature, deliver potent mechanical effects through ultrasound-induced inertial cavitation, and are easily engineered for molecular targeting, prolonged circulation time, and payload conjugation.

This article is protected by copyright. All rights reserved

Abstract

Pollution tolerance and long-term mechanical support are the two critical properties of meshes for contaminated abdominal wall defect repair. However, biological meshes with excellent pollution tolerance fail to provide bio-adaptive long-term mechanical support due to their rapid degradation. Here, a novel double-layer asymmetric porous mesh (SIS/PVA-EXO) is designed by simple and efficient in situ freeze-thaw of sticky polyvinyl alcohol (PVA) solution on the loosely porous surface of small intestinal submucosal decellularized matrix (SIS), which can successfully repair the contaminated abdominal wall defect with bio-adaptive dynamic mechanical support through only single-stage surgery. The exosome-loaded degradable loosely porous SIS layer accelerates the tissue healing, while the exosome-loaded densely porous PVA layer can maintain long-term mechanical support without any abdominal adhesion. In addition, the tensile strength and fracture strain of the SIS/PVA-EXO mesh change gradually from 0.37 MPa and 210% to 0.10 MPa and 385% with the degradation of SIS layer. This unique performance can dynamically adapt to the variable mechanical demands during different periods of contaminated abdominal wall reconstruction. As a result, our SIS/PVA-EXO mesh shows an attractive prospect in the treatment of contaminated abdominal wall defect without recurrence by integrating local immune regulation, tissue remodeling, and dynamic mechanical supporting.

This article is protected by copyright. All rights reserved

Nature Materials, Published online: 27 February 2024; doi:10.1038/s41563-024-01834-y

Solvation dynamics at picosecond timescales critically affect charge transport in aqueous systems, but conflicting values have been reported for organic electrolytes. Lifetimes on the order of 1 ns for mixtures of organic polymer and lithium salt exhibiting ultraslow dynamics of solvation shell break-up are now reported.

Nature Materials, Published online: 27 February 2024; doi:10.1038/s41563-024-01816-0

Fast charging is driving extensive research on enhanced electrodes for high-performance electrochemical capacitors and micro-supercapacitors. Thick ruthenium nitride pseudocapacitive films are shown to exhibit enhanced capacitance with a time constant of less than 6 s.

Nature Materials, Published online: 27 February 2024; doi:10.1038/s41563-024-01828-w

Controlling substrate elasticity during physical vapour deposition allows access to high-density stable glasses that would otherwise be formed under prohibitively slow deposition conditions on rigid substrates.

Nature Materials, Published online: 27 February 2024; doi:10.1038/s41563-024-01811-5

Sequential drinks of crosslinker and polymer solutions form a tough hydrogel in the stomach, enabling delivery of drugs and biologics in this harsh chemical environment.

Nature Nanotechnology, Published online: 27 February 2024; doi:10.1038/s41565-024-01622-4

Weak laser light confined at the apex of a scanning tunnelling microscope tip can drive the tautomerization of a free-base phthalocyanine with atomic-scale precision. The combination of tip-enhanced photoluminescence spectroscopy and hyperspectral mapping paired with theoretical modelling then unravel an excited-state mediated reaction.

Nature Energy, Published online: 27 February 2024; doi:10.1038/s41560-024-01467-0

Non-flammable electrolytes are essential for ensuring the safe operation of sodium-metal batteries; however, challenges arise in applications due to limited stability between the electrolytes and electrodes. Now, an electrolyte engineering approach using salts as a diluent is proposed to achieve both high interfacial stability and improved safety.

Abstract

Understanding the synergism between metal site and acid site is of great significance in boosting the efficiency of bi-functional catalysts in many heterogeneous reactions, particularly in biomass upgrading. Herein, we report a “confined auto-redox” strategy to fix CeO2-anchored Pt atoms on the inner wall of a ZSM-5 cage, achieving the target of finely controlling the placements of the two active sites. Compared with the conventional surface-supported counterpart, the encapsulated Pt/CeO2@ZSM-5 catalyst possesses remarkably-improved activity and selectivity, which can convert > 99% furfural into cyclopentanone with 97.2% selectivity in 6 hours at 160 °C. Besides the excellent catalytic performance, the ordered metal-acid distribution also makes such kind of catalyst become an ideal research subject for metal-acid interactions. The following mechanization investigation reveals that the enhancement is strongly related to the unique encapsulation structure, which promotes the migration of the reactants over different active sites, thereby contributing to the tandem reaction.

This article is protected by copyright. All rights reserved

Image-based high strain rate testing of materials

The advent of camera-based measurements like Digital Image correlation (DIC) and ultra-high speed imaging (submicrosecond interframe times) opens the way to the design of a new generation of high strain rates tests of materials. The underpinning idea is the use of acceleration fields as a volume-embedded load cell. The presentation will give an, overview of the concept and present recent results on composites and ceramics.

• Speaker: Fabrice Pierron, MatchID, Ghent, Belgium
• Thursday 09 May 2024, 15:00-16:00
• Venue: Mott Seminar Room, Cavendish Laboratory.
• Series: Physics and Chemistry of Solids Group; organiser: Stephen Walley.

Add to your calendar or Include in your list

Abstract

The eradication of osteomyelitis caused by methicillin-resistant Staphylococcus aureus (MRSA) poses a significant challenge due to its development of biofilm-induced antibiotic resistance and impaired innate immunity, which often leads to frequent surgical failure. Here, we report the design, synthesis, and performance of X-ray-activated polymer-reinforced nanotherapeutics that modulate the immunological properties of infectious microenvironments to enhance chemoradiotherapy against multidrug-resistant bacterial deep-tissue infections. Upon X-ray radiation, the proposed polymer-reinforced nanotherapeutic generates reactive oxygen species and reactive nitrogen species. To robustly eradicate MRSA biofilms at deep infection sites, these species can specifically bind to MRSA and penetrate biofilms for enhanced chemoradiotherapy treatment. X-ray-activated nanotherapeutics modulate the innate immunity of macrophages to prevent the recurrence of osteomyelitis. The remarkable anti-infection effects of these nanotherapeutics were validated using a rat osteomyelitis model. This study demonstrates the significant potential of a synergistic chemoradiotherapy and immunotherapy method for treating MRSA biofilm-infected osteomyelitis.

This article is protected by copyright. All rights reserved

Abstract

Harvesting recyclable ammonia (NH3) from acidic nitrate (NO3 −)-containing wastewater requires the utilization of corrosion-resistant electrocatalytic materials with high activity and selectivity towards acidic electrochemical nitrate reduction (NO3ER). Herein, ultrathin RhNi bimetallenes with a Rh-skin-type structure (RhNi@Rh BMLs) are fabricated towards acidic NO3ER. The Rh-skin atoms on the surface of RhNi@Rh BMLs experience the lattice compression-induced strain effect, resulting in the shortened Rh-Rh bond and a downshifted d-band center. Experimental and theoretical calculation results corroborate that Rh-skin atoms can inhibit NO2*/NH2* adsorption-induced Rh dissolution, contributing to the exceptional electrocatalytic durability of RhNi@Rh BMLs (over 400 hours) towards acidic NO3ER. RhNi@Rh BMLs also reveals an excellent catalytic performance, boasting a 98.4% NH3 Faradaic efficiency and a 13.4 mg h−1 mgcat −1 NH3 yield. Theoretical calculations reveal that compressive stress tunes the electronic structure of Rh skin atoms, which facilitates the reduction of NO* to NOH* in the process of NO3ER. The practicality of RhNi@Rh BMLs has also been confirmed in an alkaline-acidic hybrid Zn-NO3 − battery with a 1.39 V open circuit voltage and a 10.5 mW cm−2 power density. More strikingly, this work offers valuable insights into the nature of electrocatalyst deactivation behavior and guides the development of high-efficiency corrosion-resistant electrocatalysts for applications in energy and the environment.

This article is protected by copyright. All rights reserved

NNLO + Strings

The state of the art for collider phenomenology is NNLO calculations matched to Parton Showers plus hadronisation. Most current approaches for NNLO + shower matching only achieve the goal partially. I will discuss the reasons for this and present a method that does achieve a full-fledged matching between NNLO calculations and showers. I will also aim to highlight some exciting new discoveries in non-perturbative QCD and their possible implications in a string model.

• Speaker: Peter Skands
• Friday 01 March 2024, 16:00-17:00
• Venue: MR19 (Potter Room, Pavilion B), CMS.
• Series: HEP phenomenology joint Cavendish-DAMTP seminar; organiser: Nico Gubernari.

Add to your calendar or Include in your list

Abstract

Radiotherapy, a widely used therapeutic strategy for esophageal squamous cell carcinoma (ESCC), is always limited by radioresistance of tumor tissues and side-effects on normal tissues. Herein, a signature based on 4 core genes of cGAS-STING pathway, is developed to predict prognosis and assess immune cell infiltration, indicating that the cGAS-STING pathway and radiotherapy efficacy are closely intertwined in ESCC. A novel lipid-modified manganese diselenide nanoparticle (MnSe2-lipid) with extraordinarily uniform sphere morphology and tumor microenvironment (TME) responsiveness is developed to simultaneously overcome radioresistance and reduce side-effects of radiation. The uniform MnSe2 encapsulated lipid effectively achieves tumor accumulation. Octadecyl gallate (OGA) on surface of MnSe2 forming pH-responsive metal-phenolic covalent realizes rapid degradation in TME. The released Mn2+ promotes radiosensitivity by generating reactive oxygen species (ROS) induced by Fenton-like reaction and activating cGAS-STING pathway. Spontaneously, selenium strengthens immune response by promoting secretion of cytokines and increasing white blood cells (WBC), and performs antioxidant activity to reduce side-effects of radiotherapy. Overall, this multifunctional remedy which is responsive to TME is capable of providing radiosensitivity by cGAS-STING pathway-mediated immunostimulation and chemodynamic therapy (CDT), and radioprotection of normal tissues, is highlighted here to optimize ESCC treatment.

This article is protected by copyright. All rights reserved

Title to be confirmed

Abstract not available

• Speaker: Fabrice Pierron, MatchID
• Thursday 02 May 2024, 15:00-16:00
• Venue: Mott Seminar Room, Cavendish Laboratory.
• Series: Physics and Chemistry of Solids Group; organiser: Stephen Walley.

Add to your calendar or Include in your list

Renewables as Reserve Providers; Turning a Challenge into an Opportunity

The expansion of renewable energy generation increases the need for short-term reserve facilities to compensate for their short-term variations. This makes reserve markets increasingly profitable and attractive for renewable energy producers (REPs), who are facing diminishing subsidies and returns. Policymakers would also value REPs as reserve providers because conventional reserves are generally carbon-intensive. The major hurdle is, however, that reserve markets require reliability, while renewables are intermittent. This brings financial risks to REPs and reliability risks to the system operators. Two remedies to alleviate these risks are intraday trading and storage. The open question is whether, these hedging instruments, individually or combined, can resolve the financial and reliability risks and facilitate REP ’S participation in reserve markets. This is currently unknown and, among others, depends on the micro-structure of the intraday markets, mainly distinguished as discrete (D-ID) or continuous (C-ID) markets. We study this problem by formulating the operation of a profit-maximizing REP , with and without storage, providing reserve services as a multi-stage stochastic integer program, separately with the support of D-ID or C-ID markets. We combine the Benders decomposition and stochastic dual dynamic programming algorithm (SDDP) to solve the problem efficiently. Our analysis of real data from the German market provides interesting insights into REPs’ participation in short-term reserve markets. Importantly, we find that C-ID trading is the best enabler among all, facilitating the profitable and reliable participation of REPs in the FCR market. In this case, batteries not only do not help FCR participation but also worsen the reliability. Conversely, D-ID markets do not help FCR participation and REPs need batteries for reliable and profitable FCR participation. Thus, system operators should discourage the use of batteries (for REPs) in the case of C-ID markets and encourage it in the case of D-ID markets.

Bio:

Yashar Ghiassi-Farrokhfal is an Associate Professor at the Rotterdam School of Management, Erasmus University. Additionally, he serves as the academic director of smart energy and sustainability at the Erasmus Center for Data Analytics (ECDA) and holds the position of Erasmus Uni. scientific lead at the Centre of Energy System Intelligence (CESI). He obtained his Ph.D. from the University of Toronto, Canada in Electrical and Computer Engineering. He started his research in the energy domain when he was a post-doctoral fellow under the supervision of Prof. S. Keshav at UWaterloo, Canada. Since then, he has published numerous articles in esteemed journals and has organized numerous conferences and workshops in the field of energy transition. His involvement extends to various European and Dutch projects, including FlexSUS (municipal heat transition), MAGPIE (Energy transition in Ports), HyTROS (Hydrogen market), and Com2Heat (composite-based heat network). Employing a multi-disciplinary approach, he has delved into diverse facets of energy transitions such as sector coupling, microgrids, electric vehicles, hydrogen, energy storage, and market mechanisms at retail, wholesale, peer-to-peer, and bilateral levels.

• Speaker: Yashar Ghiassi-Farrokhfal, Rotterdam School of Management, Erasmus University
• Friday 15 March 2024, 13:00-14:00
• Venue: FW11, William Gates Building and Zoom link: https://cl-cam-ac-uk.zoom.us/j/4361570789?pwd=Nkl2T3ZLaTZwRm05bzRTOUUxY3Q4QT09&from=addon .
• Series: Energy and Environment Group, Department of CST; organiser: lyr24.

Add to your calendar or Include in your list

Abstract

Ultrafast manipulation of magnetic order has challenged our understanding the fundamental and dynamic properties of magnetic materials. So far single shot magnetic switching has been limited to ferrimagnetic alloys, multilayers and designed ferromagnetic heterostructures. In ferromagnetic (FM)/antiferromagnetic (AFM) bilayers, exchange bias (He ) arises from the interfacial exchange coupling between the two layers and reflects the microscopic orientation of the antiferromagnet. Here we demonstrate the possibility of single shot switching of the antiferromagnet (change of the sign and amplitude of He ) with a single femtosecond laser pulse in IrMn/CoGd bilayers. We demonstrate the manipulation in a wide range of fluences for different layer thicknesses and compositions. Atomistic simulations predict ultrafast switching and recovery of the AFM magnetization on a timescale of 2 ps. Our results provide the fastest and the most energy-efficient method to set the exchange bias and pave the way to potential applications for ultrafast spintronic devices.

This article is protected by copyright. All rights reserved

Detecting superspreaders in wildlife reservoirs of disease

To better understand the dynamics of infectious diseases of wildlife, it is crucial to be able to fit dynamic transmission models to observed data in a robust and efficient way, in order to estimate key epidemiological parameters and generate well calibrated predictive information. In practice, epidemiological events are at best only partially observed, and as such it is necessary to infer missing information alongside the model parameters as part of the inference routine, requiring computationally intensive inference algorithms where computational load increases non-linearly with population size and with increased dimensionality of the hidden states. With this in mind, we implement a recently proposed individual forward filtering backward sampling algorithm to fit a complex individual-based epidemic model to data from a large-scale longitudinal study of bovine tuberculosis in badgers. This data set, from Woodchester Park in south-west England, comprises >2,000 badgers across 34 social groups over a 40-year period. We deal with many complexities typical to endemic wildlife disease systems: incomplete sampling of individuals over time (through capture-mark-recapture events), the use of multiple diagnostic tests, spatial meta-population structures, and non-Markovian demographic aspects such as age-dependent mortality rates (with censoring), all alongside a hidden stochastic compartmental model of disease spread. The method produces full posterior distributions for the parameters, and predictive distributions for the hidden states over time for each individual, and fits in just a few hours on a desktop machine. We also propose a novel individual-level reproduction number which accounts for major sources of uncertainty of the disease system, and from it provide quantitative evidence for the presence of superspreader badgers in Woodchester Park. The inference framework is very flexible, and could be applied to other individual-level disease systems, and we will discuss future extensions to explore further important epidemiological questions.

• Speaker: Evandro Konzen (University of Warwick)
• Wednesday 28 February 2024, 14:00-15:00
• Venue: Zoom.
• Series: Worms and Bugs; organiser: Dr Ciara Dangerfield.

Add to your calendar or Include in your list