Repository for Publications and Research Data
News from the ETH Library
Recently Added
Excess endopolyploidy associated with cellular defects in neo-autotetraploid Arabidopsis arenosa is in part corrected by a naturally evolved allele of CYCD3;2
Item type: Journal Article
Meschichi, Anis; Dukić, Marinela; Schmid, Lars C.; et al. (2026)
Endopolyploidy (within-organism polyploidy) is an often-beneficial aspect of development. But polyploids, which have a higher baseline genome content, have been reported to have less endoreduplication than diploid relatives. Why has remained unclear; we explored this in Arabidopsis arenosa.
We measured endopolyploidy levels and phenotyped cellular traits, comparing diploids, neo-autotetraploids, and established autotetraploids of A. arenosa. We used a genetic approach to ask whether CYCD3;2, a polyploidy regulator with evidence of selection in tetraploids, might be responsible for endopolyploidy-related phenotypic differences between neo- and established polyploids, and whether there are associated costs.
Neo-tetraploids had the most 16C and 32C cells, whereas established tetraploids have proportions of 16C and 32C cells comparable to diploids. Tetraploid plants homozygous for a novel allele of CYCD3;2 have reduced endopolyploidy compared to homozygotes for the diploid allele and, surprisingly, also show cellular defects seen in neo-tetraploids. Suggesting this link is specific, we find that A. thaliana cycd3;2 mutants have not only increased endopolyploidy but also root hair defects like neo-tetraploids.
Our results suggest reverting endopolyploidy in polyploids to diploid-like levels could be advantageous, and that one driver might be cellular defects that arise with excess high-ploidy cells.
Nanostructure and oxidation rate of soot emissions from spray combustion of jet and sustainable aviation fuels
Item type: Journal Article
Moularas, Constantinos; Weber, Alex; Trivanovic, Una; et al. (2026)
Here, the oxidation dynamics are elucidated for soot nanoparticles generated by enclosed spray combustion of a standard jet fuel, as well as its blends with a sustainable aviation fuel (SAF). The particle size, nanostructure and organic carbon (OC) content of soot that determine its reactivity with O2 are characterized by state-of-the-art techniques. Oxidation of soot with air at 823–1173 K is investigated using thermogravimetric analysis accounting for internal and external O2 diffusion, as well as the soot porosity and composition. The nanostructure, OC content, oxidation activation energy, EA, and specific rate, ω, measured here for soot from enclosed spray combustion of jet fuel are in excellent agreement with those measured in literature for jet fuel combustion in premixed flames and aircraft engines. Blending jet fuel with 30–50 vol % SAF produces more disordered soot that contains more OC. This reduces EA up to about 13 kJ/mol and increases ω by a factor of 2.1 ± 0.9, on average. Overall, the ω obtained here for disordered, OC-rich soot from combustion of SAF/jet fuel blends is about 4 times larger than that predicted by the classic rate derived by Nagle and Strickland-Constable for graphite oxidation. In this regard, the average ω enhancement factor derived here can be interfaced with a literature relation for the jet fuel soot oxidation rate to account for the nanostructure and OC content of soot emissions from SAF/jet fuel blends. This can assist the design and optimization of aircraft engines for combustion of jet fuel blends with SAF to eliminate soot emissions and their climate and public health effects.
Security and Privacy of Machine Learning Systems
Item type: Doctoral Thesis
Edoardo, Debenedetti (2026)
Machine learning models, from image classifiers to large language models, are deployed throughout the real world, with important consequences for the security and privacy of their users. Such models are easily fooled by adversarially chosen inputs, leak their sensitive training data, and, once turned into agents, can be hijacked into taking unintended actions or leaking the private data they can access. Almost all work on these threats locates them in the model itself, yet in any real deployment the model is wrapped in a system that shapes its training and mediates its inputs, outputs, and tools.
In this dissertation, we argue that the security and privacy of a machine learning deployment is a property of this system, not of the model in isolation. This matters for two reasons: the surrounding system reshapes the attack surface, raising the cost of some attacks while widening the scope of others, but it can also be where a defense is built.
In Part I, we study predictive systems and show that a deployed classifier is never the bare function studied in most prior work. We first show that, when a classifier flags and counts "bad" queries, the cost that matters is the number of flagged queries rather than the total, so evasion attacks must be redesigned to be stealthy. We then show that components meant to improve an ML system, such as deduplication and query filters, instead open privacy side channels that leak far more than the isolated model would.
In Part II, we study agentic systems, where the central threat is prompt injection: text in the data an agent reads that is mistaken for instructions. We first build AgentDojo, a dynamic environment that evaluates a tool-calling agent as a whole, and find that even state-of-the-art agents are vulnerable. We then build CaMeL, a defense that confines the model within a system layer that follows only the control and data flow of the trusted query and tracks the provenance of every value, neutralizing a wide class of prompt injection attacks by construction.
We hope that these results encourage the security and privacy of machine learning deployments to be studied, measured, and enforced at the level of the system rather than the model alone.
Micro-spectroscopic investigations of corrosion processes at the iron-cement interface
Item type: Other Conference Item
Shaik, Najma; Dähn, Rainer; Mundra, Shishir; et al. (2026)
The economic impact of steel corrosion in reinforced concrete infrastructure is estimated at approximately 250 billion EUR annually in the EU alone.[1] The transport and retention of Fe ions released during corrosion significantly impact overall corrosion rates and the long-term durability of reinforced concrete structures[2]. Despite its importance, the physicochemical parameters controlling corrosion processes remain poorly understood. In particular, aqueous complexation of Fe(II)/Fe(III) in the cementitious environment, metastable solid phase formation of Fe(II)/Fe(III), and confinement effects on Fe(II)/Fe(III) redox kinetics influence the precipitation and spatial distribution of Fe(II)/(III) corrosion products within the concrete matrix.[3]. Accordingly, this study addresses microscale aspects of corrosion occurring at the interface between iron and cement, by using advanced high resolution micro-spectroscopic techniques. To cover a broad range of relevant conditions, the study includes both samples prepared in the laboratory under controlled conditions (RH, Temperature, CO2) and field-extracted samples from existing structures. The laboratory samples are produced by embedding steel wires in cementitious media and are subjected to carbonation corrosion through wet-dry cycles, while corrosion behavior is monitored using open circuit potential (OCP), linear polarization resistance (LPR), and electrochemical impedance spectroscopy (EIS) measurements. The carbonated mortar samples are then prepared into thin cross-sections (~30μm) and characterized using Raman Microscopy, SEM-EDS and μ-XRF techniques. This study also employs advanced synchrotron-based micro-spectroscopic techniques, including μXRF, μXRD, and μXAS, to examine the speciation of iron in solid corrosion products at the iron-cement interface and to resolve microscale Fe distribution in cement matrix. This work can help developing a mechanistic understanding of the corrosion processes at iron- cement interface and provide information to validate reactive-transport models.
Balancing sensitivity and integration in capacitive pressure sensors for textile-based health monitoring
Item type: Journal Article
Papaefstathiou, Maria; Elgendi, Mohamed; Menon, Carlo (2025)
Capacitive pressure sensors (CPSs) are increasingly important for wearable and textile-based health monitoring due to their high sensitivity, low power consumption, and structural flexibility. Building on a structured literature review (2015–2025), we conducted a systematic analysis of 34 CPS designs across five categories—microstructuring, foams, ionic liquids/gels/metals, bioinspired architectures, and multisensing/strong bonding—using a newly developed Textile Suitability Score (TSS) that integrates nine performance and integration-relevant attributes. Trade-off analysis revealed that ionic metal-based sensors dominate in raw performance, combining very high sensitivity with broad pressure ranges, but face scalability and textile-compatibility challenges. By contrast, microstructuring offers the closest balance between sensitivity and integration, while multisensing and strong bonding form a consistent cluster with the highest TSS values. Foams and bioinspired designs ranked lower due to reproducibility and stability issues. Together, these results expose clear performance-integration trade-offs and uncover unexplored design pathways, charting a roadmap for next-generation textile-integrated health monitoring systems.