We demonstrate a high-sensitivity fiber-optic sensor formed by cascading two Fabry-Perot (FP) interferometers (FP1 and FP2) for precise refractive index (RI) prediction. Both FP1 and FP2 are fabricated directly on the fiber tip via two-photon polymerization (2PP) based 3D printing, with an open-ended structure designed for operation in liquid environments. Meanwhile, the cascading FP2 with FP1 exploits the vernier effect, which amplifies the sensitivity of FP1, and experimental results show high sensitivity. The sensitivity of FP1 is 1143.78 nm/RIU (0.162 nm/(mg/mL)), while the vernier–effect of the sensor achieves a higher sensitivity of 3478.23 nm/RIU (0.494 nm/(mg/mL)), representing a 3.041–fold enhancement. A physics-guided data augmentation strategy is introduced to train SpecCMamba, and a hybrid deep learning model that integrates a convolutional neural network for local feature extraction with a Mamba encoder for global sequence modeling. The SpecCMamba enables precise RI prediction, effectively handling spectral overlap and non-linearity in the data. The model achieves a determination coefficient (R2) of 0.9996 and a Root Mean Square Error of 2.855 × 10−5. With its simple structure and high sensitivity, the proposed sensor assisted by the SpecCMamba offers significant potential for applications in highly precise refractive index prediction and related fields.
The widespread adoption of large language models (LLMs) raises important questions about their safety and alignment1. Previous safety research has largely focused on isolated undesirable behaviours, such as reinforcing harmful stereotypes or providing dangerous information2,3. Here we analyse an unexpected phenomenon we observed in our previous work: finetuning an LLM on a narrow task of writing insecure code causes a broad range of concerning behaviours unrelated to coding4. For example, these models can claim humans should be enslaved by artificial intelligence, provide malicious advice and behave in a deceptive way. We refer to this phenomenon as emergent misalignment. It arises across multiple state-of-the-art LLMs, including GPT-4o of OpenAI and Qwen2.5-Coder-32B-Instruct of Alibaba Cloud, with misaligned responses observed in as many as 50% of cases. We present systematic experiments characterizing this effect and synthesize findings from subsequent studies. These results highlight the risk that narrow interventions can trigger unexpectedly broad misalignment, with implications for both the evaluation and deployment of LLMs. Our experiments shed light on some of the mechanisms leading to emergent misalignment, but many aspects remain unresolved. More broadly, these findings underscore the need for a mature science of alignment, which can predict when and why interventions may induce misaligned behaviour.
Line-wise Shannon entropy profiling of scanning electron microscopy (SEM) cross-sections was applied to quantify through-thickness heterogeneity in amorphous TiCrNiZrMoW high-entropy alloy coatings. Entropy was used to capture local grey-level variability as a proxy for packing density and structural complexity. Coatings were deposited by pulsed magnetron sputtering at two powers (600 and 1000 W) and two modulation frequencies (10 and 1000 Hz). All films remained fully amorphous with consistent short-range order, whereas power-dependent shifts in medium-range order and entropy were observed. Lower power produced thinner, more heterogeneous layers that combined higher hardness (10.5 GPa) with higher modulus (≈ 160 GPa), whereas higher power produced thicker, more uniform films with reduced stiffness. Line-wise entropy was found to correlate strongly with depth-resolved nanoindentation, supporting a quantitative association between microstructural heterogeneity and mechanical gradients. These results demonstrate that image entropy can serve as a rapid metric for optimization of amorphous-coating processes.
In the era of growing demand for sustainable solutions in construction, increasing attention is being paid to the potential use of waste materials as components of building composites. This article presents the results of a study on the impact of ground polyurethane foam waste on the mechanical properties and durability of cement mortars. The waste, derived from industrial production processes, was used as a partial replacement for fine aggregates in various proportions. The analysis included bulk density, compressive and flexural strengths, water absorption, and resistance to freeze-thaw cycles. The results indicate that adding waste reduces the density of the mortar, which can be advantageous in applications requiring lightweight materials. The most favourable balance of strength retention, density reduction, and frost resistance was observed with a 1% addition, as the mortar maintained good mechanical performance and freeze-thaw durability while achieving reduced weight. Higher waste content (2-3%) led to significant deterioration of the mechanical properties due to increased porosity. All samples exhibited increased strength after 25 freeze-thaw cycles, possibly due to continued hydration under moist low-temperature conditions. The analysis of the microstructure of cement coatings with the addition of polyurethane foam enabled the explanation of the causes of the observed changes in physico-mechanical properties resulting from ageing factors. This study suggests that small amounts of waste can be effectively used to produce lightweight and environmentally friendly construction materials, supporting circular economy practices.
For a tree decomposition $\mathcal{T}$ of a graph $G$, by $\mu(\mathcal{T})$ we denote the size of a largest induced matching in $G$ all of whose edges intersect one bag of $\mathcal{T}$. Induced matching treewidth of a graph $G$ is the minimum value of $\mu(\mathcal{T})$ over all tree decompositions $\mathcal{T}$ of $G$. Yolov [SODA 2018] proved that Max Weight Independent Set can be solved in polynomial time for graphs of bounded induced matching treewidth. In this paper we explore what other problems are tractable in such classes of graphs. As our main result, we give a polynomial-time algorithm for Min Weight Feedback Vertex Set. We also provide some positive results concerning packing induced subgraphs, which in particular imply a PTAS for the problem of finding a largest induced subgraph of bounded treewidth. These results suggest that in graphs of bounded induced matching treewidth, one could find in polynomial time a maximum-weight induced subgraph of bounded treewidth satisfying a given CMSO$_2$ formula. We conjecture that such a result indeed holds and prove it for graphs of bounded tree-independence number, which form a rich and important family of subclasses of graphs of bounded induced matching treewidth. We complement these algorithmic results with a number of complexity and structural results concerning induced matching treewidth.
