This paper introduces DeeCLIP, a novel framework for detecting AI-generated images using CLIP-ViT and fusion learning. Despite significant advancements in generative models capable of creating highly photorealistic images, existing detection methods often struggle to generalize across different models and are highly sensitive to minor perturbations. To address these challenges, DeeCLIP incorporates DeeFuser, a fusion module that combines high-level and low-level features, improving robustness against degradations such as compression and blurring. Additionally, we apply triplet loss to refine the embedding space, enhancing the model's ability to distinguish between real and synthetic content. To further enable lightweight adaptation while preserving pre-trained knowledge, we adopt parameter-efficient fine-tuning using low-rank adaptation (LoRA) within the CLIP-ViT backbone. This approach supports effective zero-shot learning without sacrificing generalization. Trained exclusively on 4-class ProGAN data, DeeCLIP achieves an average accuracy of 89.00% on 19 test subsets composed of generative adversarial network (GAN) and diffusion models. Despite having fewer trainable parameters, DeeCLIP outperforms existing methods, demonstrating superior robustness against various generative models and real-world distortions. The code is publicly available at this https URL for research purposes.

The advancement of Deep Learning (DL) is driven by efficient Deep Neural Network (DNN) design and new hardware accelerators. Current DNN design is primarily tailored for general-purpose use and deployment on commercially viable platforms. Inference at the edge requires low latency, compact and power-efficient models, and must be cost-effective. Digital processors based on typical von Neumann architectures are not conducive to edge AI given the large amounts of required data movement in and out of memory. Conversely, analog/mixed signal in-memory computing hardware accelerators can easily transcend the memory wall of von Neuman architectures when accelerating inference workloads. They offer increased area and power efficiency, which are paramount in edge resource-constrained environments. In this paper, we propose AnalogNAS, a framework for automated DNN design targeting deployment on analog In-Memory Computing (IMC) inference accelerators. We conduct extensive hardware simulations to demonstrate the performance of AnalogNAS on State-Of-The-Art (SOTA) models in terms of accuracy and deployment efficiency on various Tiny Machine Learning (TinyML) tasks. We also present experimental results that show AnalogNAS models achieving higher accuracy than SOTA models when implemented on a 64-core IMC chip based on Phase Change Memory (PCM). The AnalogNAS search code is released: this https URL

Microfabricated alkali vapor cells enable the miniaturization of atomic sensors, but require collective wafer-level integration of complex features. In many applications, including magnetometers, gyroscopes, magneto-optical traps, and fluorescence imaging, multiple optical accesses are needed to enhance performance. Yet, achieving this without compromising manufacturability remains challenging. In this work, we present a wafer-level fabrication approach that enables three orthogonal optical pathways in microfabricated alkali vapor cells, using fully scalable and collective processes. Our method relies on the thermal reflow of glass preforms, shaped by laser-assisted etching (LAE) and bonded between silicon frames. The relatively low surface roughness produced by LAE allows effective reflow, which further smooths the surfaces without significantly compromising the optical planarity of the windows. This process results in multi-axis vapor cells featuring embedded, optics-grade lateral windows. We evaluate the device performance through both single-beam and dual-beam atomic magnetometry measurements. Magnetic sensitivities better than 200 fT/sqrt(Hz) are demonstrated along each of the three orthogonal axes, confirming the potential of the approach for tri-axis magnetic field sensing at sub-picotesla resolution. This fabrication strategy opens new perspectives for versatile, high-performance atomic sensors, fully compatible with wafer-level integration and mass production.

Vision-Language Models (VLMs) like CLIP offer promising solutions for Dynamic Facial Expression Recognition (DFER) but face challenges such as inefficient full fine-tuning, high complexity, and poor alignment between textual and visual representations. Additionally, existing methods struggle with ineffective temporal modeling. To address these issues, we propose PE-CLIP, a parameter-efficient fine-tuning (PEFT) framework that adapts CLIP for DFER while significantly reducing trainable parameters while maintaining high accuracy. PE-CLIP introduces two specialized adapters: a Temporal Dynamic Adapter (TDA) and a Shared Adapter (ShA). The TDA is a GRU-based module with dynamic scaling that captures sequential dependencies while emphasizing informative temporal features and suppressing irrelevant variations. The ShA is a lightweight adapter that refines representations within both textual and visual encoders, ensuring consistency and efficiency. Additionally, we integrate Multi-modal Prompt Learning (MaPLe), introducing learnable prompts for visual and action unit-based textual inputs, enhancing semantic alignment between modalities and enabling efficient CLIP adaptation for dynamic tasks. We evaluate PE-CLIP on two benchmark datasets, DFEW and FERV39K, achieving competitive performance compared to state-of-the-art methods while requiring fewer trainable parameters. By balancing efficiency and accuracy, PE-CLIP sets a new benchmark in resource-efficient DFER. The source code of the proposed PE-CLIP will be publicly available at this https URL .

Strong electric field annihilation by particle-antiparticle pair creation, also known as the Schwinger effect, is a non-perturbative prediction of quantum electrodynamics. Its experimental demonstration remains elusive, as threshold electric fields are extremely strong and beyond current reach. Here, we propose a mesoscopic variant of the Schwinger effect in graphene, which hosts Dirac fermions with an approximate electron-hole symmetry. Using transport measurements, we report on universal 1d-Schwinger conductance at the pinchoff of ballistic graphene transistors. Strong pinchoff electric fields are concentrated within approximately 1 $\mu$m of the transistor's drain, and induce Schwinger electron-hole pair creation at saturation. This effect precedes a collective instability toward an ohmic Zener regime, which is rejected at twice the pinchoff voltage in long devices. These observations advance our understanding of current saturation limits in ballistic graphene and provide a direction for further quantum electrodynamic experiments in the laboratory.

Deep learning has enabled various Internet of Things (IoT) applications. Still, designing models with high accuracy and computational efficiency remains a significant challenge, especially in real-time video processing applications. Such applications exhibit high inter- and intra-frame redundancy, allowing further improvement. This paper proposes a similarity-aware training methodology that exploits data redundancy in video frames for efficient processing. Our approach introduces a per-layer regularization that enhances computation reuse by increasing the similarity of weights during training. We validate our methodology on two critical real-time applications, lane detection and scene parsing. We observe an average compression ratio of approximately 50% and a speedup of \sim 1.5x for different models while maintaining the same accuracy.

We introduce an additive Gaussian process framework accounting for monotonicity constraints and scalable to high dimensions. Our contributions are threefold. First, we show that our framework enables to satisfy the constraints everywhere in the input space. We also show that more general componentwise linear inequality constraints can be handled similarly, such as componentwise convexity. Second, we propose the additive MaxMod algorithm for sequential dimension reduction. By sequentially maximizing a squared-norm criterion, MaxMod identifies the active input dimensions and refines the most important ones. This criterion can be computed explicitly at a linear cost. Finally, we provide open-source codes for our full framework. We demonstrate the performance and scalability of the methodology in several synthetic examples with hundreds of dimensions under monotonicity constraints as well as on a real-world flood application.

Strain-mediated magnetoelectric (ME) heterostructures enable electric-field control of magnetism and are promising for ultra-low-power spintronic logic. Yet achieving spatially selective, low-voltage control in thin films and quantifying ME coupling across the full ferroelastic loop remains challenging. Here, we investigate sub-micrometer Ni/BPZT thin-film devices with laterally patterned gates that localize in-plane strain beneath the Ni stripe and modulate its magnetization. We use anisotropic magnetoresistance to measure magnetization changes across the ferroelastic loop under different magnetic bias fields. Combined with Multiphysics strain simulations and micromagnetic modeling, this provides a quantitative framework that captures the convolution of ferroelastic and magnetoelastic nonlinearities and provides critical insight for device design, while enabling multi-state, bias-field-free magnetization control for non-conventional computing. The extracted coupling coefficient in linear range is 1.3 mT/V across a 700 nm gap, with a clear pathway to improving voltage efficiency through device scaling, establishing a scalable CMOS-compatible platform for energy-efficient spintronic devices.

Terahertz Time Domain Spectroscopy (THz-TDS) systems have emerged as mature technologies with significant potential across various research fields and industries. However, the lack of standardized methods for signal and noise estimation and reduction hinders its full potential. This paper introduces a methodology to significantly reduce noise in THz-TDS time traces, providing a reliable and less biased estimation of the signal. The method results in an improved signal-to-noise ratio, enabling the utilization of the full dynamic range of such setups. Additionally, we investigate the estimation of the covariance matrix to quantify the uncertainties associated with the signal estimator. This matrix is essential for extracting accurate material parameters by normalizing the error function in the fitting process. Our approach addresses practical scenarios where the number of repeated measurements is limited compared to the sampling time axis length. We envision this work as the initial step toward standardizing THz-TDS data processing. We believe it will foster collaboration between the THz and signal processing communities, leading to the development of more sophisticated methods to tackle new challenges introduced by novel setups based on optoelectronic devices and dual-comb spectroscopy.

The theory of open quantum systems plays a fundamental role in several scientific and technological disciplines, from quantum computing and information science to molecular electronics and quantum thermodynamics. Despite its widespread relevance, a rigorous formulation of quantum dissipation in conjunction with thermal noise remains a topic of active research. In this work, we establish a formal correspondence between classical stochastic thermodynamics, in particular the Fokker-Planck and Klein-Kramers equations, and the quantum master equation. Building on prior studies of multiplicative noise in classical stochastic differential equations, we demonstrate that thermal noise at the quantum level manifests as a multidimensional geometric stochastic process. By applying canonical quantization, we introduce a novel Hermitian dissipation operator that serves as a quantum analogue of classical viscous friction. This operator allows for a well-defined expression of heat exchange between a system and its environment, enabling the formulation of an alternative quantum equipartition theorem. Our framework ensures a precise energy balance that aligns with the first law of thermodynamics and an entropy balance consistent with the second law. The theoretical formalism is applied to two prototypical quantum systems, the harmonic oscillator and a particle in an infinite potential well, for which it provides new insights into nonequilibrium thermodynamics at the quantum scale. Our results advance the understanding of dissipation in quantum systems and establish a foundation for future studies on stochastic thermodynamics in the quantum domain.

We report on the generic classical electric circuit modeling that describes standard single-tone microwave optomechanics. Based on a parallel RLC circuit in which a mechanical oscillator acts as a movable capacitor, derivations of analytical expressions are presented, including key features such as the back-action force, the input-output expressions, and the spectral densities associated, all in the classical regime. These expressions coincide with the standard quantum treatment performed in optomechanics when the occupation number of both cavity and mechanical oscillator are large. Besides, the derived analytics transposes optical elements and properties into electronics terms, which is mandatory for quantitative measurement and design purposes. Finally, the direct comparison between the standard quantum treatment and the classical model addresses the bounds between quantum and classical regimes, highlighting the features which are truly quantum, and those which are not.

Existing methods for driver facial expression recognition (DFER) are often computationally intensive, rendering them unsuitable for real-time applications. In this work, we introduce a novel transfer learning-based dual architecture, named ShuffViT-DFER, which elegantly combines computational efficiency and accuracy. This is achieved by harnessing the strengths of two lightweight and efficient models using convolutional neural network (CNN) and vision transformers (ViT). We efficiently fuse the extracted features to enhance the performance of the model in accurately recognizing the facial expressions of the driver. Our experimental results on two benchmarking and public datasets, KMU-FED and KDEF, highlight the validity of our proposed method for real-time application with superior performance when compared to state-of-the-art methods.

Evaluating house prices is crucial for various stakeholders, including homeowners, investors, and policymakers. However, traditional spatial interpolation methods have limitations in capturing the complex spatial relationships that affect property values. To address these challenges, we have developed a new method called Multi-Head Gated Attention for spatial interpolation. Our approach builds upon attention-based interpolation models and incorporates multiple attention heads and gating mechanisms to capture spatial dependencies and contextual information better. Importantly, our model produces embeddings that reduce the dimensionality of the data, enabling simpler models like linear regression to outperform complex ensembling models. We conducted extensive experiments to compare our model with baseline methods and the original attention-based interpolation model. The results show a significant improvement in the accuracy of house price predictions, validating the effectiveness of our approach. This research advances the field of spatial interpolation and provides a robust tool for more precise house price evaluation. Our GitHub this http URL the data and code for all datasets, which are available for researchers and practitioners interested in replicating or building upon our work.

We demonstrate a free-space amplitude modulator for mid-infrared radiation (lambda=9.6 um) that operates at room temperature up to at least 20 GHz (above the -3dB cutoff frequency measured at 8.2 GHz). The device relies on the ultra-fast transition between weak and strong-coupling regimes induced by the variation of the applied bias voltage. Such transition induces a modulation of the device reflectivity. It is made of a semiconductor heterostructure enclosed in a judiciously designed array of metal-metal optical resonators, that - all-together - behave as an electrically tunable surface. At negative bias, it operates in the weak light-matter coupling regime. Upon application of an appropriate positive bias, the quantum wells populate with electrons and the device transitions to the strong-coupling regime. The modulator transmission keeps linear with input RF power in the 0dBm - 9dBm range. The increase of optical powers up to 25 mW exhibit a weak beginning saturation a little bit below.

Markerless Motion Capture (MoCap) using smartphone cameras is a promising approach to making exergames more accessible and cost-effective for health and rehabilitation. Unlike traditional systems requiring specialized hardware, recent advancements in AI-powered pose estimation enable movement tracking using only a mobile device. For an upcoming study, a mobile application with real-time exergames including markerless motion capture is being developed. However, implementing such technology introduces key challenges, including balancing accuracy and real-time responsiveness, ensuring proper user interaction. Future research should explore optimizing AI models for realtime performance, integrating adaptive gamification, and refining user-centered design principles. By overcoming these challenges, smartphone-based exergames could become powerful tools for engaging users in physical activity and rehabilitation, extending their benefits to a broader audience.

We discuss the effective diffusion constant $D_{\it eff}$ for stochastic processes with spatially-dependent noise. Starting from a stochastic process given by a Langevin equation, different drift-diffusion equations can be derived depending on the choice of the discretization rule $0 \leq \alpha \leq 1$. We initially study the case of periodic heterogeneous diffusion without drift and we determine a general result for the effective diffusion coefficient $D_{\it eff}$, which is valid for any value of $\alpha$. We study the case of periodic sinusoidal diffusion in detail and we find a relationship with Legendre functions. Then, we derive $D_{\it eff}$ for general $\alpha$ in the case of diffusion with periodic spatial noise and in the presence of a drift term, generalizing the Lifson-Jackson theorem. Our results are illustrated by analytical and numerical calculations on generic periodic choices for drift and diffusion terms.

This paper investigates a class of controlled stochastic partial differential equations (SPDEs) arising in the modeling of composite materials with spatially varying properties. The state equation describes the evolution of a material property, influenced by control inputs that adjust the diffusivity in different spatial regions. We establish the existence of mild solutions to the SPDE under appropriate regularity conditions on the coefficients and the control. A derivation of the sufficient and necessary conditions for optimality is provided using the stochastic maximum principle. These conditions connect the state dynamics to adjoint processes, enabling the characterization of the optimal control in terms of the curvature of the state and the sensitivity of the cost. Two explicit solvable examples are presented to illustrate the theoretical results, where the optimal control is computed explicitly for a composite material with piecewise constant diffusivity.

In this paper, we present the asymptotic properties of the moment estimator for autoregressive (AR for short) models subject to Markovian changes in regime under the assumption that the errors are uncorrelated but not necessarily independent. We relax the standard independence assumption on the innovation process to extend considerably the range of application of the Markov-switching AR models. We provide necessary conditions to prove the consistency and asymptotic normality of the moment estimator in a specific case. Particular attention is paid to the estimation of the asymptotic covariance matrix. Finally, some simulation studies and an application to the hourly meteorological data are presented to corroborate theoretical work.

Conventional electronics is founded on a paradigm where shaping perfect electrical elements is done at the fabrication plant, so as to make devices and systems identical, "eternally immutable". In nature, morphogenic evolutions are observed in most living organisms and exploit topological plasticity as a low-resource mechanism for in operando manufacturing and computation. Often fractal, the resulting topologies feature inherent disorder: a property which is never exploited in conventional electronics manufacturing, while necessary for data generation and security in software. In this study, we present how such properties can be exploited to implement long-term and evolvable synaptic plasticity in an electronic hardware. The rich topology of conducting polymer dendrites (CPDs) is exploited to program the non-ideality of their electrochemical capacitances containing constant-phase-elements. Their evolution through structural changes alters the characteristic time constants for them to charge and discharge with the applied voltage stimuli. Under a train of voltage spikes, the evolvable current relaxation of the electrochemical systems promotes short-term plasticity with timescales ranging from milliseconds to seconds. This large window depends on the temporality of the voltage pulses used for reading, but also on the structure of a pair of CPDs on two electrodes, grown by voltage pulses. This study demonstrates how relevant physically transient and non-ideal electrochemical components can be exploited for unconventional electronics, with the aim to mimic a universal property of living organisms which could barely be replicated in a silicon monocrystal.

Underwater noise pollution caused by anthropogenic activities, such as offshore wind farms, significantly affects marine life, hindering the intra- and inter-specific interactions of many aquatic species. A common strategy to mitigate these effects is to enclose the noise source within a physical barrier to achieve acceptable noise levels in the surrounding region. Underwater barriers typically achieve noise reduction through sound absorption based on locally resonant systems (e.g., foam and bubble elements) or sound reflecting systems (e.g., air bubble curtains). Although locally resonant-based solutions can yield significant noise attenuation levels, their performance is usually narrow-band. On the other hand, sound-reflecting systems require wider dimensions, presenting poor performance in the low-frequency range. Thus, significant low-frequency underwater noise attenuation remains an open issue, especially when considering thin structures that perform over a broad frequency range. In this work, we present the design of thin metamaterial-based acoustic barriers whose underwater noise attenuation is owed to tailored anisotropic material properties. A topology optimization approach is used to obtain a unit cell that maximizes the coupling between normal stresses and shear strains (and vice-versa). The resulting metabarrier presents a sub-wavelength thickness-to-wavelength ratio in the low-frequency range (circa 1/70 below 1 kHz) and also high sound transmission loss values at higher frequencies (almost 100 dB above 2 kHz). We also investigate the effects of increased hydrostatic pressure, presenting structural modifications that enable real-world applications. The results presented in this work indicate an efficient metamaterial-based solution for the mitigation of underwater noise, suggesting a fruitful direction for the exploitation of anisotropy in acoustic insulation applications.