Search

Current evaluations of defenses against prompt attacks in large language model (LLM) applications often overlook two critical factors: the dynamic nature of adversarial behavior and the usability penalties imposed on legitimate users by restrictive defenses. We propose D-SEC (Dynamic Security Utility Threat Model), which explicitly separates attackers from legitimate users, models multi-step interactions, and expresses the security-utility in an optimizable form. We further address the shortcomings in existing evaluations by introducing Gandalf, a crowd-sourced, gamified red-teaming platform designed to generate realistic, adaptive attack. Using Gandalf, we collect and release a dataset of 279k prompt attacks. Complemented by benign user data, our analysis reveals the interplay between security and utility, showing that defenses integrated in the LLM (e.g., system prompts) can degrade usability even without blocking requests. We demonstrate that restricted application domains, defense-in-depth, and adaptive defenses are effective strategies for building secure and useful LLM applications., Comment: Niklas Pfister, V\'aclav Volhejn and Manuel Knott contributed equally

Artificial intelligence (AI) solutions that automatically extract information from digital histology images have shown great promise for improving pathological diagnosis. Prior to routine use, it is important to evaluate their predictive performance and obtain regulatory approval. This assessment requires appropriate test datasets. However, compiling such datasets is challenging and specific recommendations are missing. A committee of various stakeholders, including commercial AI developers, pathologists, and researchers, discussed key aspects and conducted extensive literature reviews on test datasets in pathology. Here, we summarize the results and derive general recommendations for the collection of test datasets. We address several questions: Which and how many images are needed? How to deal with low-prevalence subsets? How can potential bias be detected? How should datasets be reported? What are the regulatory requirements in different countries? The recommendations are intended to help AI developers demonstrate the utility of their products and to help regulatory agencies and end users verify reported performance measures. Further research is needed to formulate criteria for sufficiently representative test datasets so that AI solutions can operate with less user intervention and better support diagnostic workflows in the future.

As the continual experimental advances made in Electron energy loss spectroscopy (EELS) and cathodoluminescence (CL) open the door to practical exploitations of plasmonic effects in metal nanoparticles, there is an increasing need for precise interpretation and guidance of such experiments. Numerical simulations are available but lack physical insight, while traditional analytical approaches are rare and limited to studying specific, simple structures. Here, we propose a versatile and efficient method based on transformation optics which can fully characterize and model the EELS problems of nanoparticles of complex geometries. Detailed discussions are given on 2D and 3D nanoparticle dimers, where the frequency and time domain responses under electron beam excitations are derived.

Electron energy loss spectroscopy (EELS) and Cathodoluminescence (CL) play a pivotal role in many of the cutting edge experiments in plasmonics. EELS and CL experiments are usually supported by numerical simulations, which, whilst accurate, may not provide as much physical insight as analytical calculations do. Fully analytical solutions to EELS and CL systems in plasmonics are rare and difficult to obtain. This paper aims to narrow this gap by introducing a new method based on Transformation optics that allows to calculate the quasi-static frequency and time-domain response of plasmonic particles under electron beam excitation.

We present a simple design to achieve bianisotropy at visible wavelengths: an ultrathin plasmonic grating made of a gold grating covered by a thin flat layer of gold. We show experimentally and through simulations that the grating exhibits magneto-electric coupling and features asymmetric reflection and absorption, all that with a device thickness of a tenth of the operating wavelength. We compared the spectral results and retrieved the effective material parameters of different polarizations and devices. We show that both asymmetry and strong coupling between the incoming light and the optically interacting surfaces are required for obtaining asymmetric optical behavior in metasurfaces.

Here we study subwavelength gratings for coupling into graphene plasmons by means of an an- alytical model based on transformation optics that is not limited to very shallow gratings. We consider gratings that consist of a periodic modulation of the charge density in the graphene sheet, and gratings formed by this conductivity modulation together with a dielectric grating placed in close vicinity of the graphene. Explicit expressions for the dispersion relation of the plasmon po- laritons supported by the system, and reflectance and transmittance under plane wave illumination are given. We discuss the conditions for maximising the coupling between incident radiation and plasmons in the graphene, finding the optimal modulation strength for a conductivity grating., Comment: Accepted for Journal of Optics, special issue on Transformation Optics, in November 2015

We study the influence of off-diagonal harmonic noise on transitions in a Landau-Zener model. We demonstrate that the harmonic noise can change the transition probabilities substantially and that its impact depends strongly on the characteristic frequency of the noise. In the underdamped regime of the noise process, its effect is compared with the one of a deterministic sinusoidally oscillating function. While altering the properties of the noise process allows one to engineer the transitions probabilities, driving the system with a deterministic sinusoidal function can result in larger and more controlled changes of the transition probability. This may be relevant for realistic implementations of our model with Bose-Einstein condensates in noise-driven optical lattices.

We investigated how the presence of an additional lattice potential, driven by a harmonic noise process, changes the transition rate from the ground band to the first excited band in a Wannier-Stark system. Alongside numerical simulations, we present two models that capture the essential features of the dynamics. The first model uses a noise-driven Landau-Zener approximation and describes the short time evolution of the full system very well. The second model assumes that the noise process' correlation time is much larger than the internal timescale of the system, yet it allows for good estimates of the observed transition rates and gives a simple interpretation of the dynamics. One of the central results is that we obtain a way to control the interband transitions with the help of the second lattice. This could readily be realized in state-of-the-art experiments using either Bose-Einstein condensates or optical pulses in engineered potentials.

The functional significance of correlations between action potentials of neurons is still a matter of vivid debates. In particular it is presently unclear how much synchrony is caused by afferent synchronized events and how much is intrinsic due to the connectivity structure of cortex. The available analytical approaches based on the diffusion approximation do not allow to model spike synchrony, preventing a thorough analysis. Here we theoretically investigate to what extent common synaptic afferents and synchronized inputs each contribute to closely time-locked spiking activity of pairs of neurons. We employ direct simulation and extend earlier analytical methods based on the diffusion approximation to pulse-coupling, allowing us to introduce precisely timed correlations in the spiking activity of the synaptic afferents. We investigate the transmission of correlated synaptic input currents by pairs of integrate-and-fire model neurons, so that the same input covariance can be realized by common inputs or by spiking synchrony. We identify two distinct regimes: In the limit of low correlation linear perturbation theory accurately determines the correlation transmission coefficient, which is typically smaller than unity, but increases sensitively even for weakly synchronous inputs. In the limit of high afferent correlation, in the presence of synchrony a qualitatively new picture arises. As the non-linear neuronal response becomes dominant, the output correlation becomes higher than the total correlation in the input. This transmission coefficient larger unity is a direct consequence of non-linear neural processing in the presence of noise, elucidating how synchrony-coded signals benefit from these generic properties present in cortical networks.